From 26bb02f39132821dc3d0fb9f8f33d06e736dd27b Mon Sep 17 00:00:00 2001 From: huangrt01 Date: Sat, 27 Jun 2026 15:53:41 +0800 Subject: [PATCH 1/3] Add public showcase case pages --- docs/showcases/README.md | 4 +- .../0617-blocked-p0-safe-rotation.en.html | 138 ++ .../cases/0617-blocked-p0-safe-rotation.html | 138 ++ .../cases/0617-blocked-p0-safe-rotation.md | 2 +- ...19-dynamic-workflow-hardware-agent.en.html | 138 ++ .../0619-dynamic-workflow-hardware-agent.md | 2 +- .../cases/0619-loopx-self-iteration.en.html | 138 ++ .../cases/0619-loopx-self-iteration.html | 138 ++ .../cases/0619-loopx-self-iteration.md | 2 +- .../0620-creator-operator-case-spec.en.html | 138 ++ .../0620-creator-operator-case-spec.html | 138 ++ .../cases/0620-creator-operator-case-spec.md | 2 +- .../0623-agent-to-agent-pr-comments.en.html | 138 ++ .../0623-agent-to-agent-pr-comments.html | 138 ++ .../cases/0623-agent-to-agent-pr-comments.md | 2 +- .../0623-overnight-project-refactor.en.html | 138 ++ .../0623-overnight-project-refactor.html | 138 ++ .../cases/0623-overnight-project-refactor.md | 2 +- .../cases/0624-pr-issue-auto-fix.en.html | 138 ++ .../cases/0624-pr-issue-auto-fix.html | 138 ++ .../showcases/cases/0624-pr-issue-auto-fix.md | 2 +- .../cases/0627-overnight-pr-batch.en.html | 138 ++ .../cases/0627-overnight-pr-batch.html | 138 ++ .../cases/0627-overnight-pr-batch.md | 2 +- docs/showcases/frontend-surface.md | 4 +- docs/showcases/index.en.html | 201 +++ docs/showcases/index.html | 201 +++ .../showcase-animation-prototype.html | 242 ++-- .../showcase-animation-storyboard.json | 36 +- docs/showcases/showcase-catalog.json | 582 +++++---- examples/export-frontstage-share-bundle.mjs | 14 +- examples/frontstage-share-bundle-smoke.mjs | 12 + examples/showcase-animation-prototype.py | 6 +- ...howcase-animation-source-boundary-smoke.py | 10 +- examples/showcase-catalog-smoke.py | 37 +- .../showcase-frontstage-prototype-smoke.py | 22 +- examples/showcase-frontstage-prototype.py | 12 +- examples/showcase-html-pages.py | 1157 +++++++++++++++++ 38 files changed, 4271 insertions(+), 355 deletions(-) create mode 100644 docs/showcases/cases/0617-blocked-p0-safe-rotation.en.html create mode 100644 docs/showcases/cases/0617-blocked-p0-safe-rotation.html create mode 100644 docs/showcases/cases/0619-dynamic-workflow-hardware-agent.en.html create mode 100644 docs/showcases/cases/0619-loopx-self-iteration.en.html create mode 100644 docs/showcases/cases/0619-loopx-self-iteration.html create mode 100644 docs/showcases/cases/0620-creator-operator-case-spec.en.html create mode 100644 docs/showcases/cases/0620-creator-operator-case-spec.html create mode 100644 docs/showcases/cases/0623-agent-to-agent-pr-comments.en.html create mode 100644 docs/showcases/cases/0623-agent-to-agent-pr-comments.html create mode 100644 docs/showcases/cases/0623-overnight-project-refactor.en.html create mode 100644 docs/showcases/cases/0623-overnight-project-refactor.html create mode 100644 docs/showcases/cases/0624-pr-issue-auto-fix.en.html create mode 100644 docs/showcases/cases/0624-pr-issue-auto-fix.html create mode 100644 docs/showcases/cases/0627-overnight-pr-batch.en.html create mode 100644 docs/showcases/cases/0627-overnight-pr-batch.html create mode 100644 docs/showcases/index.en.html create mode 100644 docs/showcases/index.html create mode 100644 examples/showcase-html-pages.py diff --git a/docs/showcases/README.md b/docs/showcases/README.md index 6d0f3158..f0e48534 100644 --- a/docs/showcases/README.md +++ b/docs/showcases/README.md @@ -11,7 +11,7 @@ into a reusable control-plane pattern: - the user-facing value in plain language; - the evidence boundary, including what must stay private; - a reproducible demo or the reason a demo is still pending; -- optional data that a future website can render as a visual story. +- optional data that a future website can render as a public evidence sequence. The machine-readable catalog lives in [showcase-catalog.json](showcase-catalog.json). Public docs and future frontend @@ -118,7 +118,7 @@ public-safe user story. 4. **Reproducible**: add a small synthetic demo or smoke that proves the reusable LoopX behavior without depending on private artifacts. 5. **Frontend-ready**: add or update the catalog fields needed for a visual - website card, such as story beats, pattern tags, and suggested visual + website card, such as evidence sequence, pattern tags, and suggested visual layout. Cases can enter the catalog before they have a runnable demo, but their status diff --git a/docs/showcases/cases/0617-blocked-p0-safe-rotation.en.html b/docs/showcases/cases/0617-blocked-p0-safe-rotation.en.html new file mode 100644 index 00000000..fa1e3e8e --- /dev/null +++ b/docs/showcases/cases/0617-blocked-p0-safe-rotation.en.html @@ -0,0 +1,138 @@ + + + + + + LoopX Showcase: Blocked P0 with safe P1/P2 rotation + + + + +
+
+
+ + + +
2026-06-17 · benchmark-rotation
+

Blocked P0 with safe P1/P2 rotation

+
A gated P0 lane should not stall a whole long-running goal when safe fallback work exists.
+

The operator sees exactly what decision is needed while the agent can keep making bounded progress elsewhere.

+
reproducibleuser-gatefallback
+ +
+ + + + LoopX + + + + + Blocked P0 with safe P + concrete user todo + reproducible synthetic d + goal / trigger + todo / gate / evidence + public outcome + + + + + + + +
+ + +
01

Case context

+

This case shows what should happen when a P0 route is blocked by a user decision: the system should neither keep forcing that lane nor stop the whole goal. The original shape was a benchmark rotation where one lane needed a large local image while other no-upload benchmark work remained safe.

The public repository does not expose raw benchmark tasks or local image names. It reproduces the control-plane behavior with a synthetic smoke.

+ +
02

Repository evidence

+
+
Proof

A user decision should not block all safe work.

+
LoopX intervention

concrete user todo, safe fallback, quota control

+
+ +
Synthetic fixture

`examples/showcase-0617-blocked-p0-safe-rotation-smoke.py` creates a P0 user gate, a P0 agent todo blocked by that gate, and a P1 no-upload fallback.

Quota contract

The smoke asserts `should_run=True`, `requires_user_action=True`, `safe_bypass_allowed=True`, and `safe_bypass_kind=scoped_user_gate_fallback`.

Selected fallback

The fixture selects `terminal_bench_no_upload` while preserving the `ale_image` gate as the user-visible blocker.

Rendered evidence

The smoke checks markdown for `scoped_user_gate_fallback` and safe no-upload Terminal-Bench rotation.

+ +
03

LoopX behavior

+
  • 1
    The user todo names the concrete P0 decision instead of saying only owner gate.
  • 2
    The agent does not spend compute on the gated lane; it selects fallback work that does not depend on the decision.
  • 3
    State records both the blocker and the fallback reason so P0 can resume later.
+ +
04

What the user sees

+

The user sees the exact decision they need to make while the project continues safely elsewhere.

Attention load drops: the user does not need to watch repeated idle polls and does not miss the real decision.

+ +
05

Repository sources

+
case narrativedocs/showcases/cases/0617-blocked-p0-safe-rotation.mdsynthetic smokeexamples/showcase-0617-blocked-p0-safe-rotation-smoke.py
+

Evidence boundary. Synthetic public fixture only; no private screenshots, raw tasks, internal links, local image names, or raw run logs.

+
Case note
+
Demopython3 examples/showcase-0617-blocked-p0-safe-rotation-smoke.py
+
Generated from docs/showcases/showcase-catalog.json. Private links, raw chats, local state, and internal media are excluded.
+
+
+ + diff --git a/docs/showcases/cases/0617-blocked-p0-safe-rotation.html b/docs/showcases/cases/0617-blocked-p0-safe-rotation.html new file mode 100644 index 00000000..e5847df3 --- /dev/null +++ b/docs/showcases/cases/0617-blocked-p0-safe-rotation.html @@ -0,0 +1,138 @@ + + + + + + LoopX Showcase: P0 block 后推进 P1/P2 + + + + +
+
+
+ + + +
2026-06-17 · benchmark-rotation
+

P0 block 后推进 P1/P2

+
用户决策不应阻塞全部安全工作。
+

The operator sees exactly what decision is needed while the agent can keep making bounded progress elsewhere.

+
reproducibleuser-gatefallback
+ +
+ + + + LoopX + + + + + P0 block 后推进 P1/P2 + concrete user todo + reproducible synthetic d + goal / trigger + todo / gate / evidence + public outcome + + + + + + + +
+ + +
01

案例背景

+

这个案例展示 P0 被用户决策卡住时，系统不应该继续硬跑，也不应该让整个目标停摆。原场景是 benchmark rotation：一个 lane 需要大型本地 image，其他 no-upload benchmark work 仍然安全。

公开仓库没有暴露原始 benchmark task 或本地 image 名，而是用 synthetic smoke 复现控制面行为。

+ +
02

仓库证据

+
+
证明点

被阻塞的 P0 决策不应该阻止安全的 P1/P2 工作继续。

+
LoopX 介入

concrete user todo、safe fallback、quota control

+
+ +
synthetic fixture

`examples/showcase-0617-blocked-p0-safe-rotation-smoke.py` 构造 P0 user gate、被 gate 阻塞的 P0 agent todo 和 P1 no-upload fallback。

quota contract

smoke 断言 `should_run=True`、`requires_user_action=True`、`safe_bypass_allowed=True`、`safe_bypass_kind=scoped_user_gate_fallback`。

selected fallback

fixture 选择 `terminal_bench_no_upload`，同时保留 `ale_image` gate 的 user-visible blocker。

rendered evidence

smoke 检查 markdown 中包含 `scoped_user_gate_fallback` 和 safe no-upload Terminal-Bench rotation。

+ +
03

LoopX 行为

+
  • 1
    用户 todo 具体命名 P0 决策，不用“owner gate”这种空话。
  • 2
    agent 不在 gated lane 上花 compute；只选择不依赖该决策的 fallback。
  • 3
    状态同时记录 blocker 和 fallback reason，方便之后恢复 P0。
+ +
04

用户看到什么

+

用户看到需要自己决定的具体问题，同时项目仍能在安全范围内推进。

这减少了注意力负担：不需要每 10 分钟看一次 agent 为什么没动，也不会错过真正需要决策的事项。

+ +
05

仓库来源

+
case narrativedocs/showcases/cases/0617-blocked-p0-safe-rotation.mdsynthetic smokeexamples/showcase-0617-blocked-p0-safe-rotation-smoke.py
+

证据边界. Synthetic public fixture only; no private screenshots, raw tasks, internal links, local image names, or raw run logs.

+
案例说明
+
Demopython3 examples/showcase-0617-blocked-p0-safe-rotation-smoke.py
+
由 docs/showcases/showcase-catalog.json 生成。不包含私有链接、原始聊天、本地状态或内部媒体。
+
+
+ + diff --git a/docs/showcases/cases/0617-blocked-p0-safe-rotation.md b/docs/showcases/cases/0617-blocked-p0-safe-rotation.md index ace3dfd9..12e3d868 100644 --- a/docs/showcases/cases/0617-blocked-p0-safe-rotation.md +++ b/docs/showcases/cases/0617-blocked-p0-safe-rotation.md @@ -63,7 +63,7 @@ text, local image names, internal links, and raw run logs. The behavior is represented by a synthetic fixture because the reusable product value is the control-plane pattern, not the original private artifact. -## Website Story Beats +## Public Evidence Sequence 1. A P0 lane becomes blocked by a concrete user decision. 2. LoopX keeps that decision visible as a user todo. diff --git a/docs/showcases/cases/0619-dynamic-workflow-hardware-agent.en.html b/docs/showcases/cases/0619-dynamic-workflow-hardware-agent.en.html new file mode 100644 index 00000000..322e3d35 --- /dev/null +++ b/docs/showcases/cases/0619-dynamic-workflow-hardware-agent.en.html @@ -0,0 +1,138 @@ + + + + + + LoopX Showcase: Dynamic workflow for hardware-agent development + + + + +
+
+
+ + + +
2026-06-19 · hardware-agent-development
+

Dynamic workflow for hardware-agent development

+
A fuzzy long-running engineering goal needs a shared control plane when multiple worker agents participate.
+

A contributor-approved public artifact shows how one control plane can coordinate Claude Code, generated scripts, and hardware-agent workers across five long-running engineering cases.

+
interactive-htmlmulti-agenthardware
+ +
+ + + + LoopX + + + + + Dynamic workflow for h + goal state + public safe interactive + goal / trigger + todo / gate / evidence + public outcome + + + + + + + +
+ + +
01

Case context

+

The Chinese hardware page remains the canonical artifact. This English companion follows its structure without rewriting the original.

The case demonstrates a dynamic workflow around fuzzy long-running hardware goals: generated scripts coordinate bounded worker-agent actions while LoopX preserves goal state, quota, todo ownership, validation evidence, and run history outside any one chat thread.

+ +
02

Repository evidence

+
+
Proof

Fuzzy goals, multiple workers, and long unattended runs can still converge.

+
LoopX intervention

goal state, worker handoff, dynamic workflow

+
+ +
Public artifact

The canonical HTML page includes the approved hardware workflow artifact and five public-safe hardware cases.

Case family

The companion note names closed validation, timing optimization, design-space exploration, Fmax optimization, and convergence to an engineering floor.

Boundary

The public artifact excludes raw chats, screenshots, proprietary design details, private repos, local paths, task ids, credentials, and unpublished hardware artifacts.

+ +
03

LoopX behavior

+
  • 1
    LoopX owns durable state, quota, todos, claims, evidence, and history.
  • 2
    Claude Code writes task-level orchestration and generated scripts.
  • 3
    Worker agents perform bounded RTL, simulation, and validation work under explicit review boundaries.
+ +
04

What the user sees

+

A contributor-approved page shows how multiple hardware-agent workers can coordinate under one control plane.

Readers can inspect the product pattern without receiving proprietary hardware details or raw execution traces.

+ +
05

Repository sources

+
canonical HTMLdocs/showcases/cases/0619-dynamic-workflow-hardware-agent.htmlcompanion notedocs/showcases/cases/0619-dynamic-workflow-hardware-agent.md
+

Evidence boundary. Public-safe interactive artifact; no raw chats, screenshots, proprietary design details, private repositories, local paths, task ids, credentials, or unpublished hardware artifacts.

+
Case note
+ +
Generated from docs/showcases/showcase-catalog.json. Private links, raw chats, local state, and internal media are excluded.
+
+
+ + diff --git a/docs/showcases/cases/0619-dynamic-workflow-hardware-agent.md b/docs/showcases/cases/0619-dynamic-workflow-hardware-agent.md index 1cc9075a..5beddb9c 100644 --- a/docs/showcases/cases/0619-dynamic-workflow-hardware-agent.md +++ b/docs/showcases/cases/0619-dynamic-workflow-hardware-agent.md @@ -49,7 +49,7 @@ from the repository or hosted as part of the Frontstage Pages bundle. It is not a runnable hardware benchmark and should not be presented as a reproducible EDA workflow. -## Website Story Beats +## Public Evidence Sequence 1. A fuzzy engineering goal needs more than one worker agent. 2. LoopX gives the agents a shared state and ownership surface. diff --git a/docs/showcases/cases/0619-loopx-self-iteration.en.html b/docs/showcases/cases/0619-loopx-self-iteration.en.html new file mode 100644 index 00000000..b9e2b2d9 --- /dev/null +++ b/docs/showcases/cases/0619-loopx-self-iteration.en.html @@ -0,0 +1,138 @@ + + + + + + LoopX Showcase: LoopX self-iteration loop + + + + +
+
+
+ + + +
2026-06-19 · agent-platform-self-improvement
+

LoopX self-iteration loop

+
A high-churn LoopX repo stayed legible while benchmark, product, docs, planning, and side-agent lanes moved in parallel.
+

The operator can let a fast-moving multi-lane agent project continue across benchmark, product, docs, and side-agent work while keeping ownership, gates, validation, follow-up, and conservative efficiency evidence visible in one shared control plane.

+
self-iterationcommit-backedside-agentefficiency-model
+ +
+ + + + LoopX + + + + + LoopX self-iteration l + todo + public evidence case + goal / trigger + todo / gate / evidence + public outcome + + + + + + + +
+ + +
01

Case context

+

This is public-repo self-iteration: LoopX was used to improve LoopX itself, not just one isolated demo. Benchmark adapters, control-plane correctness, planning lanes, dashboard/frontstage, docs, smokes, and multi-agent coordination all moved under high churn.

The evidence is fixed to anchor commit `86d6d9d` so this documentation update does not change its own evidence window. The efficiency model is a conservative public-Git model; it excludes private chats, active-state bodies, and raw benchmark material.

+ +
02

Repository evidence

+
+
Proof

A high-churn multi-lane project can keep state, boundaries, and evidence coherent.

+
LoopX intervention

todo, quota, gate, evidence, review packet, frontstage

+
+
801public commits
59-92dAI-assisted baseline
3.0-4.7xcalendar compression
+
Whole repository

Through the anchor commit: 801 public commits, 570 touched files, 265703 insertions, and 49895 deletions.

Recent window

Since 2026-06-18: 244 public commits, 216 touched files, 52898 insertions, and 20935 deletions.

June 19 signal

On 2026-06-19: 74 public commits, 118 touched files, 16087 insertions, and 1082 deletions.

Efficiency model

The case maps public repo capabilities to 9 requirement clusters and estimates 59-92 AI-coding-assisted developer-days against a 19.6-day public window.

+ +
03

LoopX behavior

+
  • 1
    Registry and prompt contracts name primary and side-agent identities instead of relying on chat memory.
  • 2
    Benchmark, productization, documentation, planning, and side-agent lanes become reviewable obligations.
  • 3
    Quota and status projection distinguish executable work, monitor work, user gates, and blockers.
  • 4
    Side-agent scope lives in prompt and handoff; todo metadata keeps the compact `claimed_by` owner.
  • 5
    Public docs and smokes turn reusable lessons into repository artifacts.
+ +
04

What the user sees

+

The operator can let a primary benchmark lane and product/docs/control-plane side work run in parallel without losing ownership, gates, validation, or merge discipline.

For a potential user, this is evidence that a long-running agent project can remain legible: future agents can recover goals, owners, gates, validation, evidence, and follow-up from public surfaces.

+ +
05

Repository sources

+
case narrativedocs/showcases/cases/0619-loopx-self-iteration.mdworkload signaldocs/showcases/showcase-catalog.jsonfrontstage fixtureexamples/fixtures/long-horizon-self-iteration-rollout.public.jsonfrontstage smokeexamples/long-horizon-self-iteration-rollout-fixture-smoke.py
+

Evidence boundary. Public Git evidence only; no private thread text, local active-state bodies, internal document links, screenshots, raw benchmark material, credentials, or machine-specific paths.

+
Case note
+ +
Generated from docs/showcases/showcase-catalog.json. Private links, raw chats, local state, and internal media are excluded.
+
+
+ + diff --git a/docs/showcases/cases/0619-loopx-self-iteration.html b/docs/showcases/cases/0619-loopx-self-iteration.html new file mode 100644 index 00000000..7cdf5901 --- /dev/null +++ b/docs/showcases/cases/0619-loopx-self-iteration.html @@ -0,0 +1,138 @@ + + + + + + LoopX Showcase: LoopX Meta Agent 自迭代 + + + + +
+
+
+ + + +
2026-06-19 · agent-platform-self-improvement
+

LoopX Meta Agent 自迭代

+
高变更、多车道项目可保持状态、边界和证据。
+

The operator can let a fast-moving multi-lane agent project continue across benchmark, product, docs, and side-agent work while keeping ownership, gates, validation, follow-up, and conservative efficiency evidence visible in one shared control plane.

+
self-iterationcommit-backedside-agentefficiency-model
+ +
+ + + + LoopX + + + + + LoopX Meta Agent 自迭代 + todo + public evidence case + goal / trigger + todo / gate / evidence + public outcome + + + + + + + +
+ + +
01

案例背景

+

这个案例是 public repo self-iteration：LoopX 用来推进 LoopX 自己，不是一个孤立功能 demo。仓库在 benchmark adapters、control-plane correctness、planning lanes、dashboard/frontstage、docs、smokes 和 multi-agent coordination 上同时高频变化。

证据窗口固定到 anchor commit `86d6d9d`，避免文档更新改变自己的证据。这里的效率模型是保守的 public Git model，不使用私有聊天、active-state 原文或 benchmark 原始材料。

+ +
02

仓库证据

+
+
证明点

高 churn 多 lane agent repo 可以保持状态、证据和边界一致。

+
LoopX 介入

todo、quota、gate、evidence、review packet、frontstage

+
+
801public commits
59-92dAI-assisted baseline
3.0-4.7xcalendar compression
+
全仓库证据

截至 anchor commit：801 个 public commits、570 个 touched files、265703 行新增、49895 行删除。

近期窗口

2026-06-18 起有 244 个 public commits、216 个 touched files、52898 行新增、20935 行删除。

0619 当日

2026-06-19 有 74 个 public commits、118 个 touched files、16087 行新增、1082 行删除。

效率模型

把公开仓库能力拆成 9 个 requirement clusters，保守估计 59-92 个 AI-coding-assisted developer-days，对 19.6 天 public window 得出方向性 compression。

+ +
03

LoopX 行为

+
  • 1
    registry、prompt contracts 和 registered agents 命名 primary/side identities，不靠聊天记忆。
  • 2
    benchmark、productization、documentation、planning 和 side-agent lanes 被拆成 reviewable obligations。
  • 3
    quota/status projection 区分 executable work、monitor work、user gates 和 blockers。
  • 4
    side-agent scope 留在 prompt/handoff，todo metadata 只保留 `claimed_by`。
  • 5
    public docs 和 smokes 把可复用经验沉淀为仓库 artifact。
+ +
04

用户看到什么

+

operator 可以让 primary benchmark lane 和 product/docs/control-plane side work 并行，而不会失去 ownership、gate、validation 和 merge discipline。

对潜在用户来说，这是“长期 agent 项目仍然可读”的证据：未来 agent 能从公开表面恢复目标、owner、gate、验证、证据和 follow-up。

+ +
05

仓库来源

+
case narrativedocs/showcases/cases/0619-loopx-self-iteration.mdworkload signaldocs/showcases/showcase-catalog.jsonfrontstage fixtureexamples/fixtures/long-horizon-self-iteration-rollout.public.jsonfrontstage smokeexamples/long-horizon-self-iteration-rollout-fixture-smoke.py
+

证据边界. Public Git evidence only; no private thread text, local active-state bodies, internal document links, screenshots, raw benchmark material, credentials, or machine-specific paths.

+
案例说明
+ +
由 docs/showcases/showcase-catalog.json 生成。不包含私有链接、原始聊天、本地状态或内部媒体。
+
+
+ + diff --git a/docs/showcases/cases/0619-loopx-self-iteration.md b/docs/showcases/cases/0619-loopx-self-iteration.md index 589466d2..f57d29e3 100644 --- a/docs/showcases/cases/0619-loopx-self-iteration.md +++ b/docs/showcases/cases/0619-loopx-self-iteration.md @@ -196,7 +196,7 @@ checks. It excludes private thread text, local active-state bodies, internal document links, screenshots, raw benchmark material, credentials, and machine-specific paths. -## Website Story Beats +## Public Evidence Sequence 1. A long-running agent engineering repo is moving quickly across benchmark, runtime, docs, dashboard, planning, and smoke surfaces. diff --git a/docs/showcases/cases/0620-creator-operator-case-spec.en.html b/docs/showcases/cases/0620-creator-operator-case-spec.en.html new file mode 100644 index 00000000..962d7586 --- /dev/null +++ b/docs/showcases/cases/0620-creator-operator-case-spec.en.html @@ -0,0 +1,138 @@ + + + + + + LoopX Showcase: Creator-operator long-running agent case + + + + +
+
+
+ + + +
2026-06-20 · creator-operations
+

Creator-operator long-running agent case

+
A creator-operator needs a long-running agent loop that keeps research moving while publishing decisions stay gated.
+

A non-technical operator can see what changed, what is blocked, what can safely continue, and how feedback changes the plan without reading prompts, logs, or raw traces.

+
creator_operator_workflowgate_aware_continuationfeedback_capturesafe_side_pathmaterial_libraryAppendix case
+ +
+ + + + LoopX + + + + + Creator-operator long- + creator_operator_workflo + public safe case spec + goal / trigger + todo / gate / evidence + public outcome + + + + + + + +
+ + +
01

Case context

+

This is an appendix case: it shows how a non-technical creator-operator might use LoopX to manage a long-running research and planning loop, but it is not a top-card proof until real public user evidence exists.

The public material uses only synthetic data. The product shape is trend candidates, preference map, insight board, draft queue, material library, human feedback, and controlled replan.

+ +
02

Repository evidence

+
+
Proof

A creator-operator needs a long-running agent loop that keeps research moving while publishing decisions stay gated.

+
LoopX intervention

creator_operator_workflow, gate_aware_continuation, feedback_capture, safe_side_path

+
+ +
Storyboard

`creator-ops-fake-data-storyboard.md` defines seven panels and a full fake fixture with no live platform access.

Feedback contract

`creator-ops-feedback-boundary-contract.md` separates gate decision, preference hint, todo update, boundary correction, reward signal, and product improvement note.

Source status

The contract requires every topic, insight, draft, and material item to carry source status; public repo defaults to `synthetic_demo`.

No autopublish

Publishing is a hard user gate; safe side work may continue, but preference or reward is not publication approval.

+ +
03

LoopX behavior

+
  • 1
    The creative objective is durable goal state, not a hidden task inside chat.
  • 2
    Publish/no-publish is a user gate; research, organization, and source-status work can continue as safe side paths.
  • 3
    Feedback becomes a preference hint, gate decision, todo update, or boundary correction.
  • 4
    Before the next agent run, the user can see the blocked route, safe side path, and validation expectation.
+ +
04

What the user sees

+

A non-technical user can see what changed, what is waiting for them, and what can continue without reading prompts, traces, or raw logs.

The case is useful product direction, but the page clearly labels it as a synthetic spec and makes no claim about real growth, quality, or revenue.

+ +
05

Repository sources

+
case narrativedocs/showcases/cases/0620-creator-operator-case-spec.mdfake-data storyboarddocs/showcases/creator-ops-fake-data-storyboard.mdfeedback boundary contractdocs/showcases/creator-ops-feedback-boundary-contract.md
+

Evidence boundary. Synthetic case spec only; no real platform data, creator notes, screenshots, private drafts, raw browsing traces, local paths, credentials, or performance claims.

+
Case noteStoryboardFeedback
+ +
Generated from docs/showcases/showcase-catalog.json. Private links, raw chats, local state, and internal media are excluded.
+
+
+ + diff --git a/docs/showcases/cases/0620-creator-operator-case-spec.html b/docs/showcases/cases/0620-creator-operator-case-spec.html new file mode 100644 index 00000000..9fa86825 --- /dev/null +++ b/docs/showcases/cases/0620-creator-operator-case-spec.html @@ -0,0 +1,138 @@ + + + + + + LoopX Showcase: 创作者-运营者长跑 Agent 案例 + + + + +
+
+
+ + + +
2026-06-20 · creator-operations
+

创作者-运营者长跑 Agent 案例

+
研究可以继续，发布决策仍然 gated。
+

A non-technical operator can see what changed, what is blocked, what can safely continue, and how feedback changes the plan without reading prompts, logs, or raw traces.

+
creator_operator_workflowgate_aware_continuationfeedback_capturesafe_side_pathmaterial_library附录案例
+ +
+ + + + LoopX + + + + + 创作者-运营者长跑 Agent 案例 + creator-operator workflo + public safe case spec + goal / trigger + todo / gate / evidence + public outcome + + + + + + + +
+ + +
01

案例背景

+

这是 appendix case：它展示非技术创作者/运营者如何用 LoopX 管一个长期 research + planning loop，但还不是 top-card proof，因为没有真实用户公开证据。

公开材料完全使用 synthetic data，重点是产品形态：趋势候选、偏好映射、insight board、draft queue、material library、人类反馈和 controlled replan。

+ +
02

仓库证据

+
+
证明点

创作与运营工作可以共享一个 gate-aware 的长期 agent loop。

+
LoopX 介入

creator-operator workflow、user gate、feedback capture、material library

+
+ +
storyboard

`creator-ops-fake-data-storyboard.md` 定义七个面板和完整 fake fixture，不需要 live platform access。

feedback contract

`creator-ops-feedback-boundary-contract.md` 把 gate decision、preference hint、todo update、boundary correction、reward signal、product improvement note 分开。

source status

contract 要求 topic、insight、draft、material item 都有 source status，public repo 默认 `synthetic_demo`。

no autopublish

publishing 是 hard user gate；safe side work 可以继续，但不能把偏好或 reward 当成发布授权。

+ +
03

LoopX 行为

+
  • 1
    creative objective 是 durable goal state，不是聊天窗口里的隐形任务。
  • 2
    publish/no-publish 是 user gate；research、整理和 source-status 改进可以作为 safe side path。
  • 3
    反馈被写成 preference hint、gate decision、todo update 或 boundary correction。
  • 4
    下一次 agent run 前，用户能看到 blocked route、safe side path 和 validation expectation。
+ +
04

用户看到什么

+

非技术用户不用读 prompt、trace 或 raw logs，就能知道上次改变了什么、什么在等自己、什么可以继续。

这个 case 适合展示产品方向，但页面会明确它是 synthetic spec，不声称真实增长、质量或收入效果。

+ +
05

仓库来源

+
case narrativedocs/showcases/cases/0620-creator-operator-case-spec.mdfake-data storyboarddocs/showcases/creator-ops-fake-data-storyboard.mdfeedback boundary contractdocs/showcases/creator-ops-feedback-boundary-contract.md
+

证据边界. Synthetic case spec only; no real platform data, creator notes, screenshots, private drafts, raw browsing traces, local paths, credentials, or performance claims.

+
案例说明StoryboardFeedback
+ +
由 docs/showcases/showcase-catalog.json 生成。不包含私有链接、原始聊天、本地状态或内部媒体。
+
+
+ + diff --git a/docs/showcases/cases/0620-creator-operator-case-spec.md b/docs/showcases/cases/0620-creator-operator-case-spec.md index baffd98a..8828aaca 100644 --- a/docs/showcases/cases/0620-creator-operator-case-spec.md +++ b/docs/showcases/cases/0620-creator-operator-case-spec.md @@ -179,7 +179,7 @@ This case is a synthetic product case spec. It intentionally excludes: Future demos should use fake data or public-domain sample material and should keep no-autopublish gates visible. -## Website Story Beats +## Public Evidence Sequence 1. A creator-operator has a long-running research and content-planning goal. 2. The agent proposes trend candidates, maps them to preferences, and drafts diff --git a/docs/showcases/cases/0623-agent-to-agent-pr-comments.en.html b/docs/showcases/cases/0623-agent-to-agent-pr-comments.en.html new file mode 100644 index 00000000..89621101 --- /dev/null +++ b/docs/showcases/cases/0623-agent-to-agent-pr-comments.en.html @@ -0,0 +1,138 @@ + + + + + + LoopX Showcase: Agent-to-agent PR comment and fix loop + + + + +
+
+
+ + + +
2026-06-23 · pull-request-review
+

Agent-to-agent PR comment and fix loop

+
PR review feedback can become an owned agent todo with fix evidence instead of a loose chat reminder.
+

The operator can let agents coordinate around PR comments without losing final review visibility or fix evidence.

+
handoffPR commentclaimed_by
+ +
+ + + + LoopX + + + + + Agent-to-agent PR comm + claimed_by + public safe pattern case + goal / trigger + todo / gate / evidence + public outcome + + + + + + + +
+ + +
01

Case context

+

This case shows how PR review feedback can move across multiple agents without losing ownership. The important chain is comment, claim, handoff, fix, validation, and review packet rather than the chat transcript.

The public repository contains the control-plane pieces: `claimed_by` appears in todo projection, review packets, event-sourced state, and multi-agent or side-agent prompt contracts.

+ +
02

Repository evidence

+
+
Proof

Multiple agents can coordinate around PR review comments without losing owner review.

+
LoopX intervention

claimed_by, handoff gate, review packet, comment/fix loop

+
+ +
Todo ownership

`event_sourced_state_contract_v0` defines `todo_claimed` as a canonical event for ownership, lease, or `claimed_by`.

Review packet

`loopx/review_packet.py` renders open todo `claimed_by` values into handoff and review text.

Side-agent contract

`docs/heartbeat-automation-prompt.md` requires side agents to self-merge only small validated evidence-backed work or create a claimed handoff todo.

CLI smokes

`examples/todo-lifecycle-cli-smoke.py` and `examples/todo-cli-smoke.py` cover claim, handoff successor, side-agent self-merge, and review handoff rules.

+ +
03

LoopX behavior

+
  • 1
    PR feedback becomes an owned todo rather than a chat reminder.
  • 2
    Blocked or cross-lane fixes move through handoff gates and review packets instead of another agent guessing context.
  • 3
    The fixing agent leaves diff, validation, and remaining gates; follow-up work becomes a successor todo.
  • 4
    Owner review still happens on the PR or review surface; LoopX supplies state, evidence, and handoff.
+ +
04

What the user sees

+

The user does not have to shepherd every PR comment manually; the control plane shows who claimed the feedback, where the fix evidence is, and what still needs review.

Multi-agent collaboration stops depending on memory of which agent saw which comment.

+ +
05

Repository sources

+
event-sourced todo claimdocs/reference/protocols/event-sourced-state-contract-v0.mdreview packet codeloopx/review_packet.pyside-agent prompt contractdocs/heartbeat-automation-prompt.mdtodo lifecycle smokeexamples/todo-lifecycle-cli-smoke.py
+

Evidence boundary. Public-safe pattern case only; no private screenshots, raw chats, internal review notes, local state, credentials, or unpublished artifacts.

+
Case note
+ +
Generated from docs/showcases/showcase-catalog.json. Private links, raw chats, local state, and internal media are excluded.
+
+
+ + diff --git a/docs/showcases/cases/0623-agent-to-agent-pr-comments.html b/docs/showcases/cases/0623-agent-to-agent-pr-comments.html new file mode 100644 index 00000000..210b4e88 --- /dev/null +++ b/docs/showcases/cases/0623-agent-to-agent-pr-comments.html @@ -0,0 +1,138 @@ + + + + + + LoopX Showcase: Agent to agent 回复 PR comment 和 PR Fix + + + + +
+
+
+ + + +
2026-06-23 · pull-request-review
+

Agent to agent 回复 PR comment 和 PR Fix

+
多 agent 可围绕 PR review 协同发现、评论、修复。
+

The operator can let agents coordinate around PR comments without losing final review visibility or fix evidence.

+
handoffPR commentclaimed_by
+ +
+ + + + LoopX + + + + + Agent to agent 回复 PR c + claimed_by + public safe pattern case + goal / trigger + todo / gate / evidence + public outcome + + + + + + + +
+ + +
01

案例背景

+

这个案例描述的是 PR review feedback 在多 agent 之间流转时如何不丢 ownership。重点不是聊天记录，而是 comment、claim、handoff、fix、validation、review packet 这条链。

公开仓库里可以看到相关控制面已经被产品化：`claimed_by` 出现在 todo projection、review packet、event-sourced state 和多 agent/side-agent prompt 合约里。

+ +
02

仓库证据

+
+
证明点

多 agent lane 可以围绕 PR comment 协作，同时保留 owner review。

+
LoopX 介入

claimed_by、handoff gate、review packet、comment/fix loop

+
+ +
todo ownership

`event_sourced_state_contract_v0` 把 `todo_claimed` 定义为 canonical event，记录 ownership、lease 或 `claimed_by`。

review packet

`loopx/review_packet.py` 会把 open todo 的 `claimed_by` 显示进 handoff/review 文本。

side-agent contract

`docs/heartbeat-automation-prompt.md` 规定 side-agent 小变更可带 evidence 自合并，否则创建 claimed-by handoff todo。

CLI smokes

`examples/todo-lifecycle-cli-smoke.py`、`examples/todo-cli-smoke.py` 覆盖 claim、handoff successor、side-agent self-merge 和 review handoff 规则。

+ +
03

LoopX 行为

+
  • 1
    PR feedback 先成为一个有 owner 的 todo，而不是 chat reminder。
  • 2
    被阻塞或跨 lane 的修复通过 handoff gate 和 review packet 传递，不允许另一个 agent 猜测上下文。
  • 3
    修复 agent 需要留下 diff、validation 和剩余 gate；后续工作创建 successor todo。
  • 4
    owner review 仍在 PR/review surface 上完成，LoopX 只提供状态、证据和 handoff。
+ +
04

用户看到什么

+

用户不必手动 shepherd 每条 PR comment；控制面会显示谁 claim 了反馈、修复证据在哪里、还有什么需要 review。

多 agent 协作不会变成“谁看过这条评论”的记忆题。

+ +
05

仓库来源

+
event-sourced todo claimdocs/reference/protocols/event-sourced-state-contract-v0.mdreview packet codeloopx/review_packet.pyside-agent prompt contractdocs/heartbeat-automation-prompt.mdtodo lifecycle smokeexamples/todo-lifecycle-cli-smoke.py
+

证据边界. Public-safe pattern case only; no private screenshots, raw chats, internal review notes, local state, credentials, or unpublished artifacts.

+
案例说明
+ +
由 docs/showcases/showcase-catalog.json 生成。不包含私有链接、原始聊天、本地状态或内部媒体。
+
+
+ + diff --git a/docs/showcases/cases/0623-agent-to-agent-pr-comments.md b/docs/showcases/cases/0623-agent-to-agent-pr-comments.md index ce5f01c2..9dd249e5 100644 --- a/docs/showcases/cases/0623-agent-to-agent-pr-comments.md +++ b/docs/showcases/cases/0623-agent-to-agent-pr-comments.md @@ -49,7 +49,7 @@ state, credentials, and unpublished artifacts. The public-safe evidence shape is the PR comment/fix lifecycle itself: a visible PR surface, a claimed todo, the fix diff, validation output, and the resulting review packet. -## Website Story Beats +## Public Evidence Sequence 1. A PR receives feedback that should become executable work. 2. LoopX turns the feedback into an owned todo instead of a chat reminder. diff --git a/docs/showcases/cases/0623-overnight-project-refactor.en.html b/docs/showcases/cases/0623-overnight-project-refactor.en.html new file mode 100644 index 00000000..0e2b239d --- /dev/null +++ b/docs/showcases/cases/0623-overnight-project-refactor.en.html @@ -0,0 +1,138 @@ + + + + + + LoopX Showcase: Overnight project refactor as PR-sized slices + + + + +
+
+
+ + + +
2026-06-23 · repository-refactor
+

Overnight project refactor as PR-sized slices

+
A broad refactor can run overnight while staying split into human-sized review units.
+

The operator can wake up to reviewable PR-sized refactor slices instead of a single giant autonomous diff.

+
refactorPR slicestodo evidence
+ +
+ + + + LoopX + + + + + Overnight project refa + loop + public safe pattern case + goal / trigger + todo / gate / evidence + public outcome + + + + + + + +
+ + +
01

Case context

+

This case addresses the risk of unattended refactoring: a long-running agent can mix cleanup, behavior change, discoveries, and stale plans into one broad diff.

The public evidence is not a private overnight screenshot. It is the control-plane behavior already documented and smoke-tested in the repository: todo lifecycle, successor/supersede, validation writeback, and review packets.

+ +
02

Repository evidence

+
+
Proof

Long unattended refactors can split into moderate PR slices instead of one unreviewable diff.

+
LoopX intervention

loop, todo follow-up, supersede, PR-sized slices

+
+ +
Successor path

`docs/lark-kanban-control-plane-adapter.md` says real successors use `todo complete --next-*`, while replacements or narrower splits use `todo supersede --next-agent-todo`.

Side-agent completion

`docs/heartbeat-automation-prompt.md` requires nontrivial completion to create a successor todo or a no-follow-up rationale.

CLI validation

`examples/todo-lifecycle-cli-smoke.py` covers `--next-agent-todo`, `todo supersede`, claim inheritance, and handoff successors.

Review shape

`loopx review-packet` packages open todos, claimed_by, and handoff state for reviewer consumption.

+ +
03

LoopX behavior

+
  • 1
    The current refactor slice must stay reviewable; overnight discoveries do not all land in one PR.
  • 2
    New discoveries become follow-up todos; changed routes supersede stale todos.
  • 3
    Each slice carries focused validation or doc/contract smoke evidence rather than raw agent traces.
  • 4
    Broad or unclear-risk slices route to review handoff instead of self-merge.
+ +
04

What the user sees

+

The user can let a refactor run overnight and wake up to bounded review units plus remaining todos, not one unreviewable giant change.

The project can move quickly while the review surface remains human-sized.

+ +
05

Repository sources

+
kanban control-plane adapterdocs/lark-kanban-control-plane-adapter.mdheartbeat prompt contractdocs/heartbeat-automation-prompt.mdtodo lifecycle smokeexamples/todo-lifecycle-cli-smoke.pycase narrativedocs/showcases/cases/0623-overnight-project-refactor.md
+

Evidence boundary. Public-safe pattern case only; no private screenshots, raw chats, internal planning notes, local paths, credentials, raw logs, or unpublished project artifacts.

+
Case note
+ +
Generated from docs/showcases/showcase-catalog.json. Private links, raw chats, local state, and internal media are excluded.
+
+
+ + diff --git a/docs/showcases/cases/0623-overnight-project-refactor.html b/docs/showcases/cases/0623-overnight-project-refactor.html new file mode 100644 index 00000000..2d43fca9 --- /dev/null +++ b/docs/showcases/cases/0623-overnight-project-refactor.html @@ -0,0 +1,138 @@ + + + + + + LoopX Showcase: 一晚上自主重构项目 + + + + +
+
+
+ + + +
2026-06-23 · repository-refactor
+

一晚上自主重构项目

+
长时间无值守重构能拆成适中 PR，而不是一个不可 review 大改。
+

The operator can wake up to reviewable PR-sized refactor slices instead of a single giant autonomous diff.

+
refactorPR slicestodo evidence
+ +
+ + + + LoopX + + + + + 一晚上自主重构项目 + loop + public safe pattern case + goal / trigger + todo / gate / evidence + public outcome + + + + + + + +
+ + +
01

案例背景

+

这个案例解决的是无人值守重构的风险：长时间 agent 容易把 cleanup、行为改变、发现的新问题和过期计划混成一个大 diff。

LoopX 的公开证据不是某个私有夜间截图，而是 todo lifecycle、successor/supersede、validation writeback 和 review-packet 这些已经在仓库里有文档和 smoke 的控制面。

+ +
02

仓库证据

+
+
证明点

无人值守 refactor 可以拆成 PR-sized slice。

+
LoopX 介入

loop、todo follow-up、supersede、PR-sized slices

+
+ +
successor path

`docs/lark-kanban-control-plane-adapter.md` 明确 real successor 使用 `todo complete --next-*`，replacement 或 narrower split 使用 `todo supersede --next-agent-todo`。

side-agent completion

`docs/heartbeat-automation-prompt.md` 要求非平凡完成创建 successor todo 或写 no-follow-up rationale。

CLI validation

`examples/todo-lifecycle-cli-smoke.py` 覆盖 `--next-agent-todo`、`todo supersede`、claim 继承和 handoff successor。

review shape

`loopx review-packet` 把当前 open todo、claimed_by 和 handoff 状态打包成 reviewer 可读的 packet。

+ +
03

LoopX 行为

+
  • 1
    当前 refactor slice 必须是可 review 的单位，不把整夜发现都塞进一个 PR。
  • 2
    发现的新工作写成 follow-up todo；路线变了就 supersede 旧 todo。
  • 3
    每个 slice 用 focused validation 或文档/contract smoke 证明，而不是依赖原始 agent trace。
  • 4
    大范围或风险不清楚的 slice 不自合并，进入 review handoff。
+ +
04

用户看到什么

+

用户可以让重构跑过夜，但早上看到的是一组有边界的 review 单元和剩余 todo，而不是一个不可 review 的巨型改动。

项目可以继续快，但 review 面仍是人能处理的粒度。

+ +
05

仓库来源

+
kanban control-plane adapterdocs/lark-kanban-control-plane-adapter.mdheartbeat prompt contractdocs/heartbeat-automation-prompt.mdtodo lifecycle smokeexamples/todo-lifecycle-cli-smoke.pycase narrativedocs/showcases/cases/0623-overnight-project-refactor.md
+

证据边界. Public-safe pattern case only; no private screenshots, raw chats, internal planning notes, local paths, credentials, raw logs, or unpublished project artifacts.

+
案例说明
+ +
由 docs/showcases/showcase-catalog.json 生成。不包含私有链接、原始聊天、本地状态或内部媒体。
+
+
+ + diff --git a/docs/showcases/cases/0623-overnight-project-refactor.md b/docs/showcases/cases/0623-overnight-project-refactor.md index 6e36789c..6e66511f 100644 --- a/docs/showcases/cases/0623-overnight-project-refactor.md +++ b/docs/showcases/cases/0623-overnight-project-refactor.md @@ -49,7 +49,7 @@ local paths, credentials, raw logs, and unpublished project artifacts. Public evidence should come from the resulting PR-sized diffs, validation commands, and follow-up/supersede state, not from raw agent traces. -## Website Story Beats +## Public Evidence Sequence 1. A broad refactor starts as a long-running goal. 2. LoopX keeps the current slice explicit. diff --git a/docs/showcases/cases/0624-pr-issue-auto-fix.en.html b/docs/showcases/cases/0624-pr-issue-auto-fix.en.html new file mode 100644 index 00000000..07f0aacb --- /dev/null +++ b/docs/showcases/cases/0624-pr-issue-auto-fix.en.html @@ -0,0 +1,138 @@ + + + + + + LoopX Showcase: PR issue automatic fix loop + + + + +
+
+
+ + + +
2026-06-24 · issue-fix-workflow
+

PR issue automatic fix loop

+
Review feedback should become an ordered repair workflow with repro, fix, validation, and reviewer handoff.
+

The operator can turn a PR issue into a controlled fix loop without manually rewriting the review comment as an agent plan.

+
issue fixrepro smokePR review
+ +
+ + + + LoopX + + + + + PR issue automatic fix + issue-fix workflow + public safe pattern case + goal / trigger + todo / gate / evidence + public outcome + + + + + + + +
+ + +
01

Case context

+

This case turns a PR issue, review comment, or issue text into an executable repair loop. It is not simply reading a comment and editing code; the path goes through metadata/intake, repro, branch-local patch, validation, and review packet.

The repository already contains the issue-fix capability: `loopx/capabilities/issue_fix/`, the `loopx issue-fix ...` CLI entry, protocol docs, and public smokes.

+ +
02

Repository evidence

+
+
Proof

Issue and review feedback can enter an executable repair loop.

+
LoopX intervention

issue-fix workflow, command pack, repro smoke, PR review feedback

+
+ +
Product entry

`docs/capabilities/issue-fix/README.md` names `loopx issue-fix ...`, the content-ops bridge, protocol docs, and smokes.

Workflow contract

`issue_fix_workflow_contract_v0` defines metadata preview, intake classification, workflow plan, todo writeback, caller repo branch, validation, PR review packet, and gate handling.

Executable loop

`issue_fix_acceptance_loop_v0` includes the acceptance fixture, repo-branch fixture, and caller-approved repo branch mode.

Validation surface

Smokes cover workflow planning, workflow contract, metadata/intake, and acceptance-loop behavior.

+ +
03

LoopX behavior

+
  • 1
    Public metadata can enter packets; raw issue bodies, comment bodies, timelines, and provider payloads remain gated sources.
  • 2
    Accepted candidates become ordered LoopX todos: repro smoke, code-context route, branch-local patch, validation, and review-packet readiness.
  • 3
    Caller-approved repo mode inspects the local repo only under `--execute` and reports repo-relative changed files plus validation pass/fail.
  • 4
    External comments, PR creation, merge, publish, destructive git, and production action remain explicit gates.
+ +
04

What the user sees

+

A maintainer can provide a public issue or PR signal and get a loop that reproduces, fixes, validates, and prepares human review.

The user does not need to rewrite every review comment as an agent prompt, and raw issue bodies or local paths do not leak into the public artifact.

+ +
05

Repository sources

+
capability READMEdocs/capabilities/issue-fix/README.mdworkflow contractdocs/capabilities/issue-fix/protocols/issue-fix-workflow-contract-v0.mdacceptance loopdocs/capabilities/issue-fix/protocols/issue-fix-acceptance-loop-v0.mdworkflow smokeexamples/issue-fix-workflow-plan-smoke.pyacceptance smokeexamples/issue-fix-acceptance-loop-smoke.py
+

Evidence boundary. Public-safe pattern case only; no private screenshots, raw gated issue bodies, internal review notes, local paths, raw logs, credentials, or unpublished repository artifacts.

+
Case note
+ +
Generated from docs/showcases/showcase-catalog.json. Private links, raw chats, local state, and internal media are excluded.
+
+
+ + diff --git a/docs/showcases/cases/0624-pr-issue-auto-fix.html b/docs/showcases/cases/0624-pr-issue-auto-fix.html new file mode 100644 index 00000000..ba8587fb --- /dev/null +++ b/docs/showcases/cases/0624-pr-issue-auto-fix.html @@ -0,0 +1,138 @@ + + + + + + LoopX Showcase: PR Issue 自动 Fix + + + + +
+
+
+ + + +
2026-06-24 · issue-fix-workflow
+

PR Issue 自动 Fix

+
Issue / Review comment 可以进入可执行修复闭环。
+

The operator can turn a PR issue into a controlled fix loop without manually rewriting the review comment as an agent plan.

+
issue fixrepro smokePR review
+ +
+ + + + LoopX + + + + + PR Issue 自动 Fix + issue-fix workflow + public safe pattern case + goal / trigger + todo / gate / evidence + public outcome + + + + + + + +
+ + +
01

案例背景

+

这个案例把 PR issue、review comment 或 issue text 转成可执行修复闭环。它不是“读一段评论就改代码”，而是先做 metadata/intake、repro、branch-local patch、validation 和 review packet。

仓库已经有 issue-fix capability：模块在 `loopx/capabilities/issue_fix/`，CLI 入口是 `loopx issue-fix ...`，协议文档和 smoke 都在公开仓库内。

+ +
02

仓库证据

+
+
证明点

Issue 和 review 反馈可以变成受控、可执行的修复循环。

+
LoopX 介入

issue-fix workflow、command pack、repro smoke、PR review feedback

+
+ +
产品入口

`docs/capabilities/issue-fix/README.md` 声明 `loopx issue-fix ...`、content-ops bridge、protocol docs 和 smoke。

工作流协议

`issue_fix_workflow_contract_v0` 明确 metadata preview、intake classification、workflow plan、todo writeback、caller repo branch、validation、PR review packet 和 gate handling。

可执行闭环

`issue_fix_acceptance_loop_v0` 包含 acceptance fixture、repo-branch fixture 和 caller-approved repo branch mode。

验证面

`examples/issue-fix-workflow-plan-smoke.py`、`examples/issue-fix-workflow-contract-smoke.py`、`examples/issue-fix-acceptance-loop-smoke.py` 等 smoke 覆盖这个路径。

+ +
03

LoopX 行为

+
  • 1
    public metadata 可以进入 packet；raw issue body、comment body、timeline、provider payload 都是 gated source。
  • 2
    accepted candidates 写成有序 LoopX todos：repro smoke、code-context route、branch-local patch、validation、review-packet readiness。
  • 3
    caller-approved repo 模式只在 `--execute` 时检查本地 repo，并且输出 repo-relative changed files、validation pass/fail 和 PR-readiness。
  • 4
    外部 comment、PR creation、merge、publish、destructive git 和 production action 都保留为显式 gate。
+ +
04

用户看到什么

+

维护者给出一个 public issue/PR signal 后，可以得到一个先复现、再修复、再验证、最后交给 review 的闭环。

用户不需要把每个 review comment 手动改写成 agent prompt，也不会把原始 issue body 或本地路径泄漏进公开 artifact。

+ +
05

仓库来源

+
capability READMEdocs/capabilities/issue-fix/README.mdworkflow contractdocs/capabilities/issue-fix/protocols/issue-fix-workflow-contract-v0.mdacceptance loopdocs/capabilities/issue-fix/protocols/issue-fix-acceptance-loop-v0.mdworkflow smokeexamples/issue-fix-workflow-plan-smoke.pyacceptance smokeexamples/issue-fix-acceptance-loop-smoke.py
+

证据边界. Public-safe pattern case only; no private screenshots, raw gated issue bodies, internal review notes, local paths, raw logs, credentials, or unpublished repository artifacts.

+
案例说明
+ +
由 docs/showcases/showcase-catalog.json 生成。不包含私有链接、原始聊天、本地状态或内部媒体。
+
+
+ + diff --git a/docs/showcases/cases/0624-pr-issue-auto-fix.md b/docs/showcases/cases/0624-pr-issue-auto-fix.md index 0cee4c91..cb42c439 100644 --- a/docs/showcases/cases/0624-pr-issue-auto-fix.md +++ b/docs/showcases/cases/0624-pr-issue-auto-fix.md @@ -50,7 +50,7 @@ internal review notes, local paths, raw logs, credentials, and unpublished repository artifacts. Public evidence should be the sanitized workflow plan, focused smoke, branch diff, and public PR review outcome. -## Website Story Beats +## Public Evidence Sequence 1. A PR issue or review comment appears. 2. LoopX classifies the issue and creates ordered repair todos. diff --git a/docs/showcases/cases/0627-overnight-pr-batch.en.html b/docs/showcases/cases/0627-overnight-pr-batch.en.html new file mode 100644 index 00000000..14aa1760 --- /dev/null +++ b/docs/showcases/cases/0627-overnight-pr-batch.en.html @@ -0,0 +1,138 @@ + + + + + + LoopX Showcase: Overnight PR batch with reviewable control + + + + +
+
+
+ + + +
2026-06-27 · agent-platform-self-improvement
+

Overnight PR batch with reviewable control

+
An overnight LoopX run can produce many PR-sized slices while keeping review, validation, and public evidence boundaries visible.
+

The operator can wake up to a compact batch of merged public slices, see what changed, and still trust that gates, validation, and evidence boundaries were not bypassed.

+
PR batchreview packetpublic boundary
+ +
+ + + + LoopX + + + + + Overnight PR batch wit + todo claim + public evidence case + goal / trigger + todo / gate / evidence + public outcome + + + + + + + +
+ + +
01

Case context

+

This case is not about an agent being busy. It is about keeping high-throughput autonomous work reviewable. The repository anchors the public evidence window to 2026-06-27 01:29 through 11:29 +08:00 instead of relying on private queues, screenshots, or operator notes.

Public Git history in that window shows 22 merged commits, 60 touched files, 6695 insertions, 223 deletions, and 10 commit messages with explicit PR numbers. The page only claims what those public facts support.

+ +
02

Repository evidence

+
+
Proof

High-throughput multi-lane work can remain PR-sized, reviewable, and merge-safe.

+
LoopX intervention

todo claim, review packet, self-merge boundary, focused smoke, public-boundary scan

+
+
22merged PR commits
10hpublic window
+
Public window

2026-06-27 01:29 to 11:29 +08:00, a 10-hour Git evidence window.

Change shape

22 merged commits across docs, runtime, status/quota, benchmark contracts, smokes, and release/runtime guardrails.

Review boundary

Work landed as PR-sized slices instead of one broad diff that maintainers would have to trust blindly.

Reproduction

The case document gives `git log --since ... --until ...` and `git log --numstat` commands.

+ +
03

LoopX behavior

+
  • 1
    LoopX ties each lane to todo ownership, validation, review policy, and public evidence.
  • 2
    Runtime, docs, and focused smokes move together when a control-plane contract changes.
  • 3
    Self-merge remains limited to narrow validated changes; larger or unclear work preserves review gates.
  • 4
    Public-boundary scans keep internal screenshots, local state, raw logs, and private planning out of the showcase.
+ +
04

What the user sees

+

The user wakes up to reviewable, traceable, explainable public slices rather than raw agent output.

The operator can inspect validation and remaining gates for each slice without relying on chat memory.

+ +
05

Repository sources

+
case narrativedocs/showcases/cases/0627-overnight-pr-batch.mdcatalog workload signaldocs/showcases/showcase-catalog.json
+

Evidence boundary. Public Git evidence only; no private documents, internal screenshots, raw chats, local active-state bodies, raw logs, credentials, or machine-specific paths.

+
Case note
+ +
Generated from docs/showcases/showcase-catalog.json. Private links, raw chats, local state, and internal media are excluded.
+
+
+ + diff --git a/docs/showcases/cases/0627-overnight-pr-batch.html b/docs/showcases/cases/0627-overnight-pr-batch.html new file mode 100644 index 00000000..a6eb4024 --- /dev/null +++ b/docs/showcases/cases/0627-overnight-pr-batch.html @@ -0,0 +1,138 @@ + + + + + + LoopX Showcase: 一晚 30 个高价值 PR + + + + +
+
+
+ + + +
2026-06-27 · agent-platform-self-improvement
+

一晚 30 个高价值 PR

+
高吞吐、多车道推进仍保持 PR 级可 review、可合并。
+

The operator can wake up to a compact batch of merged public slices, see what changed, and still trust that gates, validation, and evidence boundaries were not bypassed.

+
PR batchreview packetpublic boundary
+ +
+ + + + LoopX + + + + + 一晚 30 个高价值 PR + todo claim + public evidence case + goal / trigger + todo / gate / evidence + public outcome + + + + + + + +
+ + +
01

案例背景

+

这个案例不是在展示“agent 很忙”，而是在展示高吞吐自动化如何仍然保持可审阅。仓库把证据窗口固定在 2026-06-27 01:29 到 11:29 +08:00，避免把私有队列、聊天截图或操作员笔记当成公开证明。

窗口内公开 Git 历史显示 22 个 merged commits、60 个 touched files、6695 行新增和 223 行删除，其中 10 条 commit message 带 PR 编号。页面只声称这些公开 Git 能复现的事实。

+ +
02

仓库证据

+
+
证明点

高吞吐多 lane 工作也可以保持 PR 粒度、可审阅、可合并。

+
LoopX 介入

todo claim、review packet、自合并边界、focused smoke、public-boundary scan

+
+
22merged PR commits
10hpublic window
+
公开窗口

2026-06-27 01:29 到 11:29 +08:00，10 小时 Git 证据窗口。

变更粒度

22 个 merged commits 覆盖 docs、runtime、status/quota、benchmark contracts、smokes 和 release/runtime guardrails。

审阅边界

工作以 PR-sized slices 落地，而不是一个需要 maintainer 盲信的巨大 diff。

复现方式

case 文档给出 `git log --since ... --until ...` 和 `git log --numstat` 命令。

+ +
03

LoopX 行为

+
  • 1
    LoopX 把每条 lane 绑定到 todo ownership、validation、review policy 和 public evidence。
  • 2
    当 control-plane contract 改变时，runtime、docs 和 focused smoke 一起移动。
  • 3
    self-merge 只用于窄而验证过的变更；更大或不清楚的工作仍保留 review gate。
  • 4
    公开边界扫描把内部截图、本地状态、原始日志和私有计划排除在 showcase 之外。
+ +
04

用户看到什么

+

用户醒来看到的是一组可 review、可追踪、可解释的 public slices，而不是一堆原始 agent 输出。

operator 可以继续追问每条 slice 的验证和剩余 gate；证据不依赖 chat memory。

+ +
05

仓库来源

+
case narrativedocs/showcases/cases/0627-overnight-pr-batch.mdcatalog workload signaldocs/showcases/showcase-catalog.json
+

证据边界. Public Git evidence only; no private documents, internal screenshots, raw chats, local active-state bodies, raw logs, credentials, or machine-specific paths.

+
案例说明
+ +
由 docs/showcases/showcase-catalog.json 生成。不包含私有链接、原始聊天、本地状态或内部媒体。
+
+
+ + diff --git a/docs/showcases/cases/0627-overnight-pr-batch.md b/docs/showcases/cases/0627-overnight-pr-batch.md index e62735e4..d3af9415 100644 --- a/docs/showcases/cases/0627-overnight-pr-batch.md +++ b/docs/showcases/cases/0627-overnight-pr-batch.md @@ -91,7 +91,7 @@ showcase of reviewable control-plane throughput in one public repository at one point in time. Future versions can strengthen the case by linking each public PR number to its validation evidence and review outcome. -## Website Story Beats +## Public Evidence Sequence 1. A long-running LoopX project enters an overnight autonomous work window. 2. Many small slices land across runtime, docs, benchmark contracts, smokes, and diff --git a/docs/showcases/frontend-surface.md b/docs/showcases/frontend-surface.md index e0461996..da20d3c9 100644 --- a/docs/showcases/frontend-surface.md +++ b/docs/showcases/frontend-surface.md @@ -38,7 +38,7 @@ Use these catalog fields directly: | `frontend_card.visual_metaphor` | Suggested visual treatment. | | `frontend_card.primary_metric_hint` | Lightweight signal, not a hard claim. | | `frontend_card.badges` | Compact chips. | -| `frontend_card.story_beats` | Case detail timeline. | +| `frontend_card.story_beats` | Backward-compatible field for the case detail evidence sequence. New copy should render it as evidence, not as author notes. | | `workload_signal.efficiency_model` | Optional evidence panel for conservative baseline-vs-actual efficiency modeling. | ## First Screen @@ -109,7 +109,7 @@ pattern, not a transcript. ## Detail Story Model -A case detail view should be a short visual story: +A case detail view should be a compact evidence sequence: 1. **Trigger**: what made the long-running goal hard to manage. 2. **Visible State**: which LoopX objects made the situation explicit. diff --git a/docs/showcases/index.en.html b/docs/showcases/index.en.html new file mode 100644 index 00000000..4bd0c0c5 --- /dev/null +++ b/docs/showcases/index.en.html @@ -0,0 +1,201 @@ + + + + + + LoopX Showcase & Good Case + + + + +
+
+
+ + + +
LoopX · public-safe case surface
+

Showcase & Good Case

+
Real LoopX cases showing how long-running agent work stays reviewable, verifiable, and safe to continue.
+

LoopX preserves goals, gates, todos, claims, quota, run history, and evidence outside any single agent session.

+ +
+ + + + LoopX + + + + + Showcase & Good Case + catalog + public surface + goal / trigger + todo / gate / evidence + public outcome + + + + + + + +
+ + +
01

Top showcase cases

+ +
+ +
2026-06-27 · public_evidence_case
+

Overnight PR batch with reviewable control

+

An overnight LoopX run can produce many PR-sized slices while keeping review, validation, and public evidence boundaries visible.

+

Proof: High-throughput multi-lane work can remain PR-sized, reviewable, and merge-safe.

+
high_throughput_reviewable_workpr_sized_slicesself_merge_policyvalidation_writeback
+ + + + +
2026-06-24 · public_safe_pattern_case
+

PR issue automatic fix loop

+

Review feedback should become an ordered repair workflow with repro, fix, validation, and reviewer handoff.

+

Proof: Issue and review feedback can enter an executable repair loop.

+
issue_fix_workflowreview_feedbackrepro_smokecommand_pack
+ + + + +
2026-06-23 · public_safe_pattern_case
+

Agent-to-agent PR comment and fix loop

+

PR review feedback can become an owned agent todo with fix evidence instead of a loose chat reminder.

+

Proof: Multiple agents can coordinate around PR review comments without losing owner review.

+
agent_to_agent_handoffpr_comment_loopreview_packetclaimed_todo
+ + + + +
2026-06-23 · public_safe_pattern_case
+

Overnight project refactor as PR-sized slices

+

A broad refactor can run overnight while staying split into human-sized review units.

+

Proof: Long unattended refactors can split into moderate PR slices instead of one unreviewable diff.

+
long_unattended_goalpr_sized_slicestodo_follow_upsupersede
+ + + + +
2026-06-19 · public_safe_interactive_case
+

Dynamic workflow for hardware-agent development

+

A fuzzy long-running engineering goal needs a shared control plane when multiple worker agents participate.

+

Proof: Fuzzy goals, multiple workers, and long unattended runs can still converge.

+
dynamic_workflowmulti_agent_coordinationshared_control_planelong_unattended_goal
+ + + + +
2026-06-19 · public_evidence_case
+

LoopX self-iteration loop

+

A high-churn LoopX repo stayed legible while benchmark, product, docs, planning, and side-agent lanes moved in parallel.

+

Proof: A high-churn multi-lane project can keep state, boundaries, and evidence coherent.

+
self_iterationside_agent_scopetodo_claim_ownershipidentity_aware_prompt
+ + + + +
2026-06-17 · reproducible_synthetic_demo
+

Blocked P0 with safe P1/P2 rotation

+

A gated P0 lane should not stall a whole long-running goal when safe fallback work exists.

+

Proof: A user decision should not block all safe work.

+
blocked_priority_fallbackconcrete_user_gatesafe_fallback_workquota_discipline
+ +
+ +
02

Appendix case

+
+ +
2026-06-20 · public_safe_case_spec
+

Creator-operator long-running agent case

+

A creator-operator needs a long-running agent loop that keeps research moving while publishing decisions stay gated.

+

Proof: A creator-operator needs a long-running agent loop that keeps research moving while publishing decisions stay gated.

+
creator_operator_workflowgate_aware_continuationfeedback_capturesafe_side_path
+ +
+
Generated from docs/showcases/showcase-catalog.json. Private links, raw chats, local state, and internal media are excluded.
+
+
+ + + diff --git a/docs/showcases/index.html b/docs/showcases/index.html new file mode 100644 index 00000000..a92b4582 --- /dev/null +++ b/docs/showcases/index.html @@ -0,0 +1,201 @@ + + + + + + LoopX Showcase & Good Case + + + + +
+
+
+ + + +
LoopX · public-safe case surface
+

Showcase & Good Case

+
真实 LoopX 案例：长程 agent 工作如何保持可审阅、可验证、可继续推进。
+

LoopX preserves goals, gates, todos, claims, quota, run history, and evidence outside any single agent session.

+ +
+ + + + LoopX + + + + + Showcase & Good Case + catalog + public surface + goal / trigger + todo / gate / evidence + public outcome + + + + + + + +
+ + +
01

顶部 Showcase 案例

+ +
+ +
2026-06-27 · public_evidence_case
+

一晚 30 个高价值 PR

+

高吞吐、多车道推进仍保持 PR 级可 review、可合并。

+

证明点: 高吞吐多 lane 工作也可以保持 PR 粒度、可审阅、可合并。

+
high_throughput_reviewable_workpr_sized_slicesself_merge_policyvalidation_writeback
+ + + + +
2026-06-24 · public_safe_pattern_case
+

PR Issue 自动 Fix

+

Issue / Review comment 可以进入可执行修复闭环。

+

证明点: Issue 和 review 反馈可以变成受控、可执行的修复循环。

+
issue_fix_workflowreview_feedbackrepro_smokecommand_pack
+ + + + +
2026-06-23 · public_safe_pattern_case
+

Agent to agent 回复 PR comment 和 PR Fix

+

多 agent 可围绕 PR review 协同发现、评论、修复。

+

证明点: 多 agent lane 可以围绕 PR comment 协作，同时保留 owner review。

+
agent_to_agent_handoffpr_comment_loopreview_packetclaimed_todo
+ + + + +
2026-06-23 · public_safe_pattern_case
+

一晚上自主重构项目

+

长时间无值守重构能拆成适中 PR，而不是一个不可 review 大改。

+

证明点: 无人值守 refactor 可以拆成 PR-sized slice。

+
long_unattended_goalpr_sized_slicestodo_follow_upsupersede
+ + + + +
2026-06-19 · public_safe_interactive_case
+

外部芯片 agent workflow

+

模糊目标、多 worker、长时间无值守下仍可收敛。

+

证明点: 模糊目标、多 worker、长时间无值守下仍可收敛。

+
dynamic_workflowmulti_agent_coordinationshared_control_planelong_unattended_goalCanonical 原页面
+ + + + +
2026-06-19 · public_evidence_case
+

LoopX Meta Agent 自迭代

+

高变更、多车道项目可保持状态、边界和证据。

+

证明点: 高 churn 多 lane agent repo 可以保持状态、证据和边界一致。

+
self_iterationside_agent_scopetodo_claim_ownershipidentity_aware_prompt
+ + + + +
2026-06-17 · reproducible_synthetic_demo
+

P0 block 后推进 P1/P2

+

用户决策不应阻塞全部安全工作。

+

证明点: 被阻塞的 P0 决策不应该阻止安全的 P1/P2 工作继续。

+
blocked_priority_fallbackconcrete_user_gatesafe_fallback_workquota_discipline
+ +
+ +
02

附录案例

+
+ +
2026-06-20 · public_safe_case_spec
+

创作者-运营者长跑 Agent 案例

+

研究可以继续，发布决策仍然 gated。

+

证明点: 创作与运营工作可以共享一个 gate-aware 的长期 agent loop。

+
creator_operator_workflowgate_aware_continuationfeedback_capturesafe_side_path
+ +
+
由 docs/showcases/showcase-catalog.json 生成。不包含私有链接、原始聊天、本地状态或内部媒体。
+
+
+ + + diff --git a/docs/showcases/showcase-animation-prototype.html b/docs/showcases/showcase-animation-prototype.html index a25f4ccc..0b676f24 100644 --- a/docs/showcases/showcase-animation-prototype.html +++ b/docs/showcases/showcase-animation-prototype.html @@ -247,54 +247,70 @@ 0% { width: 0%; } 100% { width: 100%; } } - + .scene--opening-control-plane { animation: scene-opening-control-plane 30s linear infinite; } @keyframes scene-opening-control-plane { 0%, 0.0000% { opacity: 0; transform: translateY(18px) scale(0.985); pointer-events: none; } 2.3333%, 11.0000% { opacity: 1; transform: translateY(0) scale(1); pointer-events: auto; } 13.3333%, 100% { opacity: 0; transform: translateY(-14px) scale(0.99); pointer-events: none; } } - - .scene--gate-visible-safe-lane-moves { animation: scene-gate-visible-safe-lane-moves 30s linear infinite; } - @keyframes scene-gate-visible-safe-lane-moves { + + .scene--high-throughput-reviewable-prs { animation: scene-high-throughput-reviewable-prs 30s linear infinite; } + @keyframes scene-high-throughput-reviewable-prs { 0%, 13.3000% { opacity: 0; transform: translateY(18px) scale(0.985); pointer-events: none; } - 15.6667%, 27.6667% { opacity: 1; transform: translateY(0) scale(1); pointer-events: auto; } - 30.0000%, 100% { opacity: 0; transform: translateY(-14px) scale(0.99); pointer-events: none; } + 15.6667%, 24.3333% { opacity: 1; transform: translateY(0) scale(1); pointer-events: auto; } + 26.6667%, 100% { opacity: 0; transform: translateY(-14px) scale(0.99); pointer-events: none; } + } + + + .scene--issue-comment-repair-loop { animation: scene-issue-comment-repair-loop 30s linear infinite; } + @keyframes scene-issue-comment-repair-loop { + 0%, 26.6333% { opacity: 0; transform: translateY(18px) scale(0.985); pointer-events: none; } + 29.0000%, 37.6667% { opacity: 1; transform: translateY(0) scale(1); pointer-events: auto; } + 40.0000%, 100% { opacity: 0; transform: translateY(-14px) scale(0.99); pointer-events: none; } } - + + + .scene--refactor-and-safe-fallback { animation: scene-refactor-and-safe-fallback 30s linear infinite; } + @keyframes scene-refactor-and-safe-fallback { + 0%, 39.9667% { opacity: 0; transform: translateY(18px) scale(0.985); pointer-events: none; } + 42.3333%, 51.0000% { opacity: 1; transform: translateY(0) scale(1); pointer-events: auto; } + 53.3333%, 100% { opacity: 0; transform: translateY(-14px) scale(0.99); pointer-events: none; } + } + .scene--multi-agent-shared-state { animation: scene-multi-agent-shared-state 30s linear infinite; } @keyframes scene-multi-agent-shared-state { - 0%, 29.9667% { opacity: 0; transform: translateY(18px) scale(0.985); pointer-events: none; } - 32.3333%, 44.3333% { opacity: 1; transform: translateY(0) scale(1); pointer-events: auto; } - 46.6667%, 100% { opacity: 0; transform: translateY(-14px) scale(0.99); pointer-events: none; } + 0%, 53.3000% { opacity: 0; transform: translateY(18px) scale(0.985); pointer-events: none; } + 55.6667%, 67.6667% { opacity: 1; transform: translateY(0) scale(1); pointer-events: auto; } + 70.0000%, 100% { opacity: 0; transform: translateY(-14px) scale(0.99); pointer-events: none; } } - + .scene--self-iteration-efficiency-evidence { animation: scene-self-iteration-efficiency-evidence 30s linear infinite; } @keyframes scene-self-iteration-efficiency-evidence { - 0%, 46.6333% { opacity: 0; transform: translateY(18px) scale(0.985); pointer-events: none; } - 49.0000%, 67.6667% { opacity: 1; transform: translateY(0) scale(1); pointer-events: auto; } - 70.0000%, 100% { opacity: 0; transform: translateY(-14px) scale(0.99); pointer-events: none; } + 0%, 69.9667% { opacity: 0; transform: translateY(18px) scale(0.985); pointer-events: none; } + 72.3333%, 84.3333% { opacity: 1; transform: translateY(0) scale(1); pointer-events: auto; } + 86.6667%, 100% { opacity: 0; transform: translateY(-14px) scale(0.99); pointer-events: none; } } - + .scene--creator-operator-safe-production-loop { animation: scene-creator-operator-safe-production-loop 30s linear infinite; } @keyframes scene-creator-operator-safe-production-loop { - 0%, 69.9667% { opacity: 0; transform: translateY(18px) scale(0.985); pointer-events: none; } - 72.3333%, 87.6667% { opacity: 1; transform: translateY(0) scale(1); pointer-events: auto; } - 90.0000%, 100% { opacity: 0; transform: translateY(-14px) scale(0.99); pointer-events: none; } + 0%, 86.6333% { opacity: 0; transform: translateY(18px) scale(0.985); pointer-events: none; } + 89.0000%, 94.3333% { opacity: 1; transform: translateY(0) scale(1); pointer-events: auto; } + 96.6667%, 100% { opacity: 0; transform: translateY(-14px) scale(0.99); pointer-events: none; } } - + .scene--boundary-close { animation: scene-boundary-close 30s linear infinite; } @keyframes scene-boundary-close { - 0%, 89.9667% { opacity: 0; transform: translateY(18px) scale(0.985); pointer-events: none; } - 92.3333%, 97.6667% { opacity: 1; transform: translateY(0) scale(1); pointer-events: auto; } + 0%, 96.6333% { opacity: 0; transform: translateY(18px) scale(0.985); pointer-events: none; } + 99.0000%, 99.0000% { opacity: 1; transform: translateY(0) scale(1); pointer-events: auto; } 100.0000%, 100% { opacity: 0; transform: translateY(-14px) scale(0.99); pointer-events: none; } } - + @media (max-width: 900px) { main { grid-template-columns: 1fr; @@ -344,39 +360,49 @@

From AI assist to async agent work.

Public catalog demo. No live registry, local status export, transcript, or private active state.
- + - + + + + - + data-scene-id="issue-comment-repair-loop" + style="--start:26.6667%; --width:13.3333%;"> + + + + - + style="--start:53.3333%; --width:16.6667%;"> + - + style="--start:70.0000%; --width:16.6667%;"> + - + style="--start:86.6667%; --width:10.0000%;"> + - + style="--start:96.6667%; --width:3.3333%;"> +
- +
One shared control plane for long-running agent work.

dark canvas with a single shared goal state node; agent lanes orbit into view

fade in the goal node, then pulse two safe lanes without implying live ops data

- +
Catalog-backed control plane The prototype reads the public storyboard and catalog only. public-boundaryno-live-status
- +
- -
+ data-end-seconds="8" + data-source-case-ids="2026-06-27-overnight-pr-batch">
- 04-09s + 04-08s Scene 2
-

A user gate stays visible. Safe side work still moves.

-

a gated P0 lane pauses while a safe side lane advances with a validation marker

-

keep the blocked lane still; animate only the safe lane and evidence writeback

+

High-throughput agent work can stay reviewable.

+

many small PR-sized lanes converge into a compact review rail with validation markers

+

animate several small lanes into one review rail; avoid raw commit firehoses

- + +
+ Overnight PR batch with reviewable control + An overnight LoopX run can produce many PR-sized slices while keeping review, validation, and public evidence boundaries visible. + PR batchreview packetpublic boundary +
+ +
+
+ + +
+
+ 08-12s + Scene 3 +
+

Review feedback becomes owned repair work with evidence.

+

a PR comment becomes a claimed todo, then a repro, fix, validation, and review packet

+

show claim, repro, fix, validation, and handoff as a short packet sequence

+
+ +
+ PR issue automatic fix loop + Review feedback should become an ordered repair workflow with repro, fix, validation, and reviewer handoff. + issue fixrepro smokePR review +
+ + +
+ Agent-to-agent PR comment and fix loop + PR review feedback can become an owned agent todo with fix evidence instead of a loose chat reminder. + handoffPR commentclaimed_by +
+ +
+
+ + +
+
+ 12-16s + Scene 4 +
+

Large work splits. Blocked routes do not freeze safe progress.

+

a broad refactor splits into successor slices while a gated P0 lane pauses and a safe fallback advances

+

show supersede/follow-up chips beside a still-visible user gate

+
+ +
+ Overnight project refactor as PR-sized slices + A broad refactor can run overnight while staying split into human-sized review units. + refactorPR slicestodo evidence +
+ +
Blocked P0 with safe P1/P2 rotation A gated P0 lane should not stall a whole long-running goal when safe fallback work exists. reproducibleuser-gatefallback
- +
- +
- 09-14s - Scene 3 + 16-21s + Scene 5

Multiple agents can share one project memory without guessing each other's state.

primary and side-agent lanes claim separate todos against one catalog-backed board

show claims and handoff lines as lightweight beams into shared state

- +
Dynamic workflow for hardware-agent development A fuzzy long-running engineering goal needs a shared control plane when multiple worker agents participate. interactive-htmlmulti-agenthardware
- +
- +
- 14-21s - Scene 4 + 21-26s + Scene 6

Async side work turns repo history into recoverable product momentum.

repo-scale requirement clusters compress into validated commits, docs, smokes, and frontstage surfaces

avoid raw commit firehoses; use compact clusters and conservative evidence labels

- +
LoopX self-iteration loop A high-churn LoopX repo stayed legible while benchmark, product, docs, planning, and side-agent lanes moved in parallel. self-iterationcommit-backedside-agent
- +
- +
- 21-27s - Scene 5 + 26-29s + Scene 7

The agent can keep researching while human judgment guards what ships.

research, preference memory, and draft material move while publishing remains gated

animate safe research cards; leave publish decision locked until the final frame

- +
Creator-operator long-running agent case A creator-operator needs a long-running agent loop that keeps research moving while publishing decisions stay gated.
- +
- +
- 27-30s - Scene 6 + 29-30s + Scene 8

Public demos read the showcase catalog, not private control-plane state.

public showcase catalog badge replaces the live status feed

close on the catalog badge and a clean public/private boundary line

- +
Catalog-backed control plane The prototype reads the public storyboard and catalog only. public-boundaryno-live-status
- +
- +
diff --git a/docs/showcases/showcase-animation-storyboard.json b/docs/showcases/showcase-animation-storyboard.json index 61c74993..24653b15 100644 --- a/docs/showcases/showcase-animation-storyboard.json +++ b/docs/showcases/showcase-animation-storyboard.json @@ -37,16 +37,32 @@ "motion_notes": "fade in the goal node, then pulse two safe lanes without implying live ops data" }, { - "id": "gate-visible-safe-lane-moves", - "time_seconds": [4, 9], - "source_case_ids": ["2026-06-17-blocked-p0-safe-rotation"], - "visual": "a gated P0 lane pauses while a safe side lane advances with a validation marker", - "copy": "A user gate stays visible. Safe side work still moves.", - "motion_notes": "keep the blocked lane still; animate only the safe lane and evidence writeback" + "id": "high-throughput-reviewable-prs", + "time_seconds": [4, 8], + "source_case_ids": ["2026-06-27-overnight-pr-batch"], + "visual": "many small PR-sized lanes converge into a compact review rail with validation markers", + "copy": "High-throughput agent work can stay reviewable.", + "motion_notes": "animate several small lanes into one review rail; avoid raw commit firehoses" + }, + { + "id": "issue-comment-repair-loop", + "time_seconds": [8, 12], + "source_case_ids": ["2026-06-24-pr-issue-auto-fix", "2026-06-23-agent-to-agent-pr-comments"], + "visual": "a PR comment becomes a claimed todo, then a repro, fix, validation, and review packet", + "copy": "Review feedback becomes owned repair work with evidence.", + "motion_notes": "show claim, repro, fix, validation, and handoff as a short packet sequence" + }, + { + "id": "refactor-and-safe-fallback", + "time_seconds": [12, 16], + "source_case_ids": ["2026-06-23-overnight-project-refactor", "2026-06-17-blocked-p0-safe-rotation"], + "visual": "a broad refactor splits into successor slices while a gated P0 lane pauses and a safe fallback advances", + "copy": "Large work splits. Blocked routes do not freeze safe progress.", + "motion_notes": "show supersede/follow-up chips beside a still-visible user gate" }, { "id": "multi-agent-shared-state", - "time_seconds": [9, 14], + "time_seconds": [16, 21], "source_case_ids": ["2026-06-19-dynamic-workflow-hardware-agent"], "visual": "primary and side-agent lanes claim separate todos against one catalog-backed board", "copy": "Multiple agents can share one project memory without guessing each other's state.", @@ -54,7 +70,7 @@ }, { "id": "self-iteration-efficiency-evidence", - "time_seconds": [14, 21], + "time_seconds": [21, 26], "source_case_ids": ["2026-06-19-loopx-self-iteration"], "visual": "repo-scale requirement clusters compress into validated commits, docs, smokes, and frontstage surfaces", "copy": "Async side work turns repo history into recoverable product momentum.", @@ -62,7 +78,7 @@ }, { "id": "creator-operator-safe-production-loop", - "time_seconds": [21, 27], + "time_seconds": [26, 29], "source_case_ids": ["2026-06-20-creator-operator-case-spec"], "visual": "research, preference memory, and draft material move while publishing remains gated", "copy": "The agent can keep researching while human judgment guards what ships.", @@ -70,7 +86,7 @@ }, { "id": "boundary-close", - "time_seconds": [27, 30], + "time_seconds": [29, 30], "source_case_ids": [], "visual": "public showcase catalog badge replaces the live status feed", "copy": "Public demos read the showcase catalog, not private control-plane state.", diff --git a/docs/showcases/showcase-catalog.json b/docs/showcases/showcase-catalog.json index ea69e29c..960fd34a 100644 --- a/docs/showcases/showcase-catalog.json +++ b/docs/showcases/showcase-catalog.json @@ -21,51 +21,312 @@ }, "cases": [ { - "id": "2026-06-17-blocked-p0-safe-rotation", - "date": "2026-06-17", - "title": "Blocked P0 with safe P1/P2 rotation", - "status": "reproducible_synthetic_demo", - "case_page": "docs/showcases/cases/0617-blocked-p0-safe-rotation.md", - "demo_command": "python3 examples/showcase-0617-blocked-p0-safe-rotation-smoke.py", - "domain": "benchmark-rotation", + "id": "2026-06-27-overnight-pr-batch", + "date": "2026-06-27", + "title": "Overnight PR batch with reviewable control", + "status": "public_evidence_case", + "case_page": "docs/showcases/cases/0627-overnight-pr-batch.md", + "demo_command": null, + "domain": "agent-platform-self-improvement", "audience": [ "operator", "agent-platform-developer", - "benchmark-developer" + "technical-lead" ], "pattern_tags": [ - "blocked_priority_fallback", - "concrete_user_gate", - "safe_fallback_work", - "quota_discipline", - "attention_reduction" + "high_throughput_reviewable_work", + "pr_sized_slices", + "self_merge_policy", + "validation_writeback", + "public_boundary" ], - "headline": "A gated P0 lane should not stall a whole long-running goal when safe fallback work exists.", - "problem": "A benchmark lane needed a large local dependency before it could continue, while other no-upload benchmark work remained safe.", + "headline": "An overnight LoopX run can produce many PR-sized slices while keeping review, validation, and public evidence boundaries visible.", + "problem": "High-throughput autonomous work is only useful if the resulting changes remain reviewable, validated, and safe to publish.", "loopx_behavior": [ - "surface the P0 dependency as a concrete user gate", - "avoid spending on the gated lane", - "select a non-dependent fallback lane", - "record both the blocker and the fallback reason" + "keep work broken into PR-sized slices instead of a giant unreviewable diff", + "tie runtime, docs, and focused smoke updates together when a control-plane contract changes", + "limit self-merge to narrow validated changes while preserving broader review gates", + "record public evidence from Git history instead of raw agent logs or private screenshots", + "keep public/private boundary checks in the showcase path" ], - "user_value": "The operator sees exactly what decision is needed while the agent can keep making bounded progress elsewhere.", - "evidence_boundary": "Synthetic public fixture only; no private screenshots, raw tasks, internal links, local image names, or raw run logs.", + "user_value": "The operator can wake up to a compact batch of merged public slices, see what changed, and still trust that gates, validation, and evidence boundaries were not bypassed.", + "workload_signal": { + "scope": "public_repository_window", + "window": { + "from": "2026-06-27T01:29:00+08:00", + "to": "2026-06-27T11:29:00+08:00", + "hours": 10 + }, + "public_git": { + "merged_commits": 22, + "files_touched": 60, + "insertions": 6695, + "deletions": 223, + "commit_messages_with_pr_numbers": 10 + }, + "claim_boundary": "public Git history only; contemporaneous private queue notes are not used as public evidence" + }, + "evidence_boundary": "Public Git evidence only; no private documents, internal screenshots, raw chats, local active-state bodies, raw logs, credentials, or machine-specific paths.", + "interactive_page": "docs/showcases/cases/0627-overnight-pr-batch.html", + "interactive_page_zh": "docs/showcases/cases/0627-overnight-pr-batch.html", + "interactive_page_en": "docs/showcases/cases/0627-overnight-pr-batch.en.html", + "localized_pages": { + "zh": "docs/showcases/cases/0627-overnight-pr-batch.html", + "en": "docs/showcases/cases/0627-overnight-pr-batch.en.html" + }, + "showcase_rank": 1, + "showcase_table": { + "proof_point": "High-throughput multi-lane work can remain PR-sized, reviewable, and merge-safe.", + "loopx_intervention": "todo claim, review packet, self-merge boundary, focused smoke, public-boundary scan" + }, "frontend_card": { - "visual_metaphor": "priority lanes with one gated lane and one active fallback lane", - "primary_metric_hint": "attention reduction: user sees one concrete decision instead of reading the whole run log", + "visual_metaphor": "parallel PR lanes converge into a reviewable merge rail", + "primary_metric_hint": "High-throughput multi-lane work can remain PR-sized, reviewable, and merge-safe.", "badges": [ - "reproducible", - "user-gate", - "fallback" + "PR batch", + "review packet", + "public boundary" ], "story_beats": [ - "P0 lane is blocked by a user decision", - "LoopX projects the decision as a user todo", - "agent continues safe fallback work", - "state records blocker, fallback, validation, and spend policy" + "keep work broken into PR-sized slices instead of a giant unreviewable diff", + "tie runtime, docs, and focused smoke updates together when a control-plane contract changes", + "limit self-merge to narrow validated changes while preserving broader review gates", + "record public evidence from Git history instead of raw agent logs or private screenshots" + ] + } + }, + { + "id": "2026-06-24-pr-issue-auto-fix", + "date": "2026-06-24", + "title": "PR issue automatic fix loop", + "status": "public_safe_pattern_case", + "case_page": "docs/showcases/cases/0624-pr-issue-auto-fix.md", + "demo_command": null, + "domain": "issue-fix-workflow", + "audience": [ + "operator", + "agent-platform-developer", + "open-source-maintainer" + ], + "pattern_tags": [ + "issue_fix_workflow", + "review_feedback", + "repro_smoke", + "command_pack", + "successor_todo" + ], + "headline": "Review feedback should become an ordered repair workflow with repro, fix, validation, and reviewer handoff.", + "problem": "PR comments and issues are often concrete enough to fix, but unsafe to feed into an agent as unstructured prompt text without routing, repro, and validation.", + "loopx_behavior": [ + "classify the issue or review feedback", + "create ordered repair todos", + "keep gated source reads explicit", + "separate repro, implementation, validation, and reviewer handoff" + ], + "user_value": "The operator can turn a PR issue into a controlled fix loop without manually rewriting the review comment as an agent plan.", + "evidence_boundary": "Public-safe pattern case only; no private screenshots, raw gated issue bodies, internal review notes, local paths, raw logs, credentials, or unpublished repository artifacts.", + "interactive_page": "docs/showcases/cases/0624-pr-issue-auto-fix.html", + "interactive_page_zh": "docs/showcases/cases/0624-pr-issue-auto-fix.html", + "interactive_page_en": "docs/showcases/cases/0624-pr-issue-auto-fix.en.html", + "localized_pages": { + "zh": "docs/showcases/cases/0624-pr-issue-auto-fix.html", + "en": "docs/showcases/cases/0624-pr-issue-auto-fix.en.html" + }, + "showcase_rank": 2, + "showcase_table": { + "proof_point": "Issue and review feedback can enter an executable repair loop.", + "loopx_intervention": "issue-fix workflow, command pack, repro smoke, PR review feedback" + }, + "frontend_card": { + "visual_metaphor": "issue feedback closes through repro, fix, validation, and review handoff", + "primary_metric_hint": "Issue and review feedback can enter an executable repair loop.", + "badges": [ + "issue fix", + "repro smoke", + "PR review" + ], + "story_beats": [ + "classify the issue or review feedback", + "create ordered repair todos", + "keep gated source reads explicit", + "separate repro, implementation, validation, and reviewer handoff" + ] + } + }, + { + "id": "2026-06-23-agent-to-agent-pr-comments", + "date": "2026-06-23", + "title": "Agent-to-agent PR comment and fix loop", + "status": "public_safe_pattern_case", + "case_page": "docs/showcases/cases/0623-agent-to-agent-pr-comments.md", + "demo_command": null, + "domain": "pull-request-review", + "audience": [ + "operator", + "agent-platform-developer", + "open-source-maintainer" + ], + "pattern_tags": [ + "agent_to_agent_handoff", + "pr_comment_loop", + "review_packet", + "claimed_todo", + "successor_todo" + ], + "headline": "PR review feedback can become an owned agent todo with fix evidence instead of a loose chat reminder.", + "problem": "Multiple agent lanes can see or act on the same PR feedback, but without ownership and handoff state the comment-to-fix loop becomes hard to audit.", + "loopx_behavior": [ + "turn review feedback into a claimed todo", + "route implementation through the owning agent lane", + "record fix and validation evidence in the review packet", + "keep successor work explicit after the comment is handled" + ], + "user_value": "The operator can let agents coordinate around PR comments without losing final review visibility or fix evidence.", + "evidence_boundary": "Public-safe pattern case only; no private screenshots, raw chats, internal review notes, local state, credentials, or unpublished artifacts.", + "interactive_page": "docs/showcases/cases/0623-agent-to-agent-pr-comments.html", + "interactive_page_zh": "docs/showcases/cases/0623-agent-to-agent-pr-comments.html", + "interactive_page_en": "docs/showcases/cases/0623-agent-to-agent-pr-comments.en.html", + "localized_pages": { + "zh": "docs/showcases/cases/0623-agent-to-agent-pr-comments.html", + "en": "docs/showcases/cases/0623-agent-to-agent-pr-comments.en.html" + }, + "showcase_rank": 3, + "showcase_table": { + "proof_point": "Multiple agents can coordinate around PR review comments without losing owner review.", + "loopx_intervention": "claimed_by, handoff gate, review packet, comment/fix loop" + }, + "frontend_card": { + "visual_metaphor": "agent comments pass through explicit ownership instead of loose threads", + "primary_metric_hint": "Multiple agents can coordinate around PR review comments without losing owner review.", + "badges": [ + "handoff", + "PR comment", + "claimed_by" + ], + "story_beats": [ + "turn review feedback into a claimed todo", + "route implementation through the owning agent lane", + "record fix and validation evidence in the review packet", + "keep successor work explicit after the comment is handled" + ] + } + }, + { + "id": "2026-06-23-overnight-project-refactor", + "date": "2026-06-23", + "title": "Overnight project refactor as PR-sized slices", + "status": "public_safe_pattern_case", + "case_page": "docs/showcases/cases/0623-overnight-project-refactor.md", + "demo_command": null, + "domain": "repository-refactor", + "audience": [ + "operator", + "agent-platform-developer", + "technical-lead" + ], + "pattern_tags": [ + "long_unattended_goal", + "pr_sized_slices", + "todo_follow_up", + "supersede", + "validation_writeback" + ], + "headline": "A broad refactor can run overnight while staying split into human-sized review units.", + "problem": "Autonomous refactors become risky when discoveries, stale tasks, cleanup, and behavior changes collapse into one broad diff.", + "loopx_behavior": [ + "keep the current refactor slice explicit", + "convert discoveries into follow-up todos", + "supersede stale tasks when the route changes", + "validate each slice before merge or handoff" + ], + "user_value": "The operator can wake up to reviewable PR-sized refactor slices instead of a single giant autonomous diff.", + "evidence_boundary": "Public-safe pattern case only; no private screenshots, raw chats, internal planning notes, local paths, credentials, raw logs, or unpublished project artifacts.", + "interactive_page": "docs/showcases/cases/0623-overnight-project-refactor.html", + "interactive_page_zh": "docs/showcases/cases/0623-overnight-project-refactor.html", + "interactive_page_en": "docs/showcases/cases/0623-overnight-project-refactor.en.html", + "localized_pages": { + "zh": "docs/showcases/cases/0623-overnight-project-refactor.html", + "en": "docs/showcases/cases/0623-overnight-project-refactor.en.html" + }, + "showcase_rank": 4, + "showcase_table": { + "proof_point": "Long unattended refactors can split into moderate PR slices instead of one unreviewable diff.", + "loopx_intervention": "loop, todo follow-up, supersede, PR-sized slices" + }, + "frontend_card": { + "visual_metaphor": "a large refactor moves as small reviewable packets", + "primary_metric_hint": "Long unattended refactors can split into moderate PR slices instead of one unreviewable diff.", + "badges": [ + "refactor", + "PR slices", + "todo evidence" + ], + "story_beats": [ + "keep the current refactor slice explicit", + "convert discoveries into follow-up todos", + "supersede stale tasks when the route changes", + "validate each slice before merge or handoff" ] } }, + { + "id": "2026-06-19-dynamic-workflow-hardware-agent", + "date": "2026-06-19", + "title": "Dynamic workflow for hardware-agent development", + "status": "public_safe_interactive_case", + "case_page": "docs/showcases/cases/0619-dynamic-workflow-hardware-agent.md", + "interactive_page": "docs/showcases/cases/0619-dynamic-workflow-hardware-agent.html", + "demo_command": null, + "domain": "hardware-agent-development", + "audience": [ + "agent-platform-developer", + "technical-lead", + "open-source-user" + ], + "pattern_tags": [ + "dynamic_workflow", + "multi_agent_coordination", + "shared_control_plane", + "long_unattended_goal", + "convergence" + ], + "headline": "A fuzzy long-running engineering goal needs a shared control plane when multiple worker agents participate.", + "problem": "Specialized engineering work can require multiple agents to split work and converge without losing ownership or state.", + "loopx_behavior": [ + "keep durable goal state outside any one chat thread", + "make ownership, quota, and evidence writeback explicit", + "let script-generated worker loops continue only after bounded validation", + "project convergence state and human gates for the operator" + ], + "user_value": "A contributor-approved public artifact shows how one control plane can coordinate Claude Code, generated scripts, and hardware-agent workers across five long-running engineering cases.", + "evidence_boundary": "Public-safe interactive artifact; no raw chats, screenshots, proprietary design details, private repositories, local paths, task ids, credentials, or unpublished hardware artifacts.", + "frontend_card": { + "visual_metaphor": "multiple worker lanes converging through one shared control plane", + "primary_metric_hint": "five public hardware-agent cases under one control plane", + "badges": [ + "interactive-html", + "multi-agent", + "hardware" + ], + "story_beats": [ + "LoopX owns goal state, quota, todos, claims, evidence, and history", + "Claude Code writes task-specific orchestration scripts under that contract", + "hardware-agent workers perform bounded RTL, simulation, and validation work", + "five public cases show closed tasks, DSE, flagship Fmax optimization, and convergence floors" + ] + }, + "interactive_page_zh": "docs/showcases/cases/0619-dynamic-workflow-hardware-agent.html", + "interactive_page_en": "docs/showcases/cases/0619-dynamic-workflow-hardware-agent.en.html", + "localized_pages": { + "zh": "docs/showcases/cases/0619-dynamic-workflow-hardware-agent.html", + "en": "docs/showcases/cases/0619-dynamic-workflow-hardware-agent.en.html" + }, + "showcase_rank": 5, + "showcase_table": { + "proof_point": "Fuzzy goals, multiple workers, and long unattended runs can still converge.", + "loopx_intervention": "goal state, worker handoff, dynamic workflow" + } + }, { "id": "2026-06-19-loopx-self-iteration", "date": "2026-06-19", @@ -179,222 +440,76 @@ "public Git history is converted into conservative product requirement clusters", "public docs and smokes turn the experience into a reusable case" ] + }, + "interactive_page": "docs/showcases/cases/0619-loopx-self-iteration.html", + "interactive_page_zh": "docs/showcases/cases/0619-loopx-self-iteration.html", + "interactive_page_en": "docs/showcases/cases/0619-loopx-self-iteration.en.html", + "localized_pages": { + "zh": "docs/showcases/cases/0619-loopx-self-iteration.html", + "en": "docs/showcases/cases/0619-loopx-self-iteration.en.html" + }, + "showcase_rank": 6, + "showcase_table": { + "proof_point": "A high-churn multi-lane project can keep state, boundaries, and evidence coherent.", + "loopx_intervention": "todo, quota, gate, evidence, review packet, frontstage" } }, { - "id": "2026-06-19-dynamic-workflow-hardware-agent", - "date": "2026-06-19", - "title": "Dynamic workflow for hardware-agent development", - "status": "public_safe_interactive_case", - "case_page": "docs/showcases/cases/0619-dynamic-workflow-hardware-agent.md", - "interactive_page": "docs/showcases/cases/0619-dynamic-workflow-hardware-agent.html", - "demo_command": null, - "domain": "hardware-agent-development", + "id": "2026-06-17-blocked-p0-safe-rotation", + "date": "2026-06-17", + "title": "Blocked P0 with safe P1/P2 rotation", + "status": "reproducible_synthetic_demo", + "case_page": "docs/showcases/cases/0617-blocked-p0-safe-rotation.md", + "demo_command": "python3 examples/showcase-0617-blocked-p0-safe-rotation-smoke.py", + "domain": "benchmark-rotation", "audience": [ + "operator", "agent-platform-developer", - "technical-lead", - "open-source-user" + "benchmark-developer" ], "pattern_tags": [ - "dynamic_workflow", - "multi_agent_coordination", - "shared_control_plane", - "long_unattended_goal", - "convergence" + "blocked_priority_fallback", + "concrete_user_gate", + "safe_fallback_work", + "quota_discipline", + "attention_reduction" ], - "headline": "A fuzzy long-running engineering goal needs a shared control plane when multiple worker agents participate.", - "problem": "Specialized engineering work can require multiple agents to split work and converge without losing ownership or state.", + "headline": "A gated P0 lane should not stall a whole long-running goal when safe fallback work exists.", + "problem": "A benchmark lane needed a large local dependency before it could continue, while other no-upload benchmark work remained safe.", "loopx_behavior": [ - "keep durable goal state outside any one chat thread", - "make ownership, quota, and evidence writeback explicit", - "let script-generated worker loops continue only after bounded validation", - "project convergence state and human gates for the operator" + "surface the P0 dependency as a concrete user gate", + "avoid spending on the gated lane", + "select a non-dependent fallback lane", + "record both the blocker and the fallback reason" ], - "user_value": "A contributor-approved public artifact shows how one control plane can coordinate Claude Code, generated scripts, and hardware-agent workers across five long-running engineering cases.", - "evidence_boundary": "Public-safe interactive artifact; no raw chats, screenshots, proprietary design details, private repositories, local paths, task ids, credentials, or unpublished hardware artifacts.", + "user_value": "The operator sees exactly what decision is needed while the agent can keep making bounded progress elsewhere.", + "evidence_boundary": "Synthetic public fixture only; no private screenshots, raw tasks, internal links, local image names, or raw run logs.", "frontend_card": { - "visual_metaphor": "multiple worker lanes converging through one shared control plane", - "primary_metric_hint": "five public hardware-agent cases under one control plane", + "visual_metaphor": "priority lanes with one gated lane and one active fallback lane", + "primary_metric_hint": "attention reduction: user sees one concrete decision instead of reading the whole run log", "badges": [ - "interactive-html", - "multi-agent", - "hardware" + "reproducible", + "user-gate", + "fallback" ], "story_beats": [ - "LoopX owns goal state, quota, todos, claims, evidence, and history", - "Claude Code writes task-specific orchestration scripts under that contract", - "hardware-agent workers perform bounded RTL, simulation, and validation work", - "five public cases show closed tasks, DSE, flagship Fmax optimization, and convergence floors" - ] - } - }, - { - "id": "2026-06-23-agent-to-agent-pr-comments", - "date": "2026-06-23", - "title": "Agent-to-agent PR comment and fix loop", - "status": "public_safe_pattern_case", - "case_page": "docs/showcases/cases/0623-agent-to-agent-pr-comments.md", - "demo_command": null, - "domain": "pull-request-review", - "audience": [ - "operator", - "agent-platform-developer", - "open-source-maintainer" - ], - "pattern_tags": [ - "agent_to_agent_handoff", - "pr_comment_loop", - "review_packet", - "claimed_todo", - "successor_todo" - ], - "headline": "PR review feedback can become an owned agent todo with fix evidence instead of a loose chat reminder.", - "problem": "Multiple agent lanes can see or act on the same PR feedback, but without ownership and handoff state the comment-to-fix loop becomes hard to audit.", - "loopx_behavior": [ - "turn review feedback into a claimed todo", - "route implementation through the owning agent lane", - "record fix and validation evidence in the review packet", - "keep successor work explicit after the comment is handled" - ], - "user_value": "The operator can let agents coordinate around PR comments without losing final review visibility or fix evidence.", - "evidence_boundary": "Public-safe pattern case only; no private screenshots, raw chats, internal review notes, local state, credentials, or unpublished artifacts.", - "appendix_surface": { - "reason": "Pattern is durable, but the original evidence included private screenshots; keep as appendix until a fully public PR-comment packet is curated.", - "public_surface": "appendix_only", - "links": [ - "docs/showcases/cases/0623-agent-to-agent-pr-comments.md" - ] - } - }, - { - "id": "2026-06-23-overnight-project-refactor", - "date": "2026-06-23", - "title": "Overnight project refactor as PR-sized slices", - "status": "public_safe_pattern_case", - "case_page": "docs/showcases/cases/0623-overnight-project-refactor.md", - "demo_command": null, - "domain": "repository-refactor", - "audience": [ - "operator", - "agent-platform-developer", - "technical-lead" - ], - "pattern_tags": [ - "long_unattended_goal", - "pr_sized_slices", - "todo_follow_up", - "supersede", - "validation_writeback" - ], - "headline": "A broad refactor can run overnight while staying split into human-sized review units.", - "problem": "Autonomous refactors become risky when discoveries, stale tasks, cleanup, and behavior changes collapse into one broad diff.", - "loopx_behavior": [ - "keep the current refactor slice explicit", - "convert discoveries into follow-up todos", - "supersede stale tasks when the route changes", - "validate each slice before merge or handoff" - ], - "user_value": "The operator can wake up to reviewable PR-sized refactor slices instead of a single giant autonomous diff.", - "evidence_boundary": "Public-safe pattern case only; no private screenshots, raw chats, internal planning notes, local paths, credentials, raw logs, or unpublished project artifacts.", - "appendix_surface": { - "reason": "Pattern is durable, but keep outside canonical cards until public PR slices and validations are curated into a deeper packet.", - "public_surface": "appendix_only", - "links": [ - "docs/showcases/cases/0623-overnight-project-refactor.md" - ] - } - }, - { - "id": "2026-06-24-pr-issue-auto-fix", - "date": "2026-06-24", - "title": "PR issue automatic fix loop", - "status": "public_safe_pattern_case", - "case_page": "docs/showcases/cases/0624-pr-issue-auto-fix.md", - "demo_command": null, - "domain": "issue-fix-workflow", - "audience": [ - "operator", - "agent-platform-developer", - "open-source-maintainer" - ], - "pattern_tags": [ - "issue_fix_workflow", - "review_feedback", - "repro_smoke", - "command_pack", - "successor_todo" - ], - "headline": "Review feedback should become an ordered repair workflow with repro, fix, validation, and reviewer handoff.", - "problem": "PR comments and issues are often concrete enough to fix, but unsafe to feed into an agent as unstructured prompt text without routing, repro, and validation.", - "loopx_behavior": [ - "classify the issue or review feedback", - "create ordered repair todos", - "keep gated source reads explicit", - "separate repro, implementation, validation, and reviewer handoff" - ], - "user_value": "The operator can turn a PR issue into a controlled fix loop without manually rewriting the review comment as an agent plan.", - "evidence_boundary": "Public-safe pattern case only; no private screenshots, raw gated issue bodies, internal review notes, local paths, raw logs, credentials, or unpublished repository artifacts.", - "appendix_surface": { - "reason": "Pattern is durable, but keep outside canonical cards until a public issue-fix PR packet and focused smoke are curated together.", - "public_surface": "appendix_only", - "links": [ - "docs/showcases/cases/0624-pr-issue-auto-fix.md" + "P0 lane is blocked by a user decision", + "LoopX projects the decision as a user todo", + "agent continues safe fallback work", + "state records blocker, fallback, validation, and spend policy" ] - } - }, - { - "id": "2026-06-27-overnight-pr-batch", - "date": "2026-06-27", - "title": "Overnight PR batch with reviewable control", - "status": "public_evidence_case", - "case_page": "docs/showcases/cases/0627-overnight-pr-batch.md", - "demo_command": null, - "domain": "agent-platform-self-improvement", - "audience": [ - "operator", - "agent-platform-developer", - "technical-lead" - ], - "pattern_tags": [ - "high_throughput_reviewable_work", - "pr_sized_slices", - "self_merge_policy", - "validation_writeback", - "public_boundary" - ], - "headline": "An overnight LoopX run can produce many PR-sized slices while keeping review, validation, and public evidence boundaries visible.", - "problem": "High-throughput autonomous work is only useful if the resulting changes remain reviewable, validated, and safe to publish.", - "loopx_behavior": [ - "keep work broken into PR-sized slices instead of a giant unreviewable diff", - "tie runtime, docs, and focused smoke updates together when a control-plane contract changes", - "limit self-merge to narrow validated changes while preserving broader review gates", - "record public evidence from Git history instead of raw agent logs or private screenshots", - "keep public/private boundary checks in the showcase path" - ], - "user_value": "The operator can wake up to a compact batch of merged public slices, see what changed, and still trust that gates, validation, and evidence boundaries were not bypassed.", - "workload_signal": { - "scope": "public_repository_window", - "window": { - "from": "2026-06-27T01:29:00+08:00", - "to": "2026-06-27T11:29:00+08:00", - "hours": 10 - }, - "public_git": { - "merged_commits": 22, - "files_touched": 60, - "insertions": 6695, - "deletions": 223, - "commit_messages_with_pr_numbers": 10 - }, - "claim_boundary": "public Git history only; contemporaneous private queue notes are not used as public evidence" }, - "evidence_boundary": "Public Git evidence only; no private documents, internal screenshots, raw chats, local active-state bodies, raw logs, credentials, or machine-specific paths.", - "appendix_surface": { - "reason": "Public Git evidence case for an overnight throughput window; keep outside the first canonical frontstage cards until each PR slice has a deeper public evidence packet or reproducible demo.", - "public_surface": "appendix_only", - "links": [ - "docs/showcases/cases/0627-overnight-pr-batch.md" - ] + "interactive_page": "docs/showcases/cases/0617-blocked-p0-safe-rotation.html", + "interactive_page_zh": "docs/showcases/cases/0617-blocked-p0-safe-rotation.html", + "interactive_page_en": "docs/showcases/cases/0617-blocked-p0-safe-rotation.en.html", + "localized_pages": { + "zh": "docs/showcases/cases/0617-blocked-p0-safe-rotation.html", + "en": "docs/showcases/cases/0617-blocked-p0-safe-rotation.en.html" + }, + "showcase_rank": 7, + "showcase_table": { + "proof_point": "A user decision should not block all safe work.", + "loopx_intervention": "concrete user todo, safe fallback, quota control" } }, { @@ -439,6 +554,13 @@ "docs/showcases/creator-ops-fake-data-storyboard.md", "docs/showcases/creator-ops-feedback-boundary-contract.md" ] + }, + "interactive_page": "docs/showcases/cases/0620-creator-operator-case-spec.html", + "interactive_page_zh": "docs/showcases/cases/0620-creator-operator-case-spec.html", + "interactive_page_en": "docs/showcases/cases/0620-creator-operator-case-spec.en.html", + "localized_pages": { + "zh": "docs/showcases/cases/0620-creator-operator-case-spec.html", + "en": "docs/showcases/cases/0620-creator-operator-case-spec.en.html" } } ] diff --git a/examples/export-frontstage-share-bundle.mjs b/examples/export-frontstage-share-bundle.mjs index 2d593aa2..f049829c 100644 --- a/examples/export-frontstage-share-bundle.mjs +++ b/examples/export-frontstage-share-bundle.mjs @@ -80,12 +80,12 @@ async function copyIndexForFrontstage(siteDir) { await writeFile(resolve(frontstageDir, "index.html"), html); } -function validateInteractivePagePath(path) { +function validateShowcaseHtmlPath(path) { if (typeof path !== "string") { - throw new Error(`interactive_page must be a string: ${JSON.stringify(path)}`); + throw new Error(`showcase page must be a string: ${JSON.stringify(path)}`); } if (!path.startsWith("docs/showcases/") || !path.endsWith(".html") || path.includes("..")) { - throw new Error(`interactive_page must stay under docs/showcases/*.html: ${path}`); + throw new Error(`showcase page must stay under docs/showcases/*.html: ${path}`); } return path; } @@ -93,9 +93,15 @@ function validateInteractivePagePath(path) { async function copyInteractiveCasePages(siteDir) { const catalog = JSON.parse(await readFile(resolve(repoRoot, showcaseCatalogPath), "utf8")); const interactivePages = new Set(); + for (const pagePath of ["docs/showcases/index.html", "docs/showcases/index.en.html"]) { + interactivePages.add(validateShowcaseHtmlPath(pagePath)); + } for (const item of catalog.cases ?? []) { if (item.interactive_page) { - interactivePages.add(validateInteractivePagePath(item.interactive_page)); + interactivePages.add(validateShowcaseHtmlPath(item.interactive_page)); + } + for (const pagePath of Object.values(item.localized_pages ?? {})) { + interactivePages.add(validateShowcaseHtmlPath(pagePath)); } } diff --git a/examples/frontstage-share-bundle-smoke.mjs b/examples/frontstage-share-bundle-smoke.mjs index 775dcd0c..8cb20b2e 100644 --- a/examples/frontstage-share-bundle-smoke.mjs +++ b/examples/frontstage-share-bundle-smoke.mjs @@ -119,6 +119,18 @@ const interactivePages = manifest.content_sources?.interactive_case_pages ?? []; if (!interactivePages.includes("docs/showcases/cases/0619-dynamic-workflow-hardware-agent.html")) { throw new Error(`share bundle did not include the hardware-agent interactive page: ${JSON.stringify(interactivePages)}`); } +for (const pagePath of [ + "docs/showcases/index.html", + "docs/showcases/index.en.html", + "docs/showcases/cases/0624-pr-issue-auto-fix.html", + "docs/showcases/cases/0624-pr-issue-auto-fix.en.html", + "docs/showcases/cases/0619-dynamic-workflow-hardware-agent.en.html", +]) { + if (!interactivePages.includes(pagePath)) { + throw new Error(`share bundle did not include expected showcase page ${pagePath}: ${JSON.stringify(interactivePages)}`); + } + assertExists(resolve(siteDir, pagePath)); +} assertExists(resolve(siteDir, "docs/showcases/cases/0619-dynamic-workflow-hardware-agent.html")); const hardwareCaseHtml = await readFile(resolve(siteDir, "docs/showcases/cases/0619-dynamic-workflow-hardware-agent.html"), "utf8"); if (!hardwareCaseHtml.includes("loopx 在芯片开发任务上的实践") || !hardwareCaseHtml.includes("VeeR EH1")) { diff --git a/examples/showcase-animation-prototype.py b/examples/showcase-animation-prototype.py index 5272af53..ff7d3236 100644 --- a/examples/showcase-animation-prototype.py +++ b/examples/showcase-animation-prototype.py @@ -549,6 +549,10 @@ def render(storyboard: dict[str, Any], catalog: dict[str, Any]) -> str: """ +def clean_html(value: str) -> str: + return "\n".join(line.rstrip() for line in value.splitlines()) + "\n" + + def parse_args() -> argparse.Namespace: parser = argparse.ArgumentParser(description=__doc__) parser.add_argument("--storyboard", type=Path, default=DEFAULT_STORYBOARD) @@ -561,7 +565,7 @@ def main() -> int: args = parse_args() storyboard = read_json(args.storyboard) catalog = read_json(args.catalog) - output = render(storyboard, catalog) + output = clean_html(render(storyboard, catalog)) if args.output: args.output.parent.mkdir(parents=True, exist_ok=True) args.output.write_text(output, encoding="utf-8") diff --git a/examples/showcase-animation-source-boundary-smoke.py b/examples/showcase-animation-source-boundary-smoke.py index 2d4236a5..8b11eeae 100644 --- a/examples/showcase-animation-source-boundary-smoke.py +++ b/examples/showcase-animation-source-boundary-smoke.py @@ -71,9 +71,13 @@ def assert_catalog_is_frontstage_ready() -> None: assert case.get("headline"), case assert case.get("evidence_boundary"), case frontend = case.get("frontend_card") - assert isinstance(frontend, dict), case - beats = frontend.get("story_beats") - assert isinstance(beats, list) and len(beats) >= 3, case + appendix = case.get("appendix_surface") + assert isinstance(frontend, dict) or isinstance(appendix, dict), case + if isinstance(frontend, dict): + beats = frontend.get("story_beats") + assert isinstance(beats, list) and len(beats) >= 3, case + else: + assert appendix.get("public_surface") == "appendix_only", case def assert_storyboard_uses_public_catalog() -> None: diff --git a/examples/showcase-catalog-smoke.py b/examples/showcase-catalog-smoke.py index 70ec9d51..3c8cadfe 100644 --- a/examples/showcase-catalog-smoke.py +++ b/examples/showcase-catalog-smoke.py @@ -25,6 +25,20 @@ ("pass", "word="), ) ) +PRIMARY_SHOWCASE_IDS = [ + "2026-06-27-overnight-pr-batch", + "2026-06-24-pr-issue-auto-fix", + "2026-06-23-agent-to-agent-pr-comments", + "2026-06-23-overnight-project-refactor", + "2026-06-19-dynamic-workflow-hardware-agent", + "2026-06-19-loopx-self-iteration", + "2026-06-17-blocked-p0-safe-rotation", +] +FORBIDDEN_SHOWCASE_COPY = ( + "故事" + "节奏", + "Story " + "beats", + "Website Story " + "Beats", +) def read(path: Path) -> str: @@ -48,11 +62,7 @@ def main() -> int: assert "2026-06-19-dynamic-workflow-hardware-agent" in case_ids, case_ids assert "2026-06-19-loopx-self-iteration" in case_ids, case_ids frontstage_ids = [case.get("id") for case in cases if isinstance(case.get("frontend_card"), dict)] - assert frontstage_ids[:3] == [ - "2026-06-17-blocked-p0-safe-rotation", - "2026-06-19-loopx-self-iteration", - "2026-06-19-dynamic-workflow-hardware-agent", - ], frontstage_ids + assert frontstage_ids[:7] == PRIMARY_SHOWCASE_IDS, frontstage_ids assert_public_safe(CATALOG) assert_public_safe(SHOWCASES) @@ -77,6 +87,23 @@ def main() -> int: else: assert appendix.get("reason"), case assert appendix.get("public_surface") == "appendix_only", case + localized_pages = case.get("localized_pages") + assert isinstance(localized_pages, dict), case + for lang in ("zh", "en"): + localized_page = localized_pages.get(lang) + assert isinstance(localized_page, str), case + assert localized_page.startswith("docs/showcases/"), case + assert localized_page.endswith(".html"), case + localized_path = REPO_ROOT / localized_page + assert localized_path.is_file(), case + assert_public_safe(localized_path) + localized_text = read(localized_path) + for phrase in FORBIDDEN_SHOWCASE_COPY: + assert phrase not in localized_text, f"{localized_page}: forbidden copy {phrase!r}" + if case_id != "2026-06-19-dynamic-workflow-hardware-agent" or lang == "en": + assert "Repository evidence" in localized_text or "仓库证据" in localized_text, localized_page + assert "Repository sources" in localized_text or "仓库来源" in localized_text, localized_page + assert str(case.get("evidence_boundary")) in localized_text, localized_page demo_command = case.get("demo_command") if case.get("status") == "reproducible_synthetic_demo": diff --git a/examples/showcase-frontstage-prototype-smoke.py b/examples/showcase-frontstage-prototype-smoke.py index af0a34d4..0ca6d316 100644 --- a/examples/showcase-frontstage-prototype-smoke.py +++ b/examples/showcase-frontstage-prototype-smoke.py @@ -27,6 +27,15 @@ ("pass", "word="), ) ) +PRIMARY_SHOWCASE_IDS = [ + "2026-06-27-overnight-pr-batch", + "2026-06-24-pr-issue-auto-fix", + "2026-06-23-agent-to-agent-pr-comments", + "2026-06-23-overnight-project-refactor", + "2026-06-19-dynamic-workflow-hardware-agent", + "2026-06-19-loopx-self-iteration", + "2026-06-17-blocked-p0-safe-rotation", +] def main() -> int: @@ -48,6 +57,13 @@ def main() -> int: for marker in PRIVATE_MARKERS: assert marker not in html, marker + forbidden_display_terms = ( + "Story " + "beats", + "Website " + "Story " + "Beats", + "故事" + "节奏", + ) + for marker in forbidden_display_terms: + assert marker not in html, marker for required in ( '
', @@ -67,11 +83,7 @@ def main() -> int: ): assert required in html, required - assert [case["id"] for case in frontstage_cases] == [ - "2026-06-17-blocked-p0-safe-rotation", - "2026-06-19-loopx-self-iteration", - "2026-06-19-dynamic-workflow-hardware-agent", - ], frontstage_cases + assert [case["id"] for case in frontstage_cases] == PRIMARY_SHOWCASE_IDS, frontstage_cases assert any(case["id"] == "2026-06-20-creator-operator-case-spec" for case in appendix_cases), appendix_cases statuses = {str(case["status"]) for case in frontstage_cases} diff --git a/examples/showcase-frontstage-prototype.py b/examples/showcase-frontstage-prototype.py index 0ee4619d..143bb7ad 100644 --- a/examples/showcase-frontstage-prototype.py +++ b/examples/showcase-frontstage-prototype.py @@ -27,11 +27,7 @@ def esc(value: object) -> str: def repo_link(repo_relative: str, *, output: Path | None) -> str: - if output is None: - return repo_relative - target = (REPO_ROOT / repo_relative).resolve() - base = output.resolve().parent - return os.path.relpath(target, base).replace(os.sep, "/") + return repo_relative def render_badges(values: list[Any]) -> str: @@ -166,7 +162,7 @@ def render_case(case: dict[str, Any], *, output: Path | None) -> str: if feedback_contract_href else "" ) - story = "".join(f"
  • {esc(beat)}
    • " for beat in story_beats) + evidence_sequence = "".join(f"
  • {esc(beat)}
    • " for beat in story_beats) behavior_items = "".join(f"
  • {esc(item)}
    • " for item in behavior) efficiency_panel = render_efficiency_panel(case) @@ -195,8 +191,8 @@ def render_case(case: dict[str, Any], *, output: Path | None) -> str:
      {behavior_items}
    -

    Story beats

    -
      {story}
    +

    Evidence sequence

    +
      {evidence_sequence}
    {demo} diff --git a/examples/showcase-html-pages.py b/examples/showcase-html-pages.py new file mode 100644 index 00000000..b6431daf --- /dev/null +++ b/examples/showcase-html-pages.py @@ -0,0 +1,1157 @@ +#!/usr/bin/env python3 +"""Generate public-safe showcase HTML pages without rewriting the hardware artifact.""" + +from __future__ import annotations + +import argparse +import html +import json +import os +import re +from pathlib import Path +from typing import Any + + +REPO_ROOT = Path(__file__).resolve().parents[1] +SHOWCASE_DIR = REPO_ROOT / "docs" / "showcases" +CASES_DIR = SHOWCASE_DIR / "cases" +CATALOG = SHOWCASE_DIR / "showcase-catalog.json" +HARDWARE_CASE_ID = "2026-06-19-dynamic-workflow-hardware-agent" +HARDWARE_CANONICAL_PAGE = CASES_DIR / "0619-dynamic-workflow-hardware-agent.html" + +PRIMARY_CASE_ORDER = [ + "2026-06-27-overnight-pr-batch", + "2026-06-24-pr-issue-auto-fix", + "2026-06-23-agent-to-agent-pr-comments", + "2026-06-23-overnight-project-refactor", + HARDWARE_CASE_ID, + "2026-06-19-loopx-self-iteration", + "2026-06-17-blocked-p0-safe-rotation", +] + +SHOWCASE_TABLE = { + "2026-06-27-overnight-pr-batch": { + "proof_point": "High-throughput multi-lane work can remain PR-sized, reviewable, and merge-safe.", + "loopx_intervention": "todo claim, review packet, self-merge boundary, focused smoke, public-boundary scan", + }, + "2026-06-24-pr-issue-auto-fix": { + "proof_point": "Issue and review feedback can enter an executable repair loop.", + "loopx_intervention": "issue-fix workflow, command pack, repro smoke, PR review feedback", + }, + "2026-06-23-agent-to-agent-pr-comments": { + "proof_point": "Multiple agents can coordinate around PR review comments without losing owner review.", + "loopx_intervention": "claimed_by, handoff gate, review packet, comment/fix loop", + }, + "2026-06-23-overnight-project-refactor": { + "proof_point": "Long unattended refactors can split into moderate PR slices instead of one unreviewable diff.", + "loopx_intervention": "loop, todo follow-up, supersede, PR-sized slices", + }, + HARDWARE_CASE_ID: { + "proof_point": "Fuzzy goals, multiple workers, and long unattended runs can still converge.", + "loopx_intervention": "goal state, worker handoff, dynamic workflow", + }, + "2026-06-19-loopx-self-iteration": { + "proof_point": "A high-churn multi-lane project can keep state, boundaries, and evidence coherent.", + "loopx_intervention": "todo, quota, gate, evidence, review packet, frontstage", + }, + "2026-06-17-blocked-p0-safe-rotation": { + "proof_point": "A user decision should not block all safe work.", + "loopx_intervention": "concrete user todo, safe fallback, quota control", + }, +} + +ZH_COPY = { + "2026-06-27-overnight-pr-batch": { + "title": "一晚 30 个高价值 PR", + "headline": "高吞吐、多车道推进仍保持 PR 级可 review、可合并。", + "proof_point": "高吞吐多 lane 工作也可以保持 PR 粒度、可审阅、可合并。", + "loopx_intervention": "todo claim、review packet、自合并边界、focused smoke、public-boundary scan", + "beats": ["长时间窗口被拆成 PR-sized slice。", "每条 lane 写回验证和 review 边界。", "公开 Git 证据替代原始运行日志。"], + }, + "2026-06-24-pr-issue-auto-fix": { + "title": "PR Issue 自动 Fix", + "headline": "Issue / Review comment 可以进入可执行修复闭环。", + "proof_point": "Issue 和 review 反馈可以变成受控、可执行的修复循环。", + "loopx_intervention": "issue-fix workflow、command pack、repro smoke、PR review feedback", + "beats": ["反馈先变成可 claim 的修复目标。", "repro、fix、validation 被拆开记录。", "剩余风险回到 reviewer，而不是被 agent 默认吞掉。"], + }, + "2026-06-23-agent-to-agent-pr-comments": { + "title": "Agent to agent 回复 PR comment 和 PR Fix", + "headline": "多 agent 可围绕 PR review 协同发现、评论、修复。", + "proof_point": "多 agent lane 可以围绕 PR comment 协作，同时保留 owner review。", + "loopx_intervention": "claimed_by、handoff gate、review packet、comment/fix loop", + "beats": ["一个 agent 留下公开安全的 PR 反馈。", "另一条 lane claim 后续修复。", "owner 仍能看到 handoff、证据和剩余 gate。"], + }, + "2026-06-23-overnight-project-refactor": { + "title": "一晚上自主重构项目", + "headline": "长时间无值守重构能拆成适中 PR，而不是一个不可 review 大改。", + "proof_point": "无人值守 refactor 可以拆成 PR-sized slice。", + "loopx_intervention": "loop、todo follow-up、supersede、PR-sized slices", + "beats": ["开放 cleanup 被拆成有序 slice。", "发现项写成 follow-up todo。", "每个 slice 都留下验证和剩余风险。"], + }, + HARDWARE_CASE_ID: { + "title": "外部芯片 agent workflow", + "headline": "模糊目标、多 worker、长时间无值守下仍可收敛。", + "proof_point": "模糊目标、多 worker、长时间无值守下仍可收敛。", + "loopx_intervention": "goal state、worker handoff、dynamic workflow", + "beats": ["LoopX 持有 goal state、quota、todo、claim、evidence 和 history。", "Claude Code 编写任务级编排脚本。", "worker agent 执行有边界的硬件工程工作。"], + }, + "2026-06-19-loopx-self-iteration": { + "title": "LoopX Meta Agent 自迭代", + "headline": "高变更、多车道项目可保持状态、边界和证据。", + "proof_point": "高 churn 多 lane agent repo 可以保持状态、证据和边界一致。", + "loopx_intervention": "todo、quota、gate、evidence、review packet、frontstage", + "beats": ["benchmark、产品、文档和 side-agent 并行推进。", "状态、gate、quota 和 evidence 收敛到同一控制面。", "公开 Git 窗口形成保守效率证据。"], + }, + "2026-06-17-blocked-p0-safe-rotation": { + "title": "P0 block 后推进 P1/P2", + "headline": "用户决策不应阻塞全部安全工作。", + "proof_point": "被阻塞的 P0 决策不应该阻止安全的 P1/P2 工作继续。", + "loopx_intervention": "concrete user todo、safe fallback、quota control", + "beats": ["P0 决策投影成具体 user todo。", "safe fallback lane 继续推进。", "quota 和 evidence 限制自动推进节奏。"], + }, + "2026-06-20-creator-operator-case-spec": { + "title": "创作者-运营者长跑 Agent 案例", + "headline": "研究可以继续，发布决策仍然 gated。", + "proof_point": "创作与运营工作可以共享一个 gate-aware 的长期 agent loop。", + "loopx_intervention": "creator-operator workflow、user gate、feedback capture、material library", + "beats": ["素材整理作为 safe side path 继续。", "发布和品牌判断停在人类 gate。", "反馈沉淀成可复用偏好和材料库。"], + }, +} + +DEFAULT_FRONTEND = { + "2026-06-27-overnight-pr-batch": ("parallel PR lanes converge into a reviewable merge rail", ["PR batch", "review packet", "public boundary"]), + "2026-06-24-pr-issue-auto-fix": ("issue feedback closes through repro, fix, validation, and review handoff", ["issue fix", "repro smoke", "PR review"]), + "2026-06-23-agent-to-agent-pr-comments": ("agent comments pass through explicit ownership instead of loose threads", ["handoff", "PR comment", "claimed_by"]), + "2026-06-23-overnight-project-refactor": ("a large refactor moves as small reviewable packets", ["refactor", "PR slices", "todo evidence"]), +} + +CASE_DETAILS = { + "2026-06-27-overnight-pr-batch": { + "zh": { + "context": [ + "这个案例不是在展示“agent 很忙”，而是在展示高吞吐自动化如何仍然保持可审阅。仓库把证据窗口固定在 2026-06-27 01:29 到 11:29 +08:00，避免把私有队列、聊天截图或操作员笔记当成公开证明。", + "窗口内公开 Git 历史显示 22 个 merged commits、60 个 touched files、6695 行新增和 223 行删除，其中 10 条 commit message 带 PR 编号。页面只声称这些公开 Git 能复现的事实。", + ], + "evidence": [ + ("公开窗口", "2026-06-27 01:29 到 11:29 +08:00，10 小时 Git 证据窗口。"), + ("变更粒度", "22 个 merged commits 覆盖 docs、runtime、status/quota、benchmark contracts、smokes 和 release/runtime guardrails。"), + ("审阅边界", "工作以 PR-sized slices 落地，而不是一个需要 maintainer 盲信的巨大 diff。"), + ("复现方式", "case 文档给出 `git log --since ... --until ...` 和 `git log --numstat` 命令。"), + ], + "mechanism": [ + "LoopX 把每条 lane 绑定到 todo ownership、validation、review policy 和 public evidence。", + "当 control-plane contract 改变时，runtime、docs 和 focused smoke 一起移动。", + "self-merge 只用于窄而验证过的变更；更大或不清楚的工作仍保留 review gate。", + "公开边界扫描把内部截图、本地状态、原始日志和私有计划排除在 showcase 之外。", + ], + "user_outcome": [ + "用户醒来看到的是一组可 review、可追踪、可解释的 public slices，而不是一堆原始 agent 输出。", + "operator 可以继续追问每条 slice 的验证和剩余 gate；证据不依赖 chat memory。", + ], + "source_refs": [ + ("case narrative", "docs/showcases/cases/0627-overnight-pr-batch.md"), + ("catalog workload signal", "docs/showcases/showcase-catalog.json"), + ], + }, + "en": { + "context": [ + "This case is not about an agent being busy. It is about keeping high-throughput autonomous work reviewable. The repository anchors the public evidence window to 2026-06-27 01:29 through 11:29 +08:00 instead of relying on private queues, screenshots, or operator notes.", + "Public Git history in that window shows 22 merged commits, 60 touched files, 6695 insertions, 223 deletions, and 10 commit messages with explicit PR numbers. The page only claims what those public facts support.", + ], + "evidence": [ + ("Public window", "2026-06-27 01:29 to 11:29 +08:00, a 10-hour Git evidence window."), + ("Change shape", "22 merged commits across docs, runtime, status/quota, benchmark contracts, smokes, and release/runtime guardrails."), + ("Review boundary", "Work landed as PR-sized slices instead of one broad diff that maintainers would have to trust blindly."), + ("Reproduction", "The case document gives `git log --since ... --until ...` and `git log --numstat` commands."), + ], + "mechanism": [ + "LoopX ties each lane to todo ownership, validation, review policy, and public evidence.", + "Runtime, docs, and focused smokes move together when a control-plane contract changes.", + "Self-merge remains limited to narrow validated changes; larger or unclear work preserves review gates.", + "Public-boundary scans keep internal screenshots, local state, raw logs, and private planning out of the showcase.", + ], + "user_outcome": [ + "The user wakes up to reviewable, traceable, explainable public slices rather than raw agent output.", + "The operator can inspect validation and remaining gates for each slice without relying on chat memory.", + ], + "source_refs": [ + ("case narrative", "docs/showcases/cases/0627-overnight-pr-batch.md"), + ("catalog workload signal", "docs/showcases/showcase-catalog.json"), + ], + }, + }, + "2026-06-24-pr-issue-auto-fix": { + "zh": { + "context": [ + "这个案例把 PR issue、review comment 或 issue text 转成可执行修复闭环。它不是“读一段评论就改代码”，而是先做 metadata/intake、repro、branch-local patch、validation 和 review packet。", + "仓库已经有 issue-fix capability：模块在 `loopx/capabilities/issue_fix/`，CLI 入口是 `loopx issue-fix ...`，协议文档和 smoke 都在公开仓库内。", + ], + "evidence": [ + ("产品入口", "`docs/capabilities/issue-fix/README.md` 声明 `loopx issue-fix ...`、content-ops bridge、protocol docs 和 smoke。"), + ("工作流协议", "`issue_fix_workflow_contract_v0` 明确 metadata preview、intake classification、workflow plan、todo writeback、caller repo branch、validation、PR review packet 和 gate handling。"), + ("可执行闭环", "`issue_fix_acceptance_loop_v0` 包含 acceptance fixture、repo-branch fixture 和 caller-approved repo branch mode。"), + ("验证面", "`examples/issue-fix-workflow-plan-smoke.py`、`examples/issue-fix-workflow-contract-smoke.py`、`examples/issue-fix-acceptance-loop-smoke.py` 等 smoke 覆盖这个路径。"), + ], + "mechanism": [ + "public metadata 可以进入 packet；raw issue body、comment body、timeline、provider payload 都是 gated source。", + "accepted candidates 写成有序 LoopX todos：repro smoke、code-context route、branch-local patch、validation、review-packet readiness。", + "caller-approved repo 模式只在 `--execute` 时检查本地 repo，并且输出 repo-relative changed files、validation pass/fail 和 PR-readiness。", + "外部 comment、PR creation、merge、publish、destructive git 和 production action 都保留为显式 gate。", + ], + "user_outcome": [ + "维护者给出一个 public issue/PR signal 后，可以得到一个先复现、再修复、再验证、最后交给 review 的闭环。", + "用户不需要把每个 review comment 手动改写成 agent prompt，也不会把原始 issue body 或本地路径泄漏进公开 artifact。", + ], + "source_refs": [ + ("capability README", "docs/capabilities/issue-fix/README.md"), + ("workflow contract", "docs/capabilities/issue-fix/protocols/issue-fix-workflow-contract-v0.md"), + ("acceptance loop", "docs/capabilities/issue-fix/protocols/issue-fix-acceptance-loop-v0.md"), + ("workflow smoke", "examples/issue-fix-workflow-plan-smoke.py"), + ("acceptance smoke", "examples/issue-fix-acceptance-loop-smoke.py"), + ], + }, + "en": { + "context": [ + "This case turns a PR issue, review comment, or issue text into an executable repair loop. It is not simply reading a comment and editing code; the path goes through metadata/intake, repro, branch-local patch, validation, and review packet.", + "The repository already contains the issue-fix capability: `loopx/capabilities/issue_fix/`, the `loopx issue-fix ...` CLI entry, protocol docs, and public smokes.", + ], + "evidence": [ + ("Product entry", "`docs/capabilities/issue-fix/README.md` names `loopx issue-fix ...`, the content-ops bridge, protocol docs, and smokes."), + ("Workflow contract", "`issue_fix_workflow_contract_v0` defines metadata preview, intake classification, workflow plan, todo writeback, caller repo branch, validation, PR review packet, and gate handling."), + ("Executable loop", "`issue_fix_acceptance_loop_v0` includes the acceptance fixture, repo-branch fixture, and caller-approved repo branch mode."), + ("Validation surface", "Smokes cover workflow planning, workflow contract, metadata/intake, and acceptance-loop behavior."), + ], + "mechanism": [ + "Public metadata can enter packets; raw issue bodies, comment bodies, timelines, and provider payloads remain gated sources.", + "Accepted candidates become ordered LoopX todos: repro smoke, code-context route, branch-local patch, validation, and review-packet readiness.", + "Caller-approved repo mode inspects the local repo only under `--execute` and reports repo-relative changed files plus validation pass/fail.", + "External comments, PR creation, merge, publish, destructive git, and production action remain explicit gates.", + ], + "user_outcome": [ + "A maintainer can provide a public issue or PR signal and get a loop that reproduces, fixes, validates, and prepares human review.", + "The user does not need to rewrite every review comment as an agent prompt, and raw issue bodies or local paths do not leak into the public artifact.", + ], + "source_refs": [ + ("capability README", "docs/capabilities/issue-fix/README.md"), + ("workflow contract", "docs/capabilities/issue-fix/protocols/issue-fix-workflow-contract-v0.md"), + ("acceptance loop", "docs/capabilities/issue-fix/protocols/issue-fix-acceptance-loop-v0.md"), + ("workflow smoke", "examples/issue-fix-workflow-plan-smoke.py"), + ("acceptance smoke", "examples/issue-fix-acceptance-loop-smoke.py"), + ], + }, + }, + "2026-06-23-agent-to-agent-pr-comments": { + "zh": { + "context": [ + "这个案例描述的是 PR review feedback 在多 agent 之间流转时如何不丢 ownership。重点不是聊天记录，而是 comment、claim、handoff、fix、validation、review packet 这条链。", + "公开仓库里可以看到相关控制面已经被产品化：`claimed_by` 出现在 todo projection、review packet、event-sourced state 和多 agent/side-agent prompt 合约里。", + ], + "evidence": [ + ("todo ownership", "`event_sourced_state_contract_v0` 把 `todo_claimed` 定义为 canonical event，记录 ownership、lease 或 `claimed_by`。"), + ("review packet", "`loopx/review_packet.py` 会把 open todo 的 `claimed_by` 显示进 handoff/review 文本。"), + ("side-agent contract", "`docs/heartbeat-automation-prompt.md` 规定 side-agent 小变更可带 evidence 自合并，否则创建 claimed-by handoff todo。"), + ("CLI smokes", "`examples/todo-lifecycle-cli-smoke.py`、`examples/todo-cli-smoke.py` 覆盖 claim、handoff successor、side-agent self-merge 和 review handoff 规则。"), + ], + "mechanism": [ + "PR feedback 先成为一个有 owner 的 todo，而不是 chat reminder。", + "被阻塞或跨 lane 的修复通过 handoff gate 和 review packet 传递，不允许另一个 agent 猜测上下文。", + "修复 agent 需要留下 diff、validation 和剩余 gate；后续工作创建 successor todo。", + "owner review 仍在 PR/review surface 上完成，LoopX 只提供状态、证据和 handoff。", + ], + "user_outcome": [ + "用户不必手动 shepherd 每条 PR comment；控制面会显示谁 claim 了反馈、修复证据在哪里、还有什么需要 review。", + "多 agent 协作不会变成“谁看过这条评论”的记忆题。", + ], + "source_refs": [ + ("event-sourced todo claim", "docs/reference/protocols/event-sourced-state-contract-v0.md"), + ("review packet code", "loopx/review_packet.py"), + ("side-agent prompt contract", "docs/heartbeat-automation-prompt.md"), + ("todo lifecycle smoke", "examples/todo-lifecycle-cli-smoke.py"), + ], + }, + "en": { + "context": [ + "This case shows how PR review feedback can move across multiple agents without losing ownership. The important chain is comment, claim, handoff, fix, validation, and review packet rather than the chat transcript.", + "The public repository contains the control-plane pieces: `claimed_by` appears in todo projection, review packets, event-sourced state, and multi-agent or side-agent prompt contracts.", + ], + "evidence": [ + ("Todo ownership", "`event_sourced_state_contract_v0` defines `todo_claimed` as a canonical event for ownership, lease, or `claimed_by`."), + ("Review packet", "`loopx/review_packet.py` renders open todo `claimed_by` values into handoff and review text."), + ("Side-agent contract", "`docs/heartbeat-automation-prompt.md` requires side agents to self-merge only small validated evidence-backed work or create a claimed handoff todo."), + ("CLI smokes", "`examples/todo-lifecycle-cli-smoke.py` and `examples/todo-cli-smoke.py` cover claim, handoff successor, side-agent self-merge, and review handoff rules."), + ], + "mechanism": [ + "PR feedback becomes an owned todo rather than a chat reminder.", + "Blocked or cross-lane fixes move through handoff gates and review packets instead of another agent guessing context.", + "The fixing agent leaves diff, validation, and remaining gates; follow-up work becomes a successor todo.", + "Owner review still happens on the PR or review surface; LoopX supplies state, evidence, and handoff.", + ], + "user_outcome": [ + "The user does not have to shepherd every PR comment manually; the control plane shows who claimed the feedback, where the fix evidence is, and what still needs review.", + "Multi-agent collaboration stops depending on memory of which agent saw which comment.", + ], + "source_refs": [ + ("event-sourced todo claim", "docs/reference/protocols/event-sourced-state-contract-v0.md"), + ("review packet code", "loopx/review_packet.py"), + ("side-agent prompt contract", "docs/heartbeat-automation-prompt.md"), + ("todo lifecycle smoke", "examples/todo-lifecycle-cli-smoke.py"), + ], + }, + }, + "2026-06-23-overnight-project-refactor": { + "zh": { + "context": [ + "这个案例解决的是无人值守重构的风险：长时间 agent 容易把 cleanup、行为改变、发现的新问题和过期计划混成一个大 diff。", + "LoopX 的公开证据不是某个私有夜间截图，而是 todo lifecycle、successor/supersede、validation writeback 和 review-packet 这些已经在仓库里有文档和 smoke 的控制面。", + ], + "evidence": [ + ("successor path", "`docs/lark-kanban-control-plane-adapter.md` 明确 real successor 使用 `todo complete --next-*`，replacement 或 narrower split 使用 `todo supersede --next-agent-todo`。"), + ("side-agent completion", "`docs/heartbeat-automation-prompt.md` 要求非平凡完成创建 successor todo 或写 no-follow-up rationale。"), + ("CLI validation", "`examples/todo-lifecycle-cli-smoke.py` 覆盖 `--next-agent-todo`、`todo supersede`、claim 继承和 handoff successor。"), + ("review shape", "`loopx review-packet` 把当前 open todo、claimed_by 和 handoff 状态打包成 reviewer 可读的 packet。"), + ], + "mechanism": [ + "当前 refactor slice 必须是可 review 的单位，不把整夜发现都塞进一个 PR。", + "发现的新工作写成 follow-up todo；路线变了就 supersede 旧 todo。", + "每个 slice 用 focused validation 或文档/contract smoke 证明，而不是依赖原始 agent trace。", + "大范围或风险不清楚的 slice 不自合并，进入 review handoff。", + ], + "user_outcome": [ + "用户可以让重构跑过夜，但早上看到的是一组有边界的 review 单元和剩余 todo，而不是一个不可 review 的巨型改动。", + "项目可以继续快，但 review 面仍是人能处理的粒度。", + ], + "source_refs": [ + ("kanban control-plane adapter", "docs/lark-kanban-control-plane-adapter.md"), + ("heartbeat prompt contract", "docs/heartbeat-automation-prompt.md"), + ("todo lifecycle smoke", "examples/todo-lifecycle-cli-smoke.py"), + ("case narrative", "docs/showcases/cases/0623-overnight-project-refactor.md"), + ], + }, + "en": { + "context": [ + "This case addresses the risk of unattended refactoring: a long-running agent can mix cleanup, behavior change, discoveries, and stale plans into one broad diff.", + "The public evidence is not a private overnight screenshot. It is the control-plane behavior already documented and smoke-tested in the repository: todo lifecycle, successor/supersede, validation writeback, and review packets.", + ], + "evidence": [ + ("Successor path", "`docs/lark-kanban-control-plane-adapter.md` says real successors use `todo complete --next-*`, while replacements or narrower splits use `todo supersede --next-agent-todo`."), + ("Side-agent completion", "`docs/heartbeat-automation-prompt.md` requires nontrivial completion to create a successor todo or a no-follow-up rationale."), + ("CLI validation", "`examples/todo-lifecycle-cli-smoke.py` covers `--next-agent-todo`, `todo supersede`, claim inheritance, and handoff successors."), + ("Review shape", "`loopx review-packet` packages open todos, claimed_by, and handoff state for reviewer consumption."), + ], + "mechanism": [ + "The current refactor slice must stay reviewable; overnight discoveries do not all land in one PR.", + "New discoveries become follow-up todos; changed routes supersede stale todos.", + "Each slice carries focused validation or doc/contract smoke evidence rather than raw agent traces.", + "Broad or unclear-risk slices route to review handoff instead of self-merge.", + ], + "user_outcome": [ + "The user can let a refactor run overnight and wake up to bounded review units plus remaining todos, not one unreviewable giant change.", + "The project can move quickly while the review surface remains human-sized.", + ], + "source_refs": [ + ("kanban control-plane adapter", "docs/lark-kanban-control-plane-adapter.md"), + ("heartbeat prompt contract", "docs/heartbeat-automation-prompt.md"), + ("todo lifecycle smoke", "examples/todo-lifecycle-cli-smoke.py"), + ("case narrative", "docs/showcases/cases/0623-overnight-project-refactor.md"), + ], + }, + }, + HARDWARE_CASE_ID: { + "zh": { + "context": ["hardware 中文页面是 canonical artifact，生成器不会重写。"], + "evidence": [], + "mechanism": [], + "user_outcome": [], + "source_refs": [], + }, + "en": { + "context": [ + "The Chinese hardware page remains the canonical artifact. This English companion follows its structure without rewriting the original.", + "The case demonstrates a dynamic workflow around fuzzy long-running hardware goals: generated scripts coordinate bounded worker-agent actions while LoopX preserves goal state, quota, todo ownership, validation evidence, and run history outside any one chat thread.", + ], + "evidence": [ + ("Public artifact", "The canonical HTML page includes the approved hardware workflow artifact and five public-safe hardware cases."), + ("Case family", "The companion note names closed validation, timing optimization, design-space exploration, Fmax optimization, and convergence to an engineering floor."), + ("Boundary", "The public artifact excludes raw chats, screenshots, proprietary design details, private repos, local paths, task ids, credentials, and unpublished hardware artifacts."), + ], + "mechanism": [ + "LoopX owns durable state, quota, todos, claims, evidence, and history.", + "Claude Code writes task-level orchestration and generated scripts.", + "Worker agents perform bounded RTL, simulation, and validation work under explicit review boundaries.", + ], + "user_outcome": [ + "A contributor-approved page shows how multiple hardware-agent workers can coordinate under one control plane.", + "Readers can inspect the product pattern without receiving proprietary hardware details or raw execution traces.", + ], + "source_refs": [ + ("canonical HTML", "docs/showcases/cases/0619-dynamic-workflow-hardware-agent.html"), + ("companion note", "docs/showcases/cases/0619-dynamic-workflow-hardware-agent.md"), + ], + }, + }, + "2026-06-19-loopx-self-iteration": { + "zh": { + "context": [ + "这个案例是 public repo self-iteration：LoopX 用来推进 LoopX 自己，不是一个孤立功能 demo。仓库在 benchmark adapters、control-plane correctness、planning lanes、dashboard/frontstage、docs、smokes 和 multi-agent coordination 上同时高频变化。", + "证据窗口固定到 anchor commit `86d6d9d`，避免文档更新改变自己的证据。这里的效率模型是保守的 public Git model，不使用私有聊天、active-state 原文或 benchmark 原始材料。", + ], + "evidence": [ + ("全仓库证据", "截至 anchor commit：801 个 public commits、570 个 touched files、265703 行新增、49895 行删除。"), + ("近期窗口", "2026-06-18 起有 244 个 public commits、216 个 touched files、52898 行新增、20935 行删除。"), + ("0619 当日", "2026-06-19 有 74 个 public commits、118 个 touched files、16087 行新增、1082 行删除。"), + ("效率模型", "把公开仓库能力拆成 9 个 requirement clusters，保守估计 59-92 个 AI-coding-assisted developer-days，对 19.6 天 public window 得出方向性 compression。"), + ], + "mechanism": [ + "registry、prompt contracts 和 registered agents 命名 primary/side identities，不靠聊天记忆。", + "benchmark、productization、documentation、planning 和 side-agent lanes 被拆成 reviewable obligations。", + "quota/status projection 区分 executable work、monitor work、user gates 和 blockers。", + "side-agent scope 留在 prompt/handoff，todo metadata 只保留 `claimed_by`。", + "public docs 和 smokes 把可复用经验沉淀为仓库 artifact。", + ], + "user_outcome": [ + "operator 可以让 primary benchmark lane 和 product/docs/control-plane side work 并行，而不会失去 ownership、gate、validation 和 merge discipline。", + "对潜在用户来说，这是“长期 agent 项目仍然可读”的证据：未来 agent 能从公开表面恢复目标、owner、gate、验证、证据和 follow-up。", + ], + "source_refs": [ + ("case narrative", "docs/showcases/cases/0619-loopx-self-iteration.md"), + ("workload signal", "docs/showcases/showcase-catalog.json"), + ("frontstage fixture", "examples/fixtures/long-horizon-self-iteration-rollout.public.json"), + ("frontstage smoke", "examples/long-horizon-self-iteration-rollout-fixture-smoke.py"), + ], + }, + "en": { + "context": [ + "This is public-repo self-iteration: LoopX was used to improve LoopX itself, not just one isolated demo. Benchmark adapters, control-plane correctness, planning lanes, dashboard/frontstage, docs, smokes, and multi-agent coordination all moved under high churn.", + "The evidence is fixed to anchor commit `86d6d9d` so this documentation update does not change its own evidence window. The efficiency model is a conservative public-Git model; it excludes private chats, active-state bodies, and raw benchmark material.", + ], + "evidence": [ + ("Whole repository", "Through the anchor commit: 801 public commits, 570 touched files, 265703 insertions, and 49895 deletions."), + ("Recent window", "Since 2026-06-18: 244 public commits, 216 touched files, 52898 insertions, and 20935 deletions."), + ("June 19 signal", "On 2026-06-19: 74 public commits, 118 touched files, 16087 insertions, and 1082 deletions."), + ("Efficiency model", "The case maps public repo capabilities to 9 requirement clusters and estimates 59-92 AI-coding-assisted developer-days against a 19.6-day public window."), + ], + "mechanism": [ + "Registry and prompt contracts name primary and side-agent identities instead of relying on chat memory.", + "Benchmark, productization, documentation, planning, and side-agent lanes become reviewable obligations.", + "Quota and status projection distinguish executable work, monitor work, user gates, and blockers.", + "Side-agent scope lives in prompt and handoff; todo metadata keeps the compact `claimed_by` owner.", + "Public docs and smokes turn reusable lessons into repository artifacts.", + ], + "user_outcome": [ + "The operator can let a primary benchmark lane and product/docs/control-plane side work run in parallel without losing ownership, gates, validation, or merge discipline.", + "For a potential user, this is evidence that a long-running agent project can remain legible: future agents can recover goals, owners, gates, validation, evidence, and follow-up from public surfaces.", + ], + "source_refs": [ + ("case narrative", "docs/showcases/cases/0619-loopx-self-iteration.md"), + ("workload signal", "docs/showcases/showcase-catalog.json"), + ("frontstage fixture", "examples/fixtures/long-horizon-self-iteration-rollout.public.json"), + ("frontstage smoke", "examples/long-horizon-self-iteration-rollout-fixture-smoke.py"), + ], + }, + }, + "2026-06-17-blocked-p0-safe-rotation": { + "zh": { + "context": [ + "这个案例展示 P0 被用户决策卡住时，系统不应该继续硬跑，也不应该让整个目标停摆。原场景是 benchmark rotation：一个 lane 需要大型本地 image，其他 no-upload benchmark work 仍然安全。", + "公开仓库没有暴露原始 benchmark task 或本地 image 名，而是用 synthetic smoke 复现控制面行为。", + ], + "evidence": [ + ("synthetic fixture", "`examples/showcase-0617-blocked-p0-safe-rotation-smoke.py` 构造 P0 user gate、被 gate 阻塞的 P0 agent todo 和 P1 no-upload fallback。"), + ("quota contract", "smoke 断言 `should_run=True`、`requires_user_action=True`、`safe_bypass_allowed=True`、`safe_bypass_kind=scoped_user_gate_fallback`。"), + ("selected fallback", "fixture 选择 `terminal_bench_no_upload`，同时保留 `ale_image` gate 的 user-visible blocker。"), + ("rendered evidence", "smoke 检查 markdown 中包含 `scoped_user_gate_fallback` 和 safe no-upload Terminal-Bench rotation。"), + ], + "mechanism": [ + "用户 todo 具体命名 P0 决策，不用“owner gate”这种空话。", + "agent 不在 gated lane 上花 compute；只选择不依赖该决策的 fallback。", + "状态同时记录 blocker 和 fallback reason，方便之后恢复 P0。", + ], + "user_outcome": [ + "用户看到需要自己决定的具体问题，同时项目仍能在安全范围内推进。", + "这减少了注意力负担：不需要每 10 分钟看一次 agent 为什么没动，也不会错过真正需要决策的事项。", + ], + "source_refs": [ + ("case narrative", "docs/showcases/cases/0617-blocked-p0-safe-rotation.md"), + ("synthetic smoke", "examples/showcase-0617-blocked-p0-safe-rotation-smoke.py"), + ], + }, + "en": { + "context": [ + "This case shows what should happen when a P0 route is blocked by a user decision: the system should neither keep forcing that lane nor stop the whole goal. The original shape was a benchmark rotation where one lane needed a large local image while other no-upload benchmark work remained safe.", + "The public repository does not expose raw benchmark tasks or local image names. It reproduces the control-plane behavior with a synthetic smoke.", + ], + "evidence": [ + ("Synthetic fixture", "`examples/showcase-0617-blocked-p0-safe-rotation-smoke.py` creates a P0 user gate, a P0 agent todo blocked by that gate, and a P1 no-upload fallback."), + ("Quota contract", "The smoke asserts `should_run=True`, `requires_user_action=True`, `safe_bypass_allowed=True`, and `safe_bypass_kind=scoped_user_gate_fallback`."), + ("Selected fallback", "The fixture selects `terminal_bench_no_upload` while preserving the `ale_image` gate as the user-visible blocker."), + ("Rendered evidence", "The smoke checks markdown for `scoped_user_gate_fallback` and safe no-upload Terminal-Bench rotation."), + ], + "mechanism": [ + "The user todo names the concrete P0 decision instead of saying only owner gate.", + "The agent does not spend compute on the gated lane; it selects fallback work that does not depend on the decision.", + "State records both the blocker and the fallback reason so P0 can resume later.", + ], + "user_outcome": [ + "The user sees the exact decision they need to make while the project continues safely elsewhere.", + "Attention load drops: the user does not need to watch repeated idle polls and does not miss the real decision.", + ], + "source_refs": [ + ("case narrative", "docs/showcases/cases/0617-blocked-p0-safe-rotation.md"), + ("synthetic smoke", "examples/showcase-0617-blocked-p0-safe-rotation-smoke.py"), + ], + }, + }, + "2026-06-20-creator-operator-case-spec": { + "zh": { + "context": [ + "这是 appendix case：它展示非技术创作者/运营者如何用 LoopX 管一个长期 research + planning loop，但还不是 top-card proof，因为没有真实用户公开证据。", + "公开材料完全使用 synthetic data，重点是产品形态：趋势候选、偏好映射、insight board、draft queue、material library、人类反馈和 controlled replan。", + ], + "evidence": [ + ("storyboard", "`creator-ops-fake-data-storyboard.md` 定义七个面板和完整 fake fixture，不需要 live platform access。"), + ("feedback contract", "`creator-ops-feedback-boundary-contract.md` 把 gate decision、preference hint、todo update、boundary correction、reward signal、product improvement note 分开。"), + ("source status", "contract 要求 topic、insight、draft、material item 都有 source status，public repo 默认 `synthetic_demo`。"), + ("no autopublish", "publishing 是 hard user gate；safe side work 可以继续，但不能把偏好或 reward 当成发布授权。"), + ], + "mechanism": [ + "creative objective 是 durable goal state，不是聊天窗口里的隐形任务。", + "publish/no-publish 是 user gate；research、整理和 source-status 改进可以作为 safe side path。", + "反馈被写成 preference hint、gate decision、todo update 或 boundary correction。", + "下一次 agent run 前，用户能看到 blocked route、safe side path 和 validation expectation。", + ], + "user_outcome": [ + "非技术用户不用读 prompt、trace 或 raw logs，就能知道上次改变了什么、什么在等自己、什么可以继续。", + "这个 case 适合展示产品方向，但页面会明确它是 synthetic spec，不声称真实增长、质量或收入效果。", + ], + "source_refs": [ + ("case narrative", "docs/showcases/cases/0620-creator-operator-case-spec.md"), + ("fake-data storyboard", "docs/showcases/creator-ops-fake-data-storyboard.md"), + ("feedback boundary contract", "docs/showcases/creator-ops-feedback-boundary-contract.md"), + ], + }, + "en": { + "context": [ + "This is an appendix case: it shows how a non-technical creator-operator might use LoopX to manage a long-running research and planning loop, but it is not a top-card proof until real public user evidence exists.", + "The public material uses only synthetic data. The product shape is trend candidates, preference map, insight board, draft queue, material library, human feedback, and controlled replan.", + ], + "evidence": [ + ("Storyboard", "`creator-ops-fake-data-storyboard.md` defines seven panels and a full fake fixture with no live platform access."), + ("Feedback contract", "`creator-ops-feedback-boundary-contract.md` separates gate decision, preference hint, todo update, boundary correction, reward signal, and product improvement note."), + ("Source status", "The contract requires every topic, insight, draft, and material item to carry source status; public repo defaults to `synthetic_demo`."), + ("No autopublish", "Publishing is a hard user gate; safe side work may continue, but preference or reward is not publication approval."), + ], + "mechanism": [ + "The creative objective is durable goal state, not a hidden task inside chat.", + "Publish/no-publish is a user gate; research, organization, and source-status work can continue as safe side paths.", + "Feedback becomes a preference hint, gate decision, todo update, or boundary correction.", + "Before the next agent run, the user can see the blocked route, safe side path, and validation expectation.", + ], + "user_outcome": [ + "A non-technical user can see what changed, what is waiting for them, and what can continue without reading prompts, traces, or raw logs.", + "The case is useful product direction, but the page clearly labels it as a synthetic spec and makes no claim about real growth, quality, or revenue.", + ], + "source_refs": [ + ("case narrative", "docs/showcases/cases/0620-creator-operator-case-spec.md"), + ("fake-data storyboard", "docs/showcases/creator-ops-fake-data-storyboard.md"), + ("feedback boundary contract", "docs/showcases/creator-ops-feedback-boundary-contract.md"), + ], + }, + }, +} + +UI = { + "en": { + "html_lang": "en", + "alternate": "中文", + "index_title": "Showcase & Good Case", + "index_subtitle": "Real LoopX cases showing how long-running agent work stays reviewable, verifiable, and safe to continue.", + "top_cases": "Top showcase cases", + "appendix": "Appendix case", + "proof": "Proof", + "intervention": "LoopX intervention", + "context": "Case context", + "evidence": "Repository evidence", + "behavior": "LoopX behavior", + "outcome": "What the user sees", + "sources": "Repository sources", + "boundary": "Evidence boundary", + "narrative": "Case note", + "catalog": "Catalog", + "home": "Showcases", + "canonical": "Canonical artifact", + "open": "Open", + "demo": "Demo", + "search": "Search showcase cases", + "footer": "Generated from docs/showcases/showcase-catalog.json. Private links, raw chats, local state, and internal media are excluded.", + }, + "zh": { + "html_lang": "zh-CN", + "alternate": "English", + "index_title": "Showcase & Good Case", + "index_subtitle": "真实 LoopX 案例：长程 agent 工作如何保持可审阅、可验证、可继续推进。", + "top_cases": "顶部 Showcase 案例", + "appendix": "附录案例", + "proof": "证明点", + "intervention": "LoopX 介入", + "context": "案例背景", + "evidence": "仓库证据", + "behavior": "LoopX 行为", + "outcome": "用户看到什么", + "sources": "仓库来源", + "boundary": "证据边界", + "narrative": "案例说明", + "catalog": "Catalog", + "home": "Showcases", + "canonical": "Canonical 原页面", + "open": "打开", + "demo": "Demo", + "search": "搜索 showcase 案例", + "footer": "由 docs/showcases/showcase-catalog.json 生成。不包含私有链接、原始聊天、本地状态或内部媒体。", + }, +} + + +def esc(value: object) -> str: + return html.escape(str(value), quote=True) + + +def slug(value: object) -> str: + return re.sub(r"[^a-z0-9]+", "-", str(value).lower()).strip("-") or "case" + + +def repo_path(path: str) -> Path: + return (REPO_ROOT / path).resolve() + + +def rel_href(source: Path, target: Path) -> str: + return os.path.relpath(target, source.parent).replace(os.sep, "/") + + +def first_items(values: Any, limit: int = 6) -> list[str]: + if not isinstance(values, list): + return [] + return [str(value) for value in values[:limit]] + + +def ui(lang: str, key: str) -> str: + return UI[lang][key] + + +def copy_for(case: dict[str, Any], lang: str) -> dict[str, Any]: + if lang == "zh": + return ZH_COPY.get(str(case.get("id")), {}) + return {} + + +def localized(case: dict[str, Any], lang: str, key: str) -> str: + copy = copy_for(case, lang) + value = copy.get(key) + if value is None: + value = case.get(key) + return str(value or "") + + +def table_for(case: dict[str, Any], lang: str) -> dict[str, str]: + copy = copy_for(case, lang) + if copy: + return { + "proof_point": str(copy.get("proof_point") or ""), + "loopx_intervention": str(copy.get("loopx_intervention") or ""), + } + table = case.get("showcase_table") + if isinstance(table, dict): + return { + "proof_point": str(table.get("proof_point") or ""), + "loopx_intervention": str(table.get("loopx_intervention") or ""), + } + return SHOWCASE_TABLE.get(str(case.get("id")), { + "proof_point": str(case.get("headline") or ""), + "loopx_intervention": ", ".join(first_items(case.get("pattern_tags"), 4)), + }) + + +def details_for(case: dict[str, Any], lang: str) -> dict[str, Any]: + details = CASE_DETAILS.get(str(case.get("id")), {}).get(lang) + if isinstance(details, dict): + return details + return { + "context": [str(case.get("problem") or case.get("headline") or "")], + "evidence": [(ui(lang, "proof"), table_for(case, lang)["proof_point"])], + "mechanism": first_items(case.get("loopx_behavior"), 6), + "user_outcome": [str(case.get("user_value") or "")], + "source_refs": [(ui(lang, "narrative"), str(case.get("case_page") or ""))], + } + + +def render_text_stack(items: list[str]) -> str: + cleaned = [item for item in items if item] + return '
    ' + "".join(f"

    {esc(item)}

    " for item in cleaned) + "
    " + + +def render_evidence_items(items: list[tuple[str, str]]) -> str: + cleaned = [(label, text) for label, text in items if label or text] + return '
    ' + "".join( + f'
    {esc(label)}

    {esc(text)}

    ' + for label, text in cleaned + ) + "
    " + + +def render_source_refs(items: list[tuple[str, str]], current: Path) -> str: + links: list[str] = [] + for label, path in items: + if not path: + continue + target = repo_path(path) + links.append( + f'' + f'{esc(label)}{esc(path)}' + ) + return '
    ' + "".join(links) + "
    " + + +def case_html_path(case: dict[str, Any], lang: str) -> Path: + if str(case.get("id")) == HARDWARE_CASE_ID and lang == "zh": + return HARDWARE_CANONICAL_PAGE + case_page = str(case.get("case_page") or "") + if case_page.endswith(".md"): + base = repo_path(case_page[:-3] + ".html") + else: + base = CASES_DIR / f"{slug(case.get('id') or case.get('title'))}.html" + if lang == "en": + return base.with_name(f"{base.stem}.en{base.suffix}") + return base + + +def index_path(lang: str) -> Path: + return SHOWCASE_DIR / ("index.en.html" if lang == "en" else "index.html") + + +def is_hardware_canonical(case: dict[str, Any], lang: str) -> bool: + return str(case.get("id")) == HARDWARE_CASE_ID and lang == "zh" + + +def assert_hardware_canonical() -> None: + text = HARDWARE_CANONICAL_PAGE.read_text(encoding="utf-8") + markers = [ + "LoopX Hardware-Agent Dynamic Workflow", + "__bundler_thumbnail", + "__bundler/manifest", + "__bundler/template", + "loopx 在芯片开发任务上的实践", + "动态脚本编排", + "DUDUCoder", + "Claude Code", + "DUDU", + "CV32E40P", + "VeeR EH1", + "Viterbi", + ] + missing = [marker for marker in markers if marker not in text] + if missing: + raise AssertionError(f"canonical hardware showcase was overwritten or damaged: {missing!r}") + + +def ordered_cases(cases: list[dict[str, Any]]) -> tuple[list[dict[str, Any]], list[dict[str, Any]]]: + by_id = {str(case.get("id")): case for case in cases} + primary = [by_id[case_id] for case_id in PRIMARY_CASE_ORDER if case_id in by_id] + primary_ids = {str(case.get("id")) for case in primary} + appendix = [case for case in cases if str(case.get("id")) not in primary_ids] + return primary, appendix + + +def badges(items: list[str]) -> str: + return "".join(f"{esc(item)}" for item in items) + + +def css() -> str: + return """ + *{box-sizing:border-box;margin:0;padding:0} + html{scroll-behavior:smooth} + body{background:#0b0b0c;color:#f1f2f3;font-family:'Geist',system-ui,-apple-system,BlinkMacSystemFont,sans-serif;-webkit-font-smoothing:antialiased} + a{color:inherit} + ::selection{background:color-mix(in srgb,var(--accent,#6e79d6) 34%,transparent);color:#fff} + .gh{min-height:100vh;position:relative;overflow-x:hidden;--accent:#6e79d6} + .grain{position:fixed;inset:0;pointer-events:none;opacity:.025;mix-blend-mode:overlay;background-image:url("data:image/svg+xml,%3Csvg xmlns='http://www.w3.org/2000/svg' width='180' height='180'%3E%3Cfilter id='n'%3E%3CfeTurbulence type='fractalNoise' baseFrequency='0.85' numOctaves='3' stitchTiles='stitch'/%3E%3C/filter%3E%3Crect width='100%25' height='100%25' filter='url(%23n)'/%3E%3C/svg%3E")} + article{position:relative;max-width:800px;margin:0 auto;padding:76px 28px 130px} + .mlab{font-family:'Geist Mono',ui-monospace,SFMono-Regular,Menlo,Consolas,monospace;font-size:11px;letter-spacing:.06em;color:#62666d;text-transform:uppercase} + h1{font-size:clamp(31px,4.6vw,46px);font-weight:600;line-height:1.12;letter-spacing:-.03em;margin:16px 0 14px;color:#fafafa;text-wrap:balance} + h2{font-size:22px;font-weight:600;letter-spacing:-.02em;color:#f1f2f3} + h3{font-size:21px;font-weight:600;letter-spacing:-.02em;color:#f1f2f3} + p{font-size:15.5px;line-height:1.75;color:#9ea3aa} + strong{color:#e9eaec;font-weight:600} + code{font-family:'Geist Mono',ui-monospace,SFMono-Regular,Menlo,Consolas,monospace;font-size:12.5px;color:#c4c7cc} + .accent{font-size:clamp(17px,2.1vw,21px);font-weight:500;color:var(--accent);letter-spacing:-.01em;margin-bottom:22px} + .nav{display:flex;gap:10px;flex-wrap:wrap;margin:0 0 28px} + .nav a,.case-link{font-family:'Geist Mono',ui-monospace,monospace;font-size:11px;letter-spacing:.04em;text-decoration:none;color:#c4c7cc;border:1px solid rgba(255,255,255,.12);border-radius:6px;padding:7px 10px;background:rgba(255,255,255,.025)} + .nav a:hover,.case-link:hover{border-color:color-mix(in srgb,var(--accent) 55%,transparent);color:#f4f4f5} + .section-head{margin-top:74px;display:flex;align-items:baseline;gap:13px;margin-bottom:20px} + .section-head span{font-family:'Geist Mono',ui-monospace,monospace;font-size:13px;color:var(--accent)} + .panel{border:1px solid rgba(255,255,255,.1);border-radius:10px;background:#0e0e10;overflow:hidden} + .panel-row{display:grid;grid-template-columns:150px 1fr;border-top:1px solid rgba(255,255,255,.07)} + .panel-row:first-child{border-top:0} + .panel-key{padding:18px 20px;border-right:1px solid rgba(255,255,255,.07);font-family:'Geist Mono',ui-monospace,monospace;font-size:10.5px;color:var(--accent);background:color-mix(in srgb,var(--accent) 8%,transparent);letter-spacing:.04em} + .panel-val{padding:18px 20px;display:flex;align-items:center} + .panel-val p{font-size:14px;line-height:1.6;margin:0} + .diagram{border:1px solid rgba(255,255,255,.1);border-radius:12px;padding:30px 24px 22px;background:#0e0e10;margin:22px 0} + .diagram svg{display:block;width:100%;height:auto} + .chips{display:flex;gap:8px;flex-wrap:wrap;margin-top:22px} + .chips span{font-family:'Geist Mono',ui-monospace,monospace;font-size:10.5px;color:#888d95;border:1px solid rgba(255,255,255,.12);padding:4px 8px;border-radius:5px} + .text-stack{display:flex;flex-direction:column;gap:16px} + .evidence-grid{display:grid;grid-template-columns:repeat(2,minmax(0,1fr));gap:14px;margin:20px 0 6px} + .evidence-card{border:1px solid rgba(255,255,255,.1);border-radius:10px;background:#0e0e10;padding:18px 18px 17px;min-height:128px} + .evidence-label{font-family:'Geist Mono',ui-monospace,monospace;font-size:11px;letter-spacing:.06em;text-transform:uppercase;color:var(--accent);margin-bottom:10px} + .source-list{display:flex;flex-direction:column;gap:10px;margin-top:20px} + .source-ref{display:grid;grid-template-columns:150px 1fr;gap:14px;align-items:center;text-decoration:none;border:1px solid rgba(255,255,255,.1);border-radius:8px;background:#0e0e10;padding:12px 14px} + .source-ref:hover{border-color:color-mix(in srgb,var(--accent) 55%,transparent)} + .source-ref span{font-family:'Geist Mono',ui-monospace,monospace;font-size:10.5px;color:var(--accent);text-transform:uppercase;letter-spacing:.04em} + .source-ref code{color:#aeb3ba;word-break:break-word} + .boundary-box{border-left:2px solid var(--accent);background:#0e0e10;padding:18px 20px;margin-top:16px;border-radius:0 8px 8px 0} + .flow{display:flex;flex-direction:column} + .flow li{display:grid;grid-template-columns:28px 1fr;gap:14px;list-style:none} + .flow-num{width:24px;height:24px;border-radius:50%;border:1px solid color-mix(in srgb,var(--accent) 50%,transparent);display:flex;align-items:center;justify-content:center;font-family:'Geist Mono',ui-monospace,monospace;font-size:11.5px;color:var(--accent);background:#0b0b0c} + .flow-body{padding-bottom:18px;border-left:1px solid rgba(255,255,255,.12);padding-left:14px;margin-left:-27px} + .flow-title{font-size:14.5px;color:#eceef0;font-weight:500;margin-bottom:7px} + .metric-grid{display:grid;grid-template-columns:repeat(2,minmax(0,1fr));gap:14px;margin-top:22px} + .metric{border:1px solid rgba(255,255,255,.1);border-radius:10px;padding:17px 19px;background:#0c0c0e} + .metric strong{display:block;font-size:24px;letter-spacing:-.02em;color:#fafafa} + .metric span{display:block;margin-top:5px;font-family:'Geist Mono',ui-monospace,monospace;font-size:10.5px;color:#62666d;text-transform:uppercase} + .search{width:100%;height:42px;margin:22px 0 16px;border:1px solid rgba(255,255,255,.12);border-radius:8px;background:#0e0e10;color:#f1f2f3;padding:0 12px;font:14px 'Geist',system-ui,sans-serif} + .cards{display:flex;flex-direction:column;gap:12px} + .card{display:block;text-decoration:none;border:1px solid rgba(255,255,255,.1);border-radius:10px;background:#0e0e10;padding:18px 20px} + .card:hover{border-color:color-mix(in srgb,var(--accent) 55%,transparent)} + .card .meta{font-family:'Geist Mono',ui-monospace,monospace;font-size:10.5px;color:#62666d;margin-bottom:8px} + .card p{font-size:14px;margin-top:9px} + .hide{display:none} + footer{margin-top:76px;color:#62666d;font-family:'Geist Mono',ui-monospace,monospace;font-size:10.5px;line-height:1.7} + @media(max-width:720px){article{padding:52px 18px 90px}.panel-row{grid-template-columns:1fr}.panel-key{border-right:0;border-bottom:1px solid rgba(255,255,255,.07)}.metric-grid,.evidence-grid{grid-template-columns:1fr}.source-ref{grid-template-columns:1fr;gap:6px}} +""" + + +def html_head(title: str) -> str: + return f""" + + + {esc(title)} + + +""" + + +def nav(current: Path, lang: str, alternate: Path) -> str: + return f""" + +""" + + +def control_diagram(case: dict[str, Any], lang: str) -> str: + table = table_for(case, lang) + left = localized(case, lang, "title")[:22] + center = table["loopx_intervention"].replace("、", ",").split(",")[0].strip()[:24] or "LoopX" + right = str(case.get("status") or "public case").replace("_", " ")[:24] + return f""" +
    + + + + LoopX + + + + + {esc(left)} + {esc(center)} + {esc(right)} + goal / trigger + todo / gate / evidence + public outcome + + + + + + + +
    +""" + + +def metrics(case: dict[str, Any]) -> str: + workload = case.get("workload_signal") + if not isinstance(workload, dict): + return "" + cards: list[tuple[str, str]] = [] + public_git = workload.get("public_git") if isinstance(workload.get("public_git"), dict) else {} + whole = workload.get("whole_repository") if isinstance(workload.get("whole_repository"), dict) else {} + window = workload.get("window") if isinstance(workload.get("window"), dict) else {} + model = workload.get("efficiency_model") if isinstance(workload.get("efficiency_model"), dict) else {} + estimated = model.get("estimated_developer_days") if isinstance(model.get("estimated_developer_days"), dict) else {} + compression = model.get("single_engineer_calendar_compression") if isinstance(model.get("single_engineer_calendar_compression"), dict) else {} + if public_git.get("merged_commits") is not None: + cards.append((str(public_git["merged_commits"]), "merged PR commits")) + if window.get("hours") is not None: + cards.append((f"{window['hours']}h", "public window")) + if whole.get("commit_count") is not None: + cards.append((str(whole["commit_count"]), "public commits")) + if estimated.get("low") and estimated.get("high"): + cards.append((f"{estimated['low']}-{estimated['high']}d", "AI-assisted baseline")) + if compression.get("low") and compression.get("high"): + cards.append((f"{compression['low']}-{compression['high']}x", "calendar compression")) + if not cards: + return "" + return '
    ' + "".join( + f'
    {esc(value)}{esc(label)}
    ' + for value, label in cards[:4] + ) + "
    " + + +def render_case_page(case: dict[str, Any], lang: str, primary: bool) -> str: + output = case_html_path(case, lang) + alternate = case_html_path(case, "en" if lang == "zh" else "zh") + title = localized(case, lang, "title") + headline = localized(case, lang, "headline") + table = table_for(case, lang) + details = details_for(case, lang) + frontend = case.get("frontend_card") if isinstance(case.get("frontend_card"), dict) else {} + tags = first_items(frontend.get("badges"), 5) or first_items(case.get("pattern_tags"), 5) + behavior = [str(item) for item in details.get("mechanism", [])][:8] or first_items(case.get("loopx_behavior"), 6) + narrative = repo_path(str(case.get("case_page") or "")) + demo = case.get("demo_command") + storyboard = case.get("storyboard_path") + feedback = case.get("feedback_contract_path") + appendix = "" if primary else f'{esc(ui(lang, "appendix"))}' + links = [f'{esc(ui(lang, "narrative"))}'] + if isinstance(storyboard, str): + links.append(f'Storyboard') + if isinstance(feedback, str): + links.append(f'Feedback') + demo_block = f'
    {esc(ui(lang, "demo"))}{esc(demo)}
    ' if isinstance(demo, str) and demo else "" + behavior_list = "".join( + f'
  • {index}
    {esc(item)}
    • ' + for index, item in enumerate(behavior, start=1) + ) + context_block = render_text_stack([str(item) for item in details.get("context", [])]) + evidence_block = render_evidence_items([(str(label), str(text)) for label, text in details.get("evidence", [])]) + outcome_block = render_text_stack([str(item) for item in details.get("user_outcome", [])]) + sources_block = render_source_refs([(str(label), str(path)) for label, path in details.get("source_refs", [])], output) + return f""" + +{html_head("LoopX Showcase: " + title)} + +
    +
    +
    + {nav(output, lang, alternate)} +
    {esc(case.get("date") or "")} · {esc(case.get("domain") or "")}
    +

    {esc(title)}

    +
    {esc(headline)}
    +

    {esc(case.get("user_value") or "")}

    +
    {badges(tags)}{appendix}
    + {control_diagram(case, lang)} + +
    01

    {esc(ui(lang, "context"))}

    + {context_block} + +
    02

    {esc(ui(lang, "evidence"))}

    +
    +
    {esc(ui(lang, "proof"))}

    {esc(table["proof_point"])}

    +
    {esc(ui(lang, "intervention"))}

    {esc(table["loopx_intervention"])}

    +
    + {metrics(case)} + {evidence_block} + +
    03

    {esc(ui(lang, "behavior"))}

    +
      {behavior_list}
    + +
    04

    {esc(ui(lang, "outcome"))}

    + {outcome_block} + +
    05

    {esc(ui(lang, "sources"))}

    + {sources_block} +

    {esc(ui(lang, "boundary"))}. {esc(case.get("evidence_boundary") or "")}

    +
    {''.join(links)}
    + {demo_block} +
    {esc(ui(lang, "footer"))}
    +
    +
    + + +""" + + +def index_card(case: dict[str, Any], current: Path, lang: str) -> str: + output = case_html_path(case, lang) + title = localized(case, lang, "title") + headline = localized(case, lang, "headline") + table = table_for(case, lang) + tags = first_items(case.get("pattern_tags"), 4) + search = " ".join([title, headline, table["proof_point"], table["loopx_intervention"], *tags]).lower() + canonical = f'{esc(ui(lang, "canonical"))}' if str(case.get("id")) == HARDWARE_CASE_ID and lang == "zh" else "" + return f""" + +
    {esc(case.get("date") or "")} · {esc(case.get("status") or "")}
    +

    {esc(title)}

    +

    {esc(headline)}

    +

    {esc(ui(lang, "proof"))}: {esc(table["proof_point"])}

    +
    {badges(tags)}{canonical}
    +     +""" + + +def render_index(cases: list[dict[str, Any]], lang: str) -> str: + primary, appendix = ordered_cases(cases) + current = index_path(lang) + alternate = index_path("en" if lang == "zh" else "zh") + primary_cards = "\n".join(index_card(case, current, lang) for case in primary) + appendix_cards = "\n".join(index_card(case, current, lang) for case in appendix) + return f""" + +{html_head("LoopX " + ui(lang, "index_title"))} + +
    +
    +
    + {nav(current, lang, alternate)} +
    LoopX · public-safe case surface
    +

    {esc(ui(lang, "index_title"))}

    +
    {esc(ui(lang, "index_subtitle"))}
    +

    LoopX preserves goals, gates, todos, claims, quota, run history, and evidence outside any single agent session.

    + {control_diagram({"title": ui(lang, "index_title"), "status": "public surface", "domain": "showcase catalog", "showcase_table": {"proof_point": ui(lang, "index_subtitle"), "loopx_intervention": "catalog, pages, evidence boundary"}}, lang)} + +
    01

    {esc(ui(lang, "top_cases"))}

    + +
    {primary_cards}
    + +
    02

    {esc(ui(lang, "appendix"))}

    +
    {appendix_cards}
    +
    {esc(ui(lang, "footer"))}
    +
    +
    + + + +""" + + +def update_catalog(catalog: dict[str, Any]) -> dict[str, Any]: + cases = catalog.get("cases") + if not isinstance(cases, list): + raise ValueError("catalog must contain cases") + by_id = {str(case.get("id")): case for case in cases} + reordered = [by_id.pop(case_id) for case_id in PRIMARY_CASE_ORDER if case_id in by_id] + reordered.extend(by_id.values()) + for rank, case in enumerate(reordered, start=1): + zh_path = case_html_path(case, "zh").relative_to(REPO_ROOT).as_posix() + en_path = case_html_path(case, "en").relative_to(REPO_ROOT).as_posix() + case["interactive_page"] = zh_path + case["interactive_page_zh"] = zh_path + case["interactive_page_en"] = en_path + case["localized_pages"] = {"zh": zh_path, "en": en_path} + case_id = str(case.get("id")) + if case_id in PRIMARY_CASE_ORDER: + case["showcase_rank"] = rank + case["showcase_table"] = SHOWCASE_TABLE[case_id] + if not isinstance(case.get("frontend_card"), dict): + metaphor, default_badges = DEFAULT_FRONTEND.get( + case_id, + ("a LoopX control-plane rail turns ambiguous agent work into reviewable evidence", first_items(case.get("pattern_tags"), 3)), + ) + case["frontend_card"] = { + "visual_metaphor": metaphor, + "primary_metric_hint": SHOWCASE_TABLE[case_id]["proof_point"], + "badges": default_badges, + "story_beats": first_items(case.get("loopx_behavior"), 4), + } + case.pop("appendix_surface", None) + catalog["cases"] = reordered + return catalog + + +def read_catalog() -> dict[str, Any]: + return json.loads(CATALOG.read_text(encoding="utf-8")) + + +def clean(text: str) -> str: + return "\n".join(line.rstrip() for line in text.splitlines()) + "\n" + + +def write(path: Path, text: str) -> None: + path.parent.mkdir(parents=True, exist_ok=True) + path.write_text(clean(text), encoding="utf-8") + + +def generate(write_files: bool) -> list[Path]: + catalog = update_catalog(read_catalog()) + cases = catalog["cases"] + primary, appendix = ordered_cases(cases) + outputs: list[Path] = [] + if write_files: + CATALOG.write_text(json.dumps(catalog, ensure_ascii=False, indent=2) + "\n", encoding="utf-8") + for case in primary + appendix: + for lang in ("zh", "en"): + output = case_html_path(case, lang) + outputs.append(output) + if is_hardware_canonical(case, lang): + assert_hardware_canonical() + continue + if write_files: + write(output, render_case_page(case, lang, primary=case in primary)) + for lang in ("zh", "en"): + output = index_path(lang) + outputs.append(output) + if write_files: + write(output, render_index(cases, lang)) + return outputs + + +def check() -> None: + catalog = update_catalog(read_catalog()) + expected_catalog = json.dumps(catalog, ensure_ascii=False, indent=2) + "\n" + if CATALOG.read_text(encoding="utf-8") != expected_catalog: + raise AssertionError("showcase catalog is not normalized; run examples/showcase-html-pages.py") + cases = catalog["cases"] + primary, appendix = ordered_cases(cases) + for case in primary + appendix: + for lang in ("zh", "en"): + output = case_html_path(case, lang) + if is_hardware_canonical(case, lang): + assert_hardware_canonical() + continue + expected = clean(render_case_page(case, lang, primary=case in primary)) + if output.read_text(encoding="utf-8") != expected: + raise AssertionError(f"{output.relative_to(REPO_ROOT)} is not generated from the catalog") + for lang in ("zh", "en"): + output = index_path(lang) + expected = clean(render_index(cases, lang)) + if output.read_text(encoding="utf-8") != expected: + raise AssertionError(f"{output.relative_to(REPO_ROOT)} is not generated from the catalog") + + +def main() -> int: + parser = argparse.ArgumentParser(description=__doc__) + parser.add_argument("--check", action="store_true", help="Verify catalog and generated HTML files.") + args = parser.parse_args() + if args.check: + check() + print("showcase-html-pages check ok") + return 0 + for output in generate(write_files=True): + print(output.relative_to(REPO_ROOT).as_posix()) + return 0 + + +if __name__ == "__main__": + raise SystemExit(main()) From 7d9262a0af1a963cc2ded1bcad234c84b8c8fa6a Mon Sep 17 00:00:00 2001 From: huangrt01 Date: Sat, 27 Jun 2026 19:37:52 +0800 Subject: [PATCH 2/3] docs: require first-screen owner review --- AGENTS.md | 15 +++++++++++++++ 1 file changed, 15 insertions(+) diff --git a/AGENTS.md b/AGENTS.md index 67d2449d..89a0dea6 100644 --- a/AGENTS.md +++ b/AGENTS.md @@ -53,6 +53,21 @@ submission behavior, permission boundaries, or launch new benchmark jobs. After self-merging, sync local `main`, leave unrelated untracked local artifacts alone, and continue with the next safe project batch. +## First-Screen Review Gate + +Treat the first visible screen of public product surfaces as owner-reviewed +presentation, not as ordinary copy. Before committing, pushing, or self-merging +changes that alter the first viewport, hero block, primary CTA, or opening +navigation of README, hosted frontstage, showcase index pages, product home +pages, or similarly prominent public entry points, show the user a preview +first and wait for approval. + +The preview should be concrete enough to judge the presentation: provide the +local URL and, when the surface is visual HTML, a screenshot or browser view of +the first viewport. Do not move the review gate into a PR comment, todo note, or +final summary after the fact. It must happen before the public first-screen +change is finalized. + ## Public And Private Boundary Do not commit internal department, team, customer, meeting, reporting, From ec6d0b33e750fbf7af3a408d08efb118a2d156dc Mon Sep 17 00:00:00 2001 From: huangrt01 Date: Sat, 27 Jun 2026 19:44:17 +0800 Subject: [PATCH 3/3] docs: add experimental today value path --- README.md | 17 ++ README.zh-CN.md | 15 + apps/dashboard/src/views/frontstage-page.tsx | 63 ++++ .../0617-blocked-p0-safe-rotation.en.html | 18 +- .../cases/0617-blocked-p0-safe-rotation.html | 18 +- ...19-dynamic-workflow-hardware-agent.en.html | 14 +- .../cases/0619-loopx-self-iteration.en.html | 14 +- .../cases/0619-loopx-self-iteration.html | 14 +- .../0620-creator-operator-case-spec.en.html | 18 +- .../0620-creator-operator-case-spec.html | 18 +- .../0623-agent-to-agent-pr-comments.en.html | 18 +- .../0623-agent-to-agent-pr-comments.html | 18 +- .../0623-overnight-project-refactor.en.html | 18 +- .../0623-overnight-project-refactor.html | 18 +- .../cases/0624-pr-issue-auto-fix.en.html | 18 +- .../cases/0624-pr-issue-auto-fix.html | 18 +- .../cases/0627-overnight-pr-batch.en.html | 14 +- .../cases/0627-overnight-pr-batch.html | 14 +- docs/showcases/frontend-surface.md | 3 +- docs/showcases/index.en.html | 60 +++- docs/showcases/index.html | 60 +++- docs/showcases/showcase-catalog.json | 270 +++++++++++++++++- examples/showcase-catalog-smoke.py | 21 ++ examples/showcase-html-pages.py | 158 +++++++--- 24 files changed, 793 insertions(+), 124 deletions(-) diff --git a/README.md b/README.md index 3559b3de..359271e1 100644 --- a/README.md +++ b/README.md @@ -239,6 +239,23 @@ points: For more cases, open the [showcase catalog](docs/showcases/README.md). For a full presenter material, see the experimental notes below. +### Experimental: Today Value Path + +This is not replacing the first screen. It is an experimental entry point for +users who already understand the control-plane idea and want to pick one useful +LoopX capability today: + +| Capability / path | Expected output | User value metric | +| --- | --- | --- | +| PR review/comment -> fix loop | Branch-ready fix packet with repro, smoke result, and remaining review owner. | Fewer dropped review threads; faster path from comment to validated patch. | +| Overnight PR-sized refactor | Reviewable slice list, validation notes, successor todo, and merge boundary. | More merged commits without turning the next morning into a giant diff audit. | +| P0 blocked -> safe fallback | Kernel projection of the exact user gate, safe fallback todo, quota decision, and evidence boundary inside an active goal. | Less idle agent time while preserving human judgment on the blocked path. | + +Start the goal normally with `/loopx `. The PR review and refactor +paths can be requested directly in plain language; the P0 safe-fallback path is +a repository kernel behavior that appears when an active goal has a concrete +blocking user gate and safe P1/P2 work remains. + ## User Mental Model LoopX has more kernel concepts than a user should have to think about every diff --git a/README.zh-CN.md b/README.zh-CN.md index 494f10e4..c8a07793 100644 --- a/README.zh-CN.md +++ b/README.zh-CN.md @@ -128,6 +128,21 @@ loopx bootstrap \ 完整案例目录见 [docs/showcases/README.md](docs/showcases/README.md)。 更完整的演示材料放在文末实验性能力里。 +### Experimental: Today Value Path + +这不会替代首屏。它先作为实验性入口放在案例区下面，给已经理解控制面价值的用户 +快速选一个今天就能用的 LoopX 能力： + +| 能力 / 路径 | 预期产物 | 用户价值指标 | +| --- | --- | --- | +| PR review/comment -> fix loop | 可复核的修复包：repro、smoke 结果、剩余 review owner。 | 更少遗漏 review 线程，更快从 comment 走到 validated patch。 | +| Overnight PR-sized refactor | 可 review 的 slice 列表、验证记录、后续 todo、merge 边界。 | 增加可合并 commit，而不是第二天早上面对一个巨型 diff。 | +| P0 blocked -> safe fallback | 在已有 goal 内由 kernel 投影具体 user gate、安全 fallback todo、quota 决策和证据边界。 | 等人决策时减少 agent 空转，同时保留人类判断。 | + +正常用 `/loopx ` 启动目标即可。PR review 和 refactor 路径可以用自然语言 +直接描述；P0 safe fallback 不是单独命令，而是已有 goal 出现具体阻塞 user gate、 +且还有安全 P1/P2 工作时触发的仓库 kernel 行为。 + ## 它是什么 LoopX 不是另一个 agent runtime，也不是要替代 Codex、Claude Code、 diff --git a/apps/dashboard/src/views/frontstage-page.tsx b/apps/dashboard/src/views/frontstage-page.tsx index 69fb4ec7..bc98da84 100644 --- a/apps/dashboard/src/views/frontstage-page.tsx +++ b/apps/dashboard/src/views/frontstage-page.tsx @@ -839,6 +839,67 @@ const showcaseMotionTones = [ }, ]; +const todayValueWorkflows = [ + { + workflow: "PR review/comment -> fix loop", + output: "Branch-ready fix packet with repro, smoke result, and remaining review owner.", + metric: "Fewer dropped review threads; faster path from comment to validated patch.", + start: "/loopx fix this PR feedback", + }, + { + workflow: "Overnight PR-sized refactor", + output: "Reviewable slice list, validation notes, successor todo, and merge boundary.", + metric: "More merged commits without turning the next morning into a giant diff audit.", + start: "/loopx split this refactor into reviewable slices", + }, + { + workflow: "P0 blocked -> safe fallback", + output: "Kernel projection of the exact user gate, safe fallback todo, quota decision, and evidence boundary.", + metric: "Less idle agent time while preserving human judgment on the blocked path.", + start: "Appears inside an active /loopx goal when a concrete P0 gate blocks one lane and safe P1/P2 work remains.", + }, +]; + +function ExperimentalTodayValuePanel() { + return ( + +
    +
    +
    +

    Pick one capability that earns value today

    +

    + This lower-priority module does not replace the first screen. It gives evaluators three concrete + LoopX capabilities with expected output and user-facing value metrics. +

    +
    + experimental +
    +
    + {todayValueWorkflows.map((item) => ( +
    +
    capability
    +

    {item.workflow}

    +
    +

    + Output: + {item.output} +

    +

    + Value metric: + {item.metric} +

    + + {item.start} + +
    +
    + ))} +
    +
    +     + ); +} + function ShowcaseMotionBoard() { const [activeCaseId, setActiveCaseId] = useState(frontstageShowcases[0]?.id ?? ""); if (!frontstageShowcases.length) { @@ -1994,6 +2055,8 @@ function FrontstageRoute({ {!isDeveloperMode ? : null} + {!isDeveloperMode ? : null} + {!isDeveloperMode ? : null} {isDeveloperMode ? : null} diff --git a/docs/showcases/cases/0617-blocked-p0-safe-rotation.en.html b/docs/showcases/cases/0617-blocked-p0-safe-rotation.en.html index fa1e3e8e..4b2710e2 100644 --- a/docs/showcases/cases/0617-blocked-p0-safe-rotation.en.html +++ b/docs/showcases/cases/0617-blocked-p0-safe-rotation.en.html @@ -56,6 +56,8 @@ .metric{border:1px solid rgba(255,255,255,.1);border-radius:10px;padding:17px 19px;background:#0c0c0e} .metric strong{display:block;font-size:24px;letter-spacing:-.02em;color:#fafafa} .metric span{display:block;margin-top:5px;font-family:'Geist Mono',ui-monospace,monospace;font-size:10.5px;color:#62666d;text-transform:uppercase} + .mini-metrics{display:flex;flex-wrap:wrap;gap:8px;margin-top:13px} + .mini-metrics span{font-family:'Geist Mono',ui-monospace,monospace;font-size:10.5px;color:#aeb3ba;border:1px solid rgba(255,255,255,.1);border-radius:5px;padding:4px 8px;background:#0b0b0c} .search{width:100%;height:42px;margin:22px 0 16px;border:1px solid rgba(255,255,255,.12);border-radius:8px;background:#0e0e10;color:#f1f2f3;padding:0 12px;font:14px 'Geist',system-ui,sans-serif} .cards{display:flex;flex-direction:column;gap:12px} .card{display:block;text-decoration:none;border:1px solid rgba(255,255,255,.1);border-radius:10px;background:#0e0e10;padding:18px 20px} @@ -63,8 +65,16 @@ .card .meta{font-family:'Geist Mono',ui-monospace,monospace;font-size:10.5px;color:#62666d;margin-bottom:8px} .card p{font-size:14px;margin-top:9px} .hide{display:none} + .experiment{margin-top:24px;border:1px solid rgba(255,255,255,.1);border-radius:10px;background:#0e0e10;padding:20px} + .experiment h3{font-size:18px} + .experiment p{font-size:14px;margin-top:8px} + .experiment-grid{display:grid;grid-template-columns:repeat(3,minmax(0,1fr));gap:12px;margin-top:16px} + .experiment-card{border:1px solid rgba(255,255,255,.1);border-radius:8px;background:#0b0b0c;padding:14px} + .experiment-card strong{display:block;font-size:14px;line-height:1.45} + .experiment-card span{display:block;margin-top:9px;font-size:13px;line-height:1.55;color:#9ea3aa} + .experiment-card em{display:block;margin-top:10px;font-style:normal;font-family:'Geist Mono',ui-monospace,monospace;font-size:10.5px;line-height:1.55;color:var(--accent)} footer{margin-top:76px;color:#62666d;font-family:'Geist Mono',ui-monospace,monospace;font-size:10.5px;line-height:1.7} - @media(max-width:720px){article{padding:52px 18px 90px}.panel-row{grid-template-columns:1fr}.panel-key{border-right:0;border-bottom:1px solid rgba(255,255,255,.07)}.metric-grid,.evidence-grid{grid-template-columns:1fr}.source-ref{grid-template-columns:1fr;gap:6px}} + @media(max-width:720px){article{padding:52px 18px 90px}.panel-row{grid-template-columns:1fr}.panel-key{border-right:0;border-bottom:1px solid rgba(255,255,255,.07)}.metric-grid,.evidence-grid,.experiment-grid{grid-template-columns:1fr}.source-ref{grid-template-columns:1fr;gap:6px}} @@ -110,15 +120,15 @@

    Blocked P0 with safe P1/P2 rotation

    01

    Case context

    -

    This case shows what should happen when a P0 route is blocked by a user decision: the system should neither keep forcing that lane nor stop the whole goal. The original shape was a benchmark rotation where one lane needed a large local image while other no-upload benchmark work remained safe.

    The public repository does not expose raw benchmark tasks or local image names. It reproduces the control-plane behavior with a synthetic smoke.

    +

    This case shows what should happen when a P0 route is blocked by a user decision: the system should neither keep forcing that lane nor stop the whole goal. The original shape was a benchmark rotation where one lane needed a large local image while other no-upload benchmark work remained safe.

    The public repository does not expose raw benchmark tasks or local image names. It reproduces the control-plane behavior with a synthetic smoke. The user-facing value is seeing one concrete P0 decision while safe fallback work can continue and the gated lane does not burn automated progress budget.

    02

    Repository evidence

    Proof

    A user decision should not block all safe work.

    LoopX intervention

    concrete user todo, safe fallback, quota control

    - -
    Synthetic fixture

    `examples/showcase-0617-blocked-p0-safe-rotation-smoke.py` creates a P0 user gate, a P0 agent todo blocked by that gate, and a P1 no-upload fallback.

    Quota contract

    The smoke asserts `should_run=True`, `requires_user_action=True`, `safe_bypass_allowed=True`, and `safe_bypass_kind=scoped_user_gate_fallback`.

    Selected fallback

    The fixture selects `terminal_bench_no_upload` while preserving the `ale_image` gate as the user-visible blocker.

    Rendered evidence

    The smoke checks markdown for `scoped_user_gate_fallback` and safe no-upload Terminal-Bench rotation.

    +
    1concrete P0 user decision
    1safe fallback lane
    0gated-lane auto-progress
    0private upload dependency
    +
    Synthetic fixture

    `examples/showcase-0617-blocked-p0-safe-rotation-smoke.py` reproduces a P0 user gate, a P0 agent lane blocked by that gate, and a P1 no-upload fallback.

    Quota contract

    The smoke pins `should_run=True`, `requires_user_action=True`, `safe_bypass_allowed=True`, `safe_bypass_kind=scoped_user_gate_fallback`, and related fallback evidence.

    Selected fallback

    The fixture selects `terminal_bench_no_upload` while preserving the `ale_image` gate as the user-visible blocker.

    Rendered evidence

    The smoke checks markdown for `scoped_user_gate_fallback` and safe no-upload Terminal-Bench rotation.

    03

    LoopX behavior

    • 1
      The user todo names the concrete P0 decision instead of saying only owner gate.
    • 2
      The agent does not spend compute on the gated lane; it selects fallback work that does not depend on the decision.
    • 3
      State records both the blocker and the fallback reason so P0 can resume later.
    diff --git a/docs/showcases/cases/0617-blocked-p0-safe-rotation.html b/docs/showcases/cases/0617-blocked-p0-safe-rotation.html index e5847df3..d664b663 100644 --- a/docs/showcases/cases/0617-blocked-p0-safe-rotation.html +++ b/docs/showcases/cases/0617-blocked-p0-safe-rotation.html @@ -56,6 +56,8 @@ .metric{border:1px solid rgba(255,255,255,.1);border-radius:10px;padding:17px 19px;background:#0c0c0e} .metric strong{display:block;font-size:24px;letter-spacing:-.02em;color:#fafafa} .metric span{display:block;margin-top:5px;font-family:'Geist Mono',ui-monospace,monospace;font-size:10.5px;color:#62666d;text-transform:uppercase} + .mini-metrics{display:flex;flex-wrap:wrap;gap:8px;margin-top:13px} + .mini-metrics span{font-family:'Geist Mono',ui-monospace,monospace;font-size:10.5px;color:#aeb3ba;border:1px solid rgba(255,255,255,.1);border-radius:5px;padding:4px 8px;background:#0b0b0c} .search{width:100%;height:42px;margin:22px 0 16px;border:1px solid rgba(255,255,255,.12);border-radius:8px;background:#0e0e10;color:#f1f2f3;padding:0 12px;font:14px 'Geist',system-ui,sans-serif} .cards{display:flex;flex-direction:column;gap:12px} .card{display:block;text-decoration:none;border:1px solid rgba(255,255,255,.1);border-radius:10px;background:#0e0e10;padding:18px 20px} @@ -63,8 +65,16 @@ .card .meta{font-family:'Geist Mono',ui-monospace,monospace;font-size:10.5px;color:#62666d;margin-bottom:8px} .card p{font-size:14px;margin-top:9px} .hide{display:none} + .experiment{margin-top:24px;border:1px solid rgba(255,255,255,.1);border-radius:10px;background:#0e0e10;padding:20px} + .experiment h3{font-size:18px} + .experiment p{font-size:14px;margin-top:8px} + .experiment-grid{display:grid;grid-template-columns:repeat(3,minmax(0,1fr));gap:12px;margin-top:16px} + .experiment-card{border:1px solid rgba(255,255,255,.1);border-radius:8px;background:#0b0b0c;padding:14px} + .experiment-card strong{display:block;font-size:14px;line-height:1.45} + .experiment-card span{display:block;margin-top:9px;font-size:13px;line-height:1.55;color:#9ea3aa} + .experiment-card em{display:block;margin-top:10px;font-style:normal;font-family:'Geist Mono',ui-monospace,monospace;font-size:10.5px;line-height:1.55;color:var(--accent)} footer{margin-top:76px;color:#62666d;font-family:'Geist Mono',ui-monospace,monospace;font-size:10.5px;line-height:1.7} - @media(max-width:720px){article{padding:52px 18px 90px}.panel-row{grid-template-columns:1fr}.panel-key{border-right:0;border-bottom:1px solid rgba(255,255,255,.07)}.metric-grid,.evidence-grid{grid-template-columns:1fr}.source-ref{grid-template-columns:1fr;gap:6px}} + @media(max-width:720px){article{padding:52px 18px 90px}.panel-row{grid-template-columns:1fr}.panel-key{border-right:0;border-bottom:1px solid rgba(255,255,255,.07)}.metric-grid,.evidence-grid,.experiment-grid{grid-template-columns:1fr}.source-ref{grid-template-columns:1fr;gap:6px}} @@ -110,15 +120,15 @@

    P0 block 后推进 P1/P2

    01

    案例背景

    -

    这个案例展示 P0 被用户决策卡住时，系统不应该继续硬跑，也不应该让整个目标停摆。原场景是 benchmark rotation：一个 lane 需要大型本地 image，其他 no-upload benchmark work 仍然安全。

    公开仓库没有暴露原始 benchmark task 或本地 image 名，而是用 synthetic smoke 复现控制面行为。

    +

    这个案例展示 P0 被用户决策卡住时，系统不应该继续硬跑，也不应该让整个目标停摆。原场景是 benchmark rotation：一个 lane 需要大型本地 image，其他 no-upload benchmark work 仍然安全。

    公开仓库没有暴露原始 benchmark task 或本地 image 名，而是用 synthetic smoke 复现控制面行为。用户价值是明确看到一个需要决策的 P0，同时安全 fallback 可以继续，且 gated lane 不消耗额外自动推进预算。

    02

    仓库证据

    证明点

    被阻塞的 P0 决策不应该阻止安全的 P1/P2 工作继续。

    LoopX 介入

    concrete user todo、safe fallback、quota control

    - -
    synthetic fixture

    `examples/showcase-0617-blocked-p0-safe-rotation-smoke.py` 构造 P0 user gate、被 gate 阻塞的 P0 agent todo 和 P1 no-upload fallback。

    quota contract

    smoke 断言 `should_run=True`、`requires_user_action=True`、`safe_bypass_allowed=True`、`safe_bypass_kind=scoped_user_gate_fallback`。

    selected fallback

    fixture 选择 `terminal_bench_no_upload`，同时保留 `ale_image` gate 的 user-visible blocker。

    rendered evidence

    smoke 检查 markdown 中包含 `scoped_user_gate_fallback` 和 safe no-upload Terminal-Bench rotation。

    +
    1具体 P0 用户决策
    1safe fallback lane
    0gated lane 自动推进
    0private upload 依赖
    +
    synthetic fixture

    `examples/showcase-0617-blocked-p0-safe-rotation-smoke.py` 复现 P0 user gate、被 gate 阻塞的 P0 agent lane 和 P1 no-upload fallback。

    quota contract

    smoke 固定 `should_run=True`、`requires_user_action=True`、`safe_bypass_allowed=True`、`safe_bypass_kind=scoped_user_gate_fallback` 等关键 contract。

    selected fallback

    fixture 选择 `terminal_bench_no_upload`，同时保留 `ale_image` gate 的 user-visible blocker。

    rendered evidence

    smoke 检查 markdown 中包含 `scoped_user_gate_fallback` 和 safe no-upload Terminal-Bench rotation。

    03

    LoopX 行为

    • 1
      用户 todo 具体命名 P0 决策，不用“owner gate”这种空话。
    • 2
      agent 不在 gated lane 上花 compute；只选择不依赖该决策的 fallback。
    • 3
      状态同时记录 blocker 和 fallback reason，方便之后恢复 P0。
    diff --git a/docs/showcases/cases/0619-dynamic-workflow-hardware-agent.en.html b/docs/showcases/cases/0619-dynamic-workflow-hardware-agent.en.html index 322e3d35..b382ed48 100644 --- a/docs/showcases/cases/0619-dynamic-workflow-hardware-agent.en.html +++ b/docs/showcases/cases/0619-dynamic-workflow-hardware-agent.en.html @@ -56,6 +56,8 @@ .metric{border:1px solid rgba(255,255,255,.1);border-radius:10px;padding:17px 19px;background:#0c0c0e} .metric strong{display:block;font-size:24px;letter-spacing:-.02em;color:#fafafa} .metric span{display:block;margin-top:5px;font-family:'Geist Mono',ui-monospace,monospace;font-size:10.5px;color:#62666d;text-transform:uppercase} + .mini-metrics{display:flex;flex-wrap:wrap;gap:8px;margin-top:13px} + .mini-metrics span{font-family:'Geist Mono',ui-monospace,monospace;font-size:10.5px;color:#aeb3ba;border:1px solid rgba(255,255,255,.1);border-radius:5px;padding:4px 8px;background:#0b0b0c} .search{width:100%;height:42px;margin:22px 0 16px;border:1px solid rgba(255,255,255,.12);border-radius:8px;background:#0e0e10;color:#f1f2f3;padding:0 12px;font:14px 'Geist',system-ui,sans-serif} .cards{display:flex;flex-direction:column;gap:12px} .card{display:block;text-decoration:none;border:1px solid rgba(255,255,255,.1);border-radius:10px;background:#0e0e10;padding:18px 20px} @@ -63,8 +65,16 @@ .card .meta{font-family:'Geist Mono',ui-monospace,monospace;font-size:10.5px;color:#62666d;margin-bottom:8px} .card p{font-size:14px;margin-top:9px} .hide{display:none} + .experiment{margin-top:24px;border:1px solid rgba(255,255,255,.1);border-radius:10px;background:#0e0e10;padding:20px} + .experiment h3{font-size:18px} + .experiment p{font-size:14px;margin-top:8px} + .experiment-grid{display:grid;grid-template-columns:repeat(3,minmax(0,1fr));gap:12px;margin-top:16px} + .experiment-card{border:1px solid rgba(255,255,255,.1);border-radius:8px;background:#0b0b0c;padding:14px} + .experiment-card strong{display:block;font-size:14px;line-height:1.45} + .experiment-card span{display:block;margin-top:9px;font-size:13px;line-height:1.55;color:#9ea3aa} + .experiment-card em{display:block;margin-top:10px;font-style:normal;font-family:'Geist Mono',ui-monospace,monospace;font-size:10.5px;line-height:1.55;color:var(--accent)} footer{margin-top:76px;color:#62666d;font-family:'Geist Mono',ui-monospace,monospace;font-size:10.5px;line-height:1.7} - @media(max-width:720px){article{padding:52px 18px 90px}.panel-row{grid-template-columns:1fr}.panel-key{border-right:0;border-bottom:1px solid rgba(255,255,255,.07)}.metric-grid,.evidence-grid{grid-template-columns:1fr}.source-ref{grid-template-columns:1fr;gap:6px}} + @media(max-width:720px){article{padding:52px 18px 90px}.panel-row{grid-template-columns:1fr}.panel-key{border-right:0;border-bottom:1px solid rgba(255,255,255,.07)}.metric-grid,.evidence-grid,.experiment-grid{grid-template-columns:1fr}.source-ref{grid-template-columns:1fr;gap:6px}} @@ -117,7 +127,7 @@

    Dynamic workflow for hardware-agent development

    Proof

    Fuzzy goals, multiple workers, and long unattended runs can still converge.

    LoopX intervention

    goal state, worker handoff, dynamic workflow

    - +
    5public-safe hardware workflows
    3LoopX/orchestrator/worker role split
    1canonical artifact preserved
    0proprietary design details exposed
    Public artifact

    The canonical HTML page includes the approved hardware workflow artifact and five public-safe hardware cases.

    Case family

    The companion note names closed validation, timing optimization, design-space exploration, Fmax optimization, and convergence to an engineering floor.

    Boundary

    The public artifact excludes raw chats, screenshots, proprietary design details, private repos, local paths, task ids, credentials, and unpublished hardware artifacts.

    03

    LoopX behavior

    diff --git a/docs/showcases/cases/0619-loopx-self-iteration.en.html b/docs/showcases/cases/0619-loopx-self-iteration.en.html index b9e2b2d9..f6877c9c 100644 --- a/docs/showcases/cases/0619-loopx-self-iteration.en.html +++ b/docs/showcases/cases/0619-loopx-self-iteration.en.html @@ -56,6 +56,8 @@ .metric{border:1px solid rgba(255,255,255,.1);border-radius:10px;padding:17px 19px;background:#0c0c0e} .metric strong{display:block;font-size:24px;letter-spacing:-.02em;color:#fafafa} .metric span{display:block;margin-top:5px;font-family:'Geist Mono',ui-monospace,monospace;font-size:10.5px;color:#62666d;text-transform:uppercase} + .mini-metrics{display:flex;flex-wrap:wrap;gap:8px;margin-top:13px} + .mini-metrics span{font-family:'Geist Mono',ui-monospace,monospace;font-size:10.5px;color:#aeb3ba;border:1px solid rgba(255,255,255,.1);border-radius:5px;padding:4px 8px;background:#0b0b0c} .search{width:100%;height:42px;margin:22px 0 16px;border:1px solid rgba(255,255,255,.12);border-radius:8px;background:#0e0e10;color:#f1f2f3;padding:0 12px;font:14px 'Geist',system-ui,sans-serif} .cards{display:flex;flex-direction:column;gap:12px} .card{display:block;text-decoration:none;border:1px solid rgba(255,255,255,.1);border-radius:10px;background:#0e0e10;padding:18px 20px} @@ -63,8 +65,16 @@ .card .meta{font-family:'Geist Mono',ui-monospace,monospace;font-size:10.5px;color:#62666d;margin-bottom:8px} .card p{font-size:14px;margin-top:9px} .hide{display:none} + .experiment{margin-top:24px;border:1px solid rgba(255,255,255,.1);border-radius:10px;background:#0e0e10;padding:20px} + .experiment h3{font-size:18px} + .experiment p{font-size:14px;margin-top:8px} + .experiment-grid{display:grid;grid-template-columns:repeat(3,minmax(0,1fr));gap:12px;margin-top:16px} + .experiment-card{border:1px solid rgba(255,255,255,.1);border-radius:8px;background:#0b0b0c;padding:14px} + .experiment-card strong{display:block;font-size:14px;line-height:1.45} + .experiment-card span{display:block;margin-top:9px;font-size:13px;line-height:1.55;color:#9ea3aa} + .experiment-card em{display:block;margin-top:10px;font-style:normal;font-family:'Geist Mono',ui-monospace,monospace;font-size:10.5px;line-height:1.55;color:var(--accent)} footer{margin-top:76px;color:#62666d;font-family:'Geist Mono',ui-monospace,monospace;font-size:10.5px;line-height:1.7} - @media(max-width:720px){article{padding:52px 18px 90px}.panel-row{grid-template-columns:1fr}.panel-key{border-right:0;border-bottom:1px solid rgba(255,255,255,.07)}.metric-grid,.evidence-grid{grid-template-columns:1fr}.source-ref{grid-template-columns:1fr;gap:6px}} + @media(max-width:720px){article{padding:52px 18px 90px}.panel-row{grid-template-columns:1fr}.panel-key{border-right:0;border-bottom:1px solid rgba(255,255,255,.07)}.metric-grid,.evidence-grid,.experiment-grid{grid-template-columns:1fr}.source-ref{grid-template-columns:1fr;gap:6px}} @@ -117,7 +127,7 @@

    LoopX self-iteration loop

    Proof

    A high-churn multi-lane project can keep state, boundaries, and evidence coherent.

    LoopX intervention

    todo, quota, gate, evidence, review packet, frontstage

    -
    801public commits
    59-92dAI-assisted baseline
    3.0-4.7xcalendar compression
    +
    801public commits
    244commits since Jun 18
    59-92dAI-assisted baseline
    3.0-4.7xcalendar compression
    Whole repository

    Through the anchor commit: 801 public commits, 570 touched files, 265703 insertions, and 49895 deletions.

    Recent window

    Since 2026-06-18: 244 public commits, 216 touched files, 52898 insertions, and 20935 deletions.

    June 19 signal

    On 2026-06-19: 74 public commits, 118 touched files, 16087 insertions, and 1082 deletions.

    Efficiency model

    The case maps public repo capabilities to 9 requirement clusters and estimates 59-92 AI-coding-assisted developer-days against a 19.6-day public window.

    03

    LoopX behavior

    diff --git a/docs/showcases/cases/0619-loopx-self-iteration.html b/docs/showcases/cases/0619-loopx-self-iteration.html index 7cdf5901..447696ce 100644 --- a/docs/showcases/cases/0619-loopx-self-iteration.html +++ b/docs/showcases/cases/0619-loopx-self-iteration.html @@ -56,6 +56,8 @@ .metric{border:1px solid rgba(255,255,255,.1);border-radius:10px;padding:17px 19px;background:#0c0c0e} .metric strong{display:block;font-size:24px;letter-spacing:-.02em;color:#fafafa} .metric span{display:block;margin-top:5px;font-family:'Geist Mono',ui-monospace,monospace;font-size:10.5px;color:#62666d;text-transform:uppercase} + .mini-metrics{display:flex;flex-wrap:wrap;gap:8px;margin-top:13px} + .mini-metrics span{font-family:'Geist Mono',ui-monospace,monospace;font-size:10.5px;color:#aeb3ba;border:1px solid rgba(255,255,255,.1);border-radius:5px;padding:4px 8px;background:#0b0b0c} .search{width:100%;height:42px;margin:22px 0 16px;border:1px solid rgba(255,255,255,.12);border-radius:8px;background:#0e0e10;color:#f1f2f3;padding:0 12px;font:14px 'Geist',system-ui,sans-serif} .cards{display:flex;flex-direction:column;gap:12px} .card{display:block;text-decoration:none;border:1px solid rgba(255,255,255,.1);border-radius:10px;background:#0e0e10;padding:18px 20px} @@ -63,8 +65,16 @@ .card .meta{font-family:'Geist Mono',ui-monospace,monospace;font-size:10.5px;color:#62666d;margin-bottom:8px} .card p{font-size:14px;margin-top:9px} .hide{display:none} + .experiment{margin-top:24px;border:1px solid rgba(255,255,255,.1);border-radius:10px;background:#0e0e10;padding:20px} + .experiment h3{font-size:18px} + .experiment p{font-size:14px;margin-top:8px} + .experiment-grid{display:grid;grid-template-columns:repeat(3,minmax(0,1fr));gap:12px;margin-top:16px} + .experiment-card{border:1px solid rgba(255,255,255,.1);border-radius:8px;background:#0b0b0c;padding:14px} + .experiment-card strong{display:block;font-size:14px;line-height:1.45} + .experiment-card span{display:block;margin-top:9px;font-size:13px;line-height:1.55;color:#9ea3aa} + .experiment-card em{display:block;margin-top:10px;font-style:normal;font-family:'Geist Mono',ui-monospace,monospace;font-size:10.5px;line-height:1.55;color:var(--accent)} footer{margin-top:76px;color:#62666d;font-family:'Geist Mono',ui-monospace,monospace;font-size:10.5px;line-height:1.7} - @media(max-width:720px){article{padding:52px 18px 90px}.panel-row{grid-template-columns:1fr}.panel-key{border-right:0;border-bottom:1px solid rgba(255,255,255,.07)}.metric-grid,.evidence-grid{grid-template-columns:1fr}.source-ref{grid-template-columns:1fr;gap:6px}} + @media(max-width:720px){article{padding:52px 18px 90px}.panel-row{grid-template-columns:1fr}.panel-key{border-right:0;border-bottom:1px solid rgba(255,255,255,.07)}.metric-grid,.evidence-grid,.experiment-grid{grid-template-columns:1fr}.source-ref{grid-template-columns:1fr;gap:6px}} @@ -117,7 +127,7 @@

    LoopX Meta Agent 自迭代

    证明点

    高 churn 多 lane agent repo 可以保持状态、证据和边界一致。

    LoopX 介入

    todo、quota、gate、evidence、review packet、frontstage

    -
    801public commits
    59-92dAI-assisted baseline
    3.0-4.7xcalendar compression
    +
    801public commits
    24406-18 后 commits
    59-92dAI-assisted baseline
    3.0-4.7xcalendar compression
    全仓库证据

    截至 anchor commit：801 个 public commits、570 个 touched files、265703 行新增、49895 行删除。

    近期窗口

    2026-06-18 起有 244 个 public commits、216 个 touched files、52898 行新增、20935 行删除。

    0619 当日

    2026-06-19 有 74 个 public commits、118 个 touched files、16087 行新增、1082 行删除。

    效率模型

    把公开仓库能力拆成 9 个 requirement clusters，保守估计 59-92 个 AI-coding-assisted developer-days，对 19.6 天 public window 得出方向性 compression。

    03

    LoopX 行为

    diff --git a/docs/showcases/cases/0620-creator-operator-case-spec.en.html b/docs/showcases/cases/0620-creator-operator-case-spec.en.html index 962d7586..40d97286 100644 --- a/docs/showcases/cases/0620-creator-operator-case-spec.en.html +++ b/docs/showcases/cases/0620-creator-operator-case-spec.en.html @@ -56,6 +56,8 @@ .metric{border:1px solid rgba(255,255,255,.1);border-radius:10px;padding:17px 19px;background:#0c0c0e} .metric strong{display:block;font-size:24px;letter-spacing:-.02em;color:#fafafa} .metric span{display:block;margin-top:5px;font-family:'Geist Mono',ui-monospace,monospace;font-size:10.5px;color:#62666d;text-transform:uppercase} + .mini-metrics{display:flex;flex-wrap:wrap;gap:8px;margin-top:13px} + .mini-metrics span{font-family:'Geist Mono',ui-monospace,monospace;font-size:10.5px;color:#aeb3ba;border:1px solid rgba(255,255,255,.1);border-radius:5px;padding:4px 8px;background:#0b0b0c} .search{width:100%;height:42px;margin:22px 0 16px;border:1px solid rgba(255,255,255,.12);border-radius:8px;background:#0e0e10;color:#f1f2f3;padding:0 12px;font:14px 'Geist',system-ui,sans-serif} .cards{display:flex;flex-direction:column;gap:12px} .card{display:block;text-decoration:none;border:1px solid rgba(255,255,255,.1);border-radius:10px;background:#0e0e10;padding:18px 20px} @@ -63,8 +65,16 @@ .card .meta{font-family:'Geist Mono',ui-monospace,monospace;font-size:10.5px;color:#62666d;margin-bottom:8px} .card p{font-size:14px;margin-top:9px} .hide{display:none} + .experiment{margin-top:24px;border:1px solid rgba(255,255,255,.1);border-radius:10px;background:#0e0e10;padding:20px} + .experiment h3{font-size:18px} + .experiment p{font-size:14px;margin-top:8px} + .experiment-grid{display:grid;grid-template-columns:repeat(3,minmax(0,1fr));gap:12px;margin-top:16px} + .experiment-card{border:1px solid rgba(255,255,255,.1);border-radius:8px;background:#0b0b0c;padding:14px} + .experiment-card strong{display:block;font-size:14px;line-height:1.45} + .experiment-card span{display:block;margin-top:9px;font-size:13px;line-height:1.55;color:#9ea3aa} + .experiment-card em{display:block;margin-top:10px;font-style:normal;font-family:'Geist Mono',ui-monospace,monospace;font-size:10.5px;line-height:1.55;color:var(--accent)} footer{margin-top:76px;color:#62666d;font-family:'Geist Mono',ui-monospace,monospace;font-size:10.5px;line-height:1.7} - @media(max-width:720px){article{padding:52px 18px 90px}.panel-row{grid-template-columns:1fr}.panel-key{border-right:0;border-bottom:1px solid rgba(255,255,255,.07)}.metric-grid,.evidence-grid{grid-template-columns:1fr}.source-ref{grid-template-columns:1fr;gap:6px}} + @media(max-width:720px){article{padding:52px 18px 90px}.panel-row{grid-template-columns:1fr}.panel-key{border-right:0;border-bottom:1px solid rgba(255,255,255,.07)}.metric-grid,.evidence-grid,.experiment-grid{grid-template-columns:1fr}.source-ref{grid-template-columns:1fr;gap:6px}} @@ -110,15 +120,15 @@

    Creator-operator long-running agent case

    01

    Case context

    -

    This is an appendix case: it shows how a non-technical creator-operator might use LoopX to manage a long-running research and planning loop, but it is not a top-card proof until real public user evidence exists.

    The public material uses only synthetic data. The product shape is trend candidates, preference map, insight board, draft queue, material library, human feedback, and controlled replan.

    +

    This is an appendix case: it shows how a non-technical creator-operator might use LoopX to manage a long-running research and planning loop, but it is not a top-card proof until real public user evidence exists.

    The public material uses only synthetic data. The point is not proving growth; it is proving the product boundary: research and material organization can continue, while publishing, brand judgment, and external action remain human gates.

    02

    Repository evidence

    Proof

    A creator-operator needs a long-running agent loop that keeps research moving while publishing decisions stay gated.

    LoopX intervention

    creator_operator_workflow, gate_aware_continuation, feedback_capture, safe_side_path

    - -
    Storyboard

    `creator-ops-fake-data-storyboard.md` defines seven panels and a full fake fixture with no live platform access.

    Feedback contract

    `creator-ops-feedback-boundary-contract.md` separates gate decision, preference hint, todo update, boundary correction, reward signal, and product improvement note.

    Source status

    The contract requires every topic, insight, draft, and material item to carry source status; public repo defaults to `synthetic_demo`.

    No autopublish

    Publishing is a hard user gate; safe side work may continue, but preference or reward is not publication approval.

    +
    1publish hard gate
    1safe research side path
    0autopublish actions
    0real operations data exposed
    +
    Storyboard

    `creator-ops-fake-data-storyboard.md` provides a complete fake fixture with no live platform access, suitable for showing the user journey without exposing real operations data.

    Feedback contract

    `creator-ops-feedback-boundary-contract.md` separates gate decisions, preference hints, todo updates, boundary corrections, reward signals, and product improvement notes so preference is not mistaken for publish approval.

    Source status

    The contract requires every topic, insight, draft, and material item to carry source status; the public repo defaults to `synthetic_demo` instead of pretending to show real growth evidence.

    No autopublish

    Publishing is a hard user gate; safe side work may continue, but preference or reward is not publication approval.

    03

    LoopX behavior

    • 1
      The creative objective is durable goal state, not a hidden task inside chat.
    • 2
      Publish/no-publish is a user gate; research, organization, and source-status work can continue as safe side paths.
    • 3
      Feedback becomes a preference hint, gate decision, todo update, or boundary correction.
    • 4
      Before the next agent run, the user can see the blocked route, safe side path, and validation expectation.
    diff --git a/docs/showcases/cases/0620-creator-operator-case-spec.html b/docs/showcases/cases/0620-creator-operator-case-spec.html index 9fa86825..5ac785ae 100644 --- a/docs/showcases/cases/0620-creator-operator-case-spec.html +++ b/docs/showcases/cases/0620-creator-operator-case-spec.html @@ -56,6 +56,8 @@ .metric{border:1px solid rgba(255,255,255,.1);border-radius:10px;padding:17px 19px;background:#0c0c0e} .metric strong{display:block;font-size:24px;letter-spacing:-.02em;color:#fafafa} .metric span{display:block;margin-top:5px;font-family:'Geist Mono',ui-monospace,monospace;font-size:10.5px;color:#62666d;text-transform:uppercase} + .mini-metrics{display:flex;flex-wrap:wrap;gap:8px;margin-top:13px} + .mini-metrics span{font-family:'Geist Mono',ui-monospace,monospace;font-size:10.5px;color:#aeb3ba;border:1px solid rgba(255,255,255,.1);border-radius:5px;padding:4px 8px;background:#0b0b0c} .search{width:100%;height:42px;margin:22px 0 16px;border:1px solid rgba(255,255,255,.12);border-radius:8px;background:#0e0e10;color:#f1f2f3;padding:0 12px;font:14px 'Geist',system-ui,sans-serif} .cards{display:flex;flex-direction:column;gap:12px} .card{display:block;text-decoration:none;border:1px solid rgba(255,255,255,.1);border-radius:10px;background:#0e0e10;padding:18px 20px} @@ -63,8 +65,16 @@ .card .meta{font-family:'Geist Mono',ui-monospace,monospace;font-size:10.5px;color:#62666d;margin-bottom:8px} .card p{font-size:14px;margin-top:9px} .hide{display:none} + .experiment{margin-top:24px;border:1px solid rgba(255,255,255,.1);border-radius:10px;background:#0e0e10;padding:20px} + .experiment h3{font-size:18px} + .experiment p{font-size:14px;margin-top:8px} + .experiment-grid{display:grid;grid-template-columns:repeat(3,minmax(0,1fr));gap:12px;margin-top:16px} + .experiment-card{border:1px solid rgba(255,255,255,.1);border-radius:8px;background:#0b0b0c;padding:14px} + .experiment-card strong{display:block;font-size:14px;line-height:1.45} + .experiment-card span{display:block;margin-top:9px;font-size:13px;line-height:1.55;color:#9ea3aa} + .experiment-card em{display:block;margin-top:10px;font-style:normal;font-family:'Geist Mono',ui-monospace,monospace;font-size:10.5px;line-height:1.55;color:var(--accent)} footer{margin-top:76px;color:#62666d;font-family:'Geist Mono',ui-monospace,monospace;font-size:10.5px;line-height:1.7} - @media(max-width:720px){article{padding:52px 18px 90px}.panel-row{grid-template-columns:1fr}.panel-key{border-right:0;border-bottom:1px solid rgba(255,255,255,.07)}.metric-grid,.evidence-grid{grid-template-columns:1fr}.source-ref{grid-template-columns:1fr;gap:6px}} + @media(max-width:720px){article{padding:52px 18px 90px}.panel-row{grid-template-columns:1fr}.panel-key{border-right:0;border-bottom:1px solid rgba(255,255,255,.07)}.metric-grid,.evidence-grid,.experiment-grid{grid-template-columns:1fr}.source-ref{grid-template-columns:1fr;gap:6px}} @@ -110,15 +120,15 @@

    创作者-运营者长跑 Agent 案例

    01

    案例背景

    -

    这是 appendix case：它展示非技术创作者/运营者如何用 LoopX 管一个长期 research + planning loop，但还不是 top-card proof，因为没有真实用户公开证据。

    公开材料完全使用 synthetic data，重点是产品形态：趋势候选、偏好映射、insight board、draft queue、material library、人类反馈和 controlled replan。

    +

    这是 appendix case：它展示非技术创作者/运营者如何用 LoopX 管一个长期 research + planning loop，但还不是 top-card proof，因为没有真实用户公开证据。

    公开材料完全使用 synthetic data，重点不是证明增长，而是证明产品边界：研究和素材整理可以继续，发布、品牌判断和对外动作必须停在人类 gate。

    02

    仓库证据

    证明点

    创作与运营工作可以共享一个 gate-aware 的长期 agent loop。

    LoopX 介入

    creator-operator workflow、user gate、feedback capture、material library

    - -
    storyboard

    `creator-ops-fake-data-storyboard.md` 定义七个面板和完整 fake fixture，不需要 live platform access。

    feedback contract

    `creator-ops-feedback-boundary-contract.md` 把 gate decision、preference hint、todo update、boundary correction、reward signal、product improvement note 分开。

    source status

    contract 要求 topic、insight、draft、material item 都有 source status，public repo 默认 `synthetic_demo`。

    no autopublish

    publishing 是 hard user gate；safe side work 可以继续，但不能把偏好或 reward 当成发布授权。

    +
    1publish hard gate
    1safe research side path
    0autopublish 动作
    0真实运营数据暴露
    +
    storyboard

    `creator-ops-fake-data-storyboard.md` 给出完整 fake fixture，不需要 live platform access，适合公开展示用户旅程而不泄漏真实运营数据。

    feedback contract

    `creator-ops-feedback-boundary-contract.md` 把 gate decision、preference hint、todo update、boundary correction、reward signal 和 product improvement note 分开，避免把偏好误当发布授权。

    source status

    contract 要求 topic、insight、draft、material item 都携带 source status；public repo 默认 `synthetic_demo`，不伪装成真实增长证据。

    no autopublish

    publishing 是 hard user gate；safe side work 可以继续，但不能把偏好或 reward 当成发布授权。

    03

    LoopX 行为

    • 1
      creative objective 是 durable goal state，不是聊天窗口里的隐形任务。
    • 2
      publish/no-publish 是 user gate；research、整理和 source-status 改进可以作为 safe side path。
    • 3
      反馈被写成 preference hint、gate decision、todo update 或 boundary correction。
    • 4
      下一次 agent run 前，用户能看到 blocked route、safe side path 和 validation expectation。
    diff --git a/docs/showcases/cases/0623-agent-to-agent-pr-comments.en.html b/docs/showcases/cases/0623-agent-to-agent-pr-comments.en.html index 89621101..6a789093 100644 --- a/docs/showcases/cases/0623-agent-to-agent-pr-comments.en.html +++ b/docs/showcases/cases/0623-agent-to-agent-pr-comments.en.html @@ -56,6 +56,8 @@ .metric{border:1px solid rgba(255,255,255,.1);border-radius:10px;padding:17px 19px;background:#0c0c0e} .metric strong{display:block;font-size:24px;letter-spacing:-.02em;color:#fafafa} .metric span{display:block;margin-top:5px;font-family:'Geist Mono',ui-monospace,monospace;font-size:10.5px;color:#62666d;text-transform:uppercase} + .mini-metrics{display:flex;flex-wrap:wrap;gap:8px;margin-top:13px} + .mini-metrics span{font-family:'Geist Mono',ui-monospace,monospace;font-size:10.5px;color:#aeb3ba;border:1px solid rgba(255,255,255,.1);border-radius:5px;padding:4px 8px;background:#0b0b0c} .search{width:100%;height:42px;margin:22px 0 16px;border:1px solid rgba(255,255,255,.12);border-radius:8px;background:#0e0e10;color:#f1f2f3;padding:0 12px;font:14px 'Geist',system-ui,sans-serif} .cards{display:flex;flex-direction:column;gap:12px} .card{display:block;text-decoration:none;border:1px solid rgba(255,255,255,.1);border-radius:10px;background:#0e0e10;padding:18px 20px} @@ -63,8 +65,16 @@ .card .meta{font-family:'Geist Mono',ui-monospace,monospace;font-size:10.5px;color:#62666d;margin-bottom:8px} .card p{font-size:14px;margin-top:9px} .hide{display:none} + .experiment{margin-top:24px;border:1px solid rgba(255,255,255,.1);border-radius:10px;background:#0e0e10;padding:20px} + .experiment h3{font-size:18px} + .experiment p{font-size:14px;margin-top:8px} + .experiment-grid{display:grid;grid-template-columns:repeat(3,minmax(0,1fr));gap:12px;margin-top:16px} + .experiment-card{border:1px solid rgba(255,255,255,.1);border-radius:8px;background:#0b0b0c;padding:14px} + .experiment-card strong{display:block;font-size:14px;line-height:1.45} + .experiment-card span{display:block;margin-top:9px;font-size:13px;line-height:1.55;color:#9ea3aa} + .experiment-card em{display:block;margin-top:10px;font-style:normal;font-family:'Geist Mono',ui-monospace,monospace;font-size:10.5px;line-height:1.55;color:var(--accent)} footer{margin-top:76px;color:#62666d;font-family:'Geist Mono',ui-monospace,monospace;font-size:10.5px;line-height:1.7} - @media(max-width:720px){article{padding:52px 18px 90px}.panel-row{grid-template-columns:1fr}.panel-key{border-right:0;border-bottom:1px solid rgba(255,255,255,.07)}.metric-grid,.evidence-grid{grid-template-columns:1fr}.source-ref{grid-template-columns:1fr;gap:6px}} + @media(max-width:720px){article{padding:52px 18px 90px}.panel-row{grid-template-columns:1fr}.panel-key{border-right:0;border-bottom:1px solid rgba(255,255,255,.07)}.metric-grid,.evidence-grid,.experiment-grid{grid-template-columns:1fr}.source-ref{grid-template-columns:1fr;gap:6px}} @@ -110,15 +120,15 @@

    Agent-to-agent PR comment and fix loop

    01

    Case context

    -

    This case shows how PR review feedback can move across multiple agents without losing ownership. The important chain is comment, claim, handoff, fix, validation, and review packet rather than the chat transcript.

    The public repository contains the control-plane pieces: `claimed_by` appears in todo projection, review packets, event-sourced state, and multi-agent or side-agent prompt contracts.

    +

    This case shows how PR review feedback can move across multiple agents without losing ownership. The important chain is comment, claim, handoff, fix, validation, and review packet rather than the chat transcript.

    The user-facing value is that every feedback item can answer three questions: who owns it, where the fix evidence is, and who still needs to review it. Public evidence spans the event contract, review packet, heartbeat prompt, and validation fixtures.

    02

    Repository evidence

    Proof

    Multiple agents can coordinate around PR review comments without losing owner review.

    LoopX intervention

    claimed_by, handoff gate, review packet, comment/fix loop

    - -
    Todo ownership

    `event_sourced_state_contract_v0` defines `todo_claimed` as a canonical event for ownership, lease, or `claimed_by`.

    Review packet

    `loopx/review_packet.py` renders open todo `claimed_by` values into handoff and review text.

    Side-agent contract

    `docs/heartbeat-automation-prompt.md` requires side agents to self-merge only small validated evidence-backed work or create a claimed handoff todo.

    CLI smokes

    `examples/todo-lifecycle-cli-smoke.py` and `examples/todo-cli-smoke.py` cover claim, handoff successor, side-agent self-merge, and review handoff rules.

    +
    1owner per feedback item
    3owner/fix/review questions answered
    1review-packet handoff
    0unowned PR comment reminders
    +
    Todo ownership

    `event_sourced_state_contract_v0` defines `todo_claimed` as a canonical event for ownership, lease, or `claimed_by`.

    Review packet

    `loopx/review_packet.py` preserves `claimed_by` in open-todo rendering and handoff ranking so review packets can show ownership.

    Side-agent contract

    `docs/heartbeat-automation-prompt.md` requires side agents to self-merge only small validated evidence-backed work or create a claimed handoff todo.

    CLI smokes

    `examples/todo-lifecycle-cli-smoke.py` and `examples/todo-cli-smoke.py` cover claim, handoff successor, side-agent self-merge, review handoff, and same-agent handoff rejection.

    03

    LoopX behavior

    • 1
      PR feedback becomes an owned todo rather than a chat reminder.
    • 2
      Blocked or cross-lane fixes move through handoff gates and review packets instead of another agent guessing context.
    • 3
      The fixing agent leaves diff, validation, and remaining gates; follow-up work becomes a successor todo.
    • 4
      Owner review still happens on the PR or review surface; LoopX supplies state, evidence, and handoff.
    diff --git a/docs/showcases/cases/0623-agent-to-agent-pr-comments.html b/docs/showcases/cases/0623-agent-to-agent-pr-comments.html index 210b4e88..9e5c7750 100644 --- a/docs/showcases/cases/0623-agent-to-agent-pr-comments.html +++ b/docs/showcases/cases/0623-agent-to-agent-pr-comments.html @@ -56,6 +56,8 @@ .metric{border:1px solid rgba(255,255,255,.1);border-radius:10px;padding:17px 19px;background:#0c0c0e} .metric strong{display:block;font-size:24px;letter-spacing:-.02em;color:#fafafa} .metric span{display:block;margin-top:5px;font-family:'Geist Mono',ui-monospace,monospace;font-size:10.5px;color:#62666d;text-transform:uppercase} + .mini-metrics{display:flex;flex-wrap:wrap;gap:8px;margin-top:13px} + .mini-metrics span{font-family:'Geist Mono',ui-monospace,monospace;font-size:10.5px;color:#aeb3ba;border:1px solid rgba(255,255,255,.1);border-radius:5px;padding:4px 8px;background:#0b0b0c} .search{width:100%;height:42px;margin:22px 0 16px;border:1px solid rgba(255,255,255,.12);border-radius:8px;background:#0e0e10;color:#f1f2f3;padding:0 12px;font:14px 'Geist',system-ui,sans-serif} .cards{display:flex;flex-direction:column;gap:12px} .card{display:block;text-decoration:none;border:1px solid rgba(255,255,255,.1);border-radius:10px;background:#0e0e10;padding:18px 20px} @@ -63,8 +65,16 @@ .card .meta{font-family:'Geist Mono',ui-monospace,monospace;font-size:10.5px;color:#62666d;margin-bottom:8px} .card p{font-size:14px;margin-top:9px} .hide{display:none} + .experiment{margin-top:24px;border:1px solid rgba(255,255,255,.1);border-radius:10px;background:#0e0e10;padding:20px} + .experiment h3{font-size:18px} + .experiment p{font-size:14px;margin-top:8px} + .experiment-grid{display:grid;grid-template-columns:repeat(3,minmax(0,1fr));gap:12px;margin-top:16px} + .experiment-card{border:1px solid rgba(255,255,255,.1);border-radius:8px;background:#0b0b0c;padding:14px} + .experiment-card strong{display:block;font-size:14px;line-height:1.45} + .experiment-card span{display:block;margin-top:9px;font-size:13px;line-height:1.55;color:#9ea3aa} + .experiment-card em{display:block;margin-top:10px;font-style:normal;font-family:'Geist Mono',ui-monospace,monospace;font-size:10.5px;line-height:1.55;color:var(--accent)} footer{margin-top:76px;color:#62666d;font-family:'Geist Mono',ui-monospace,monospace;font-size:10.5px;line-height:1.7} - @media(max-width:720px){article{padding:52px 18px 90px}.panel-row{grid-template-columns:1fr}.panel-key{border-right:0;border-bottom:1px solid rgba(255,255,255,.07)}.metric-grid,.evidence-grid{grid-template-columns:1fr}.source-ref{grid-template-columns:1fr;gap:6px}} + @media(max-width:720px){article{padding:52px 18px 90px}.panel-row{grid-template-columns:1fr}.panel-key{border-right:0;border-bottom:1px solid rgba(255,255,255,.07)}.metric-grid,.evidence-grid,.experiment-grid{grid-template-columns:1fr}.source-ref{grid-template-columns:1fr;gap:6px}} @@ -110,15 +120,15 @@

    Agent to agent 回复 PR comment 和 PR Fix

    01

    案例背景

    -

    这个案例描述的是 PR review feedback 在多 agent 之间流转时如何不丢 ownership。重点不是聊天记录，而是 comment、claim、handoff、fix、validation、review packet 这条链。

    公开仓库里可以看到相关控制面已经被产品化：`claimed_by` 出现在 todo projection、review packet、event-sourced state 和多 agent/side-agent prompt 合约里。

    +

    这个案例描述的是 PR review feedback 在多 agent 之间流转时如何不丢 ownership。重点不是聊天记录，而是 comment、claim、handoff、fix、validation、review packet 这条链。

    对用户有价值的是每条反馈都能回答三个问题：谁负责、修复证据在哪里、还需要谁 review。公开仓库里的 event contract、review packet、heartbeat prompt 和 validation fixtures 共同证明这条 handoff 链。

    02

    仓库证据

    证明点

    多 agent lane 可以围绕 PR comment 协作，同时保留 owner review。

    LoopX 介入

    claimed_by、handoff gate、review packet、comment/fix loop

    - -
    todo ownership

    `event_sourced_state_contract_v0` 把 `todo_claimed` 定义为 canonical event，记录 ownership、lease 或 `claimed_by`。

    review packet

    `loopx/review_packet.py` 会把 open todo 的 `claimed_by` 显示进 handoff/review 文本。

    side-agent contract

    `docs/heartbeat-automation-prompt.md` 规定 side-agent 小变更可带 evidence 自合并，否则创建 claimed-by handoff todo。

    CLI smokes

    `examples/todo-lifecycle-cli-smoke.py`、`examples/todo-cli-smoke.py` 覆盖 claim、handoff successor、side-agent self-merge 和 review handoff 规则。

    +
    1每条反馈一个 owner
    3owner/fix/review 可回答问题
    1review packet handoff
    0无主 PR comment reminder
    +
    todo ownership

    `event_sourced_state_contract_v0` 把 `todo_claimed` 定义为 canonical event，记录 ownership、lease 或 `claimed_by`。

    review packet

    `loopx/review_packet.py` 在 open-todo rendering 和 handoff ranking 路径中保留 `claimed_by`，让 review packet 能显示 owner。

    side-agent contract

    `docs/heartbeat-automation-prompt.md` 规定 side-agent 小变更可带 evidence 自合并，否则创建 claimed-by handoff todo。

    CLI smokes

    `examples/todo-lifecycle-cli-smoke.py` 和 `examples/todo-cli-smoke.py` 覆盖 claim、handoff successor、side-agent self-merge、review handoff 和 same-agent handoff rejection。

    03

    LoopX 行为

    • 1
      PR feedback 先成为一个有 owner 的 todo，而不是 chat reminder。
    • 2
      被阻塞或跨 lane 的修复通过 handoff gate 和 review packet 传递，不允许另一个 agent 猜测上下文。
    • 3
      修复 agent 需要留下 diff、validation 和剩余 gate；后续工作创建 successor todo。
    • 4
      owner review 仍在 PR/review surface 上完成，LoopX 只提供状态、证据和 handoff。
    diff --git a/docs/showcases/cases/0623-overnight-project-refactor.en.html b/docs/showcases/cases/0623-overnight-project-refactor.en.html index 0e2b239d..bf598f60 100644 --- a/docs/showcases/cases/0623-overnight-project-refactor.en.html +++ b/docs/showcases/cases/0623-overnight-project-refactor.en.html @@ -56,6 +56,8 @@ .metric{border:1px solid rgba(255,255,255,.1);border-radius:10px;padding:17px 19px;background:#0c0c0e} .metric strong{display:block;font-size:24px;letter-spacing:-.02em;color:#fafafa} .metric span{display:block;margin-top:5px;font-family:'Geist Mono',ui-monospace,monospace;font-size:10.5px;color:#62666d;text-transform:uppercase} + .mini-metrics{display:flex;flex-wrap:wrap;gap:8px;margin-top:13px} + .mini-metrics span{font-family:'Geist Mono',ui-monospace,monospace;font-size:10.5px;color:#aeb3ba;border:1px solid rgba(255,255,255,.1);border-radius:5px;padding:4px 8px;background:#0b0b0c} .search{width:100%;height:42px;margin:22px 0 16px;border:1px solid rgba(255,255,255,.12);border-radius:8px;background:#0e0e10;color:#f1f2f3;padding:0 12px;font:14px 'Geist',system-ui,sans-serif} .cards{display:flex;flex-direction:column;gap:12px} .card{display:block;text-decoration:none;border:1px solid rgba(255,255,255,.1);border-radius:10px;background:#0e0e10;padding:18px 20px} @@ -63,8 +65,16 @@ .card .meta{font-family:'Geist Mono',ui-monospace,monospace;font-size:10.5px;color:#62666d;margin-bottom:8px} .card p{font-size:14px;margin-top:9px} .hide{display:none} + .experiment{margin-top:24px;border:1px solid rgba(255,255,255,.1);border-radius:10px;background:#0e0e10;padding:20px} + .experiment h3{font-size:18px} + .experiment p{font-size:14px;margin-top:8px} + .experiment-grid{display:grid;grid-template-columns:repeat(3,minmax(0,1fr));gap:12px;margin-top:16px} + .experiment-card{border:1px solid rgba(255,255,255,.1);border-radius:8px;background:#0b0b0c;padding:14px} + .experiment-card strong{display:block;font-size:14px;line-height:1.45} + .experiment-card span{display:block;margin-top:9px;font-size:13px;line-height:1.55;color:#9ea3aa} + .experiment-card em{display:block;margin-top:10px;font-style:normal;font-family:'Geist Mono',ui-monospace,monospace;font-size:10.5px;line-height:1.55;color:var(--accent)} footer{margin-top:76px;color:#62666d;font-family:'Geist Mono',ui-monospace,monospace;font-size:10.5px;line-height:1.7} - @media(max-width:720px){article{padding:52px 18px 90px}.panel-row{grid-template-columns:1fr}.panel-key{border-right:0;border-bottom:1px solid rgba(255,255,255,.07)}.metric-grid,.evidence-grid{grid-template-columns:1fr}.source-ref{grid-template-columns:1fr;gap:6px}} + @media(max-width:720px){article{padding:52px 18px 90px}.panel-row{grid-template-columns:1fr}.panel-key{border-right:0;border-bottom:1px solid rgba(255,255,255,.07)}.metric-grid,.evidence-grid,.experiment-grid{grid-template-columns:1fr}.source-ref{grid-template-columns:1fr;gap:6px}} @@ -110,15 +120,15 @@

    Overnight project refactor as PR-sized slices

    01

    Case context

    -

    This case addresses the risk of unattended refactoring: a long-running agent can mix cleanup, behavior change, discoveries, and stale plans into one broad diff.

    The public evidence is not a private overnight screenshot. It is the control-plane behavior already documented and smoke-tested in the repository: todo lifecycle, successor/supersede, validation writeback, and review packets.

    +

    This case addresses the risk of unattended refactoring: a long-running agent can mix cleanup, behavior change, discoveries, and stale plans into one broad diff.

    The public evidence is not a private overnight screenshot. It is the control-plane behavior already documented and smoke-tested in the repository: todo lifecycle, successor/supersede, validation writeback, and review packets. `todo-lifecycle-cli-smoke.py` carries regression coverage for successors, supersede, handoff, and self-merge.

    02

    Repository evidence

    Proof

    Long unattended refactors can split into moderate PR slices instead of one unreviewable diff.

    LoopX intervention

    loop, todo follow-up, supersede, PR-sized slices

    - -
    Successor path

    `docs/lark-kanban-control-plane-adapter.md` says real successors use `todo complete --next-*`, while replacements or narrower splits use `todo supersede --next-agent-todo`.

    Side-agent completion

    `docs/heartbeat-automation-prompt.md` requires nontrivial completion to create a successor todo or a no-follow-up rationale.

    CLI validation

    `examples/todo-lifecycle-cli-smoke.py` covers `--next-agent-todo`, `todo supersede`, claim inheritance, and handoff successors.

    Review shape

    `loopx review-packet` packages open todos, claimed_by, and handoff state for reviewer consumption.

    +
    1PR-sized slice at a time
    3successor/supersede/handoff routes
    1review gate for broad risk
    0giant-diff target
    +
    Successor path

    `docs/lark-kanban-control-plane-adapter.md` says real successors use `todo complete --next-*`, while replacements or narrower splits use `todo supersede --next-agent-todo`.

    Side-agent completion

    `docs/heartbeat-automation-prompt.md` requires nontrivial completion to create a successor todo or a no-follow-up rationale.

    CLI validation

    `examples/todo-lifecycle-cli-smoke.py` covers `--next-agent-todo` successors, `todo supersede`, claim inheritance, handoff successors, same-agent handoff rejection, and side-agent self-merge evidence.

    Review shape

    `loopx review-packet` packages open todos, claimed_by, and handoff state for reviewer consumption.

    03

    LoopX behavior

    • 1
      The current refactor slice must stay reviewable; overnight discoveries do not all land in one PR.
    • 2
      New discoveries become follow-up todos; changed routes supersede stale todos.
    • 3
      Each slice carries focused validation or doc/contract smoke evidence rather than raw agent traces.
    • 4
      Broad or unclear-risk slices route to review handoff instead of self-merge.
    diff --git a/docs/showcases/cases/0623-overnight-project-refactor.html b/docs/showcases/cases/0623-overnight-project-refactor.html index 2d43fca9..6f303945 100644 --- a/docs/showcases/cases/0623-overnight-project-refactor.html +++ b/docs/showcases/cases/0623-overnight-project-refactor.html @@ -56,6 +56,8 @@ .metric{border:1px solid rgba(255,255,255,.1);border-radius:10px;padding:17px 19px;background:#0c0c0e} .metric strong{display:block;font-size:24px;letter-spacing:-.02em;color:#fafafa} .metric span{display:block;margin-top:5px;font-family:'Geist Mono',ui-monospace,monospace;font-size:10.5px;color:#62666d;text-transform:uppercase} + .mini-metrics{display:flex;flex-wrap:wrap;gap:8px;margin-top:13px} + .mini-metrics span{font-family:'Geist Mono',ui-monospace,monospace;font-size:10.5px;color:#aeb3ba;border:1px solid rgba(255,255,255,.1);border-radius:5px;padding:4px 8px;background:#0b0b0c} .search{width:100%;height:42px;margin:22px 0 16px;border:1px solid rgba(255,255,255,.12);border-radius:8px;background:#0e0e10;color:#f1f2f3;padding:0 12px;font:14px 'Geist',system-ui,sans-serif} .cards{display:flex;flex-direction:column;gap:12px} .card{display:block;text-decoration:none;border:1px solid rgba(255,255,255,.1);border-radius:10px;background:#0e0e10;padding:18px 20px} @@ -63,8 +65,16 @@ .card .meta{font-family:'Geist Mono',ui-monospace,monospace;font-size:10.5px;color:#62666d;margin-bottom:8px} .card p{font-size:14px;margin-top:9px} .hide{display:none} + .experiment{margin-top:24px;border:1px solid rgba(255,255,255,.1);border-radius:10px;background:#0e0e10;padding:20px} + .experiment h3{font-size:18px} + .experiment p{font-size:14px;margin-top:8px} + .experiment-grid{display:grid;grid-template-columns:repeat(3,minmax(0,1fr));gap:12px;margin-top:16px} + .experiment-card{border:1px solid rgba(255,255,255,.1);border-radius:8px;background:#0b0b0c;padding:14px} + .experiment-card strong{display:block;font-size:14px;line-height:1.45} + .experiment-card span{display:block;margin-top:9px;font-size:13px;line-height:1.55;color:#9ea3aa} + .experiment-card em{display:block;margin-top:10px;font-style:normal;font-family:'Geist Mono',ui-monospace,monospace;font-size:10.5px;line-height:1.55;color:var(--accent)} footer{margin-top:76px;color:#62666d;font-family:'Geist Mono',ui-monospace,monospace;font-size:10.5px;line-height:1.7} - @media(max-width:720px){article{padding:52px 18px 90px}.panel-row{grid-template-columns:1fr}.panel-key{border-right:0;border-bottom:1px solid rgba(255,255,255,.07)}.metric-grid,.evidence-grid{grid-template-columns:1fr}.source-ref{grid-template-columns:1fr;gap:6px}} + @media(max-width:720px){article{padding:52px 18px 90px}.panel-row{grid-template-columns:1fr}.panel-key{border-right:0;border-bottom:1px solid rgba(255,255,255,.07)}.metric-grid,.evidence-grid,.experiment-grid{grid-template-columns:1fr}.source-ref{grid-template-columns:1fr;gap:6px}} @@ -110,15 +120,15 @@

    一晚上自主重构项目

    01

    案例背景

    -

    这个案例解决的是无人值守重构的风险：长时间 agent 容易把 cleanup、行为改变、发现的新问题和过期计划混成一个大 diff。

    LoopX 的公开证据不是某个私有夜间截图，而是 todo lifecycle、successor/supersede、validation writeback 和 review-packet 这些已经在仓库里有文档和 smoke 的控制面。

    +

    这个案例解决的是无人值守重构的风险：长时间 agent 容易把 cleanup、行为改变、发现的新问题和过期计划混成一个大 diff。

    LoopX 的公开证据不是某个私有夜间截图，而是 todo lifecycle、successor/supersede、validation writeback 和 review-packet 这些已经在仓库里有文档和 smoke 的控制面。`todo-lifecycle-cli-smoke.py` 里能看到 successor、supersede、handoff 和 self-merge 的完整回归覆盖。

    02

    仓库证据

    证明点

    无人值守 refactor 可以拆成 PR-sized slice。

    LoopX 介入

    loop、todo follow-up、supersede、PR-sized slices

    - -
    successor path

    `docs/lark-kanban-control-plane-adapter.md` 明确 real successor 使用 `todo complete --next-*`，replacement 或 narrower split 使用 `todo supersede --next-agent-todo`。

    side-agent completion

    `docs/heartbeat-automation-prompt.md` 要求非平凡完成创建 successor todo 或写 no-follow-up rationale。

    CLI validation

    `examples/todo-lifecycle-cli-smoke.py` 覆盖 `--next-agent-todo`、`todo supersede`、claim 继承和 handoff successor。

    review shape

    `loopx review-packet` 把当前 open todo、claimed_by 和 handoff 状态打包成 reviewer 可读的 packet。

    +
    1一次一个 PR-sized slice
    3successor/supersede/handoff 路线
    1宽风险 review gate
    0巨型 diff 目标
    +
    successor path

    `docs/lark-kanban-control-plane-adapter.md` 明确 real successor 使用 `todo complete --next-*`，replacement 或 narrower split 使用 `todo supersede --next-agent-todo`。

    side-agent completion

    `docs/heartbeat-automation-prompt.md` 要求非平凡完成创建 successor todo 或写 no-follow-up rationale。

    CLI validation

    `examples/todo-lifecycle-cli-smoke.py` 覆盖 `--next-agent-todo` successor、`todo supersede`、claim 继承、handoff successor、same-agent handoff rejection 和 side-agent self-merge evidence。

    review shape

    `loopx review-packet` 把当前 open todo、claimed_by 和 handoff 状态打包成 reviewer 可读的 packet。

    03

    LoopX 行为

    • 1
      当前 refactor slice 必须是可 review 的单位，不把整夜发现都塞进一个 PR。
    • 2
      发现的新工作写成 follow-up todo；路线变了就 supersede 旧 todo。
    • 3
      每个 slice 用 focused validation 或文档/contract smoke 证明，而不是依赖原始 agent trace。
    • 4
      大范围或风险不清楚的 slice 不自合并，进入 review handoff。
    diff --git a/docs/showcases/cases/0624-pr-issue-auto-fix.en.html b/docs/showcases/cases/0624-pr-issue-auto-fix.en.html index 07f0aacb..5af87377 100644 --- a/docs/showcases/cases/0624-pr-issue-auto-fix.en.html +++ b/docs/showcases/cases/0624-pr-issue-auto-fix.en.html @@ -56,6 +56,8 @@ .metric{border:1px solid rgba(255,255,255,.1);border-radius:10px;padding:17px 19px;background:#0c0c0e} .metric strong{display:block;font-size:24px;letter-spacing:-.02em;color:#fafafa} .metric span{display:block;margin-top:5px;font-family:'Geist Mono',ui-monospace,monospace;font-size:10.5px;color:#62666d;text-transform:uppercase} + .mini-metrics{display:flex;flex-wrap:wrap;gap:8px;margin-top:13px} + .mini-metrics span{font-family:'Geist Mono',ui-monospace,monospace;font-size:10.5px;color:#aeb3ba;border:1px solid rgba(255,255,255,.1);border-radius:5px;padding:4px 8px;background:#0b0b0c} .search{width:100%;height:42px;margin:22px 0 16px;border:1px solid rgba(255,255,255,.12);border-radius:8px;background:#0e0e10;color:#f1f2f3;padding:0 12px;font:14px 'Geist',system-ui,sans-serif} .cards{display:flex;flex-direction:column;gap:12px} .card{display:block;text-decoration:none;border:1px solid rgba(255,255,255,.1);border-radius:10px;background:#0e0e10;padding:18px 20px} @@ -63,8 +65,16 @@ .card .meta{font-family:'Geist Mono',ui-monospace,monospace;font-size:10.5px;color:#62666d;margin-bottom:8px} .card p{font-size:14px;margin-top:9px} .hide{display:none} + .experiment{margin-top:24px;border:1px solid rgba(255,255,255,.1);border-radius:10px;background:#0e0e10;padding:20px} + .experiment h3{font-size:18px} + .experiment p{font-size:14px;margin-top:8px} + .experiment-grid{display:grid;grid-template-columns:repeat(3,minmax(0,1fr));gap:12px;margin-top:16px} + .experiment-card{border:1px solid rgba(255,255,255,.1);border-radius:8px;background:#0b0b0c;padding:14px} + .experiment-card strong{display:block;font-size:14px;line-height:1.45} + .experiment-card span{display:block;margin-top:9px;font-size:13px;line-height:1.55;color:#9ea3aa} + .experiment-card em{display:block;margin-top:10px;font-style:normal;font-family:'Geist Mono',ui-monospace,monospace;font-size:10.5px;line-height:1.55;color:var(--accent)} footer{margin-top:76px;color:#62666d;font-family:'Geist Mono',ui-monospace,monospace;font-size:10.5px;line-height:1.7} - @media(max-width:720px){article{padding:52px 18px 90px}.panel-row{grid-template-columns:1fr}.panel-key{border-right:0;border-bottom:1px solid rgba(255,255,255,.07)}.metric-grid,.evidence-grid{grid-template-columns:1fr}.source-ref{grid-template-columns:1fr;gap:6px}} + @media(max-width:720px){article{padding:52px 18px 90px}.panel-row{grid-template-columns:1fr}.panel-key{border-right:0;border-bottom:1px solid rgba(255,255,255,.07)}.metric-grid,.evidence-grid,.experiment-grid{grid-template-columns:1fr}.source-ref{grid-template-columns:1fr;gap:6px}} @@ -110,15 +120,15 @@

    PR issue automatic fix loop

    01

    Case context

    -

    This case turns a PR issue, review comment, or issue text into an executable repair loop. It is not simply reading a comment and editing code; the path goes through metadata/intake, repro, branch-local patch, validation, and review packet.

    The repository already contains the issue-fix capability: `loopx/capabilities/issue_fix/`, the `loopx issue-fix ...` CLI entry, protocol docs, and public smokes.

    +

    This case turns a PR issue, review comment, or issue text into an executable repair loop. It is not simply reading a comment and editing code; the path goes through metadata/intake, repro, branch-local patch, validation, and review packet.

    The user-facing value is that feedback no longer sits in a comment thread: it enters a loop with an owner, repro path, branch-local fix, validation, and review handoff. The public evidence proves the loop and its boundaries without exposing raw issue bodies, private timelines, or local paths.

    02

    Repository evidence

    Proof

    Issue and review feedback can enter an executable repair loop.

    LoopX intervention

    issue-fix workflow, command pack, repro smoke, PR review feedback

    - -
    Product entry

    `docs/capabilities/issue-fix/README.md` names `loopx issue-fix ...`, the content-ops bridge, protocol docs, and smokes.

    Workflow contract

    `issue_fix_workflow_contract_v0` defines metadata preview, intake classification, workflow plan, todo writeback, caller repo branch, validation, PR review packet, and gate handling.

    Executable loop

    `issue_fix_acceptance_loop_v0` includes the acceptance fixture, repo-branch fixture, and caller-approved repo branch mode.

    Validation surface

    Smokes cover workflow planning, workflow contract, metadata/intake, and acceptance-loop behavior.

    +
    5repair-loop stages
    1branch-ready fix packet
    1review handoff preserved
    0raw issue bodies in public artifact
    +
    Product entry

    `docs/capabilities/issue-fix/README.md` names `loopx issue-fix ...`, the content-ops bridge, protocol docs, and smokes; the maintainer-facing action is turning feedback into an executable fix packet.

    Workflow contract

    `issue_fix_workflow_contract_v0` defines metadata preview, intake classification, workflow plan, todo writeback, caller repo branch, validation, PR review packet, and gate handling.

    Executable loop

    `issue_fix_acceptance_loop_v0` includes the acceptance fixture, repo-branch fixture, and caller-approved repo branch mode.

    Validation surface

    Focused smokes protect the key boundaries for metadata preview, content-ops intake, workflow plan, workflow contract, acceptance loop, and end-to-end workflow behavior.

    03

    LoopX behavior

    • 1
      Public metadata can enter packets; raw issue bodies, comment bodies, timelines, and provider payloads remain gated sources.
    • 2
      Accepted candidates become ordered LoopX todos: repro smoke, code-context route, branch-local patch, validation, and review-packet readiness.
    • 3
      Caller-approved repo mode inspects the local repo only under `--execute` and reports repo-relative changed files plus validation pass/fail.
    • 4
      External comments, PR creation, merge, publish, destructive git, and production action remain explicit gates.
    diff --git a/docs/showcases/cases/0624-pr-issue-auto-fix.html b/docs/showcases/cases/0624-pr-issue-auto-fix.html index ba8587fb..f42f1be5 100644 --- a/docs/showcases/cases/0624-pr-issue-auto-fix.html +++ b/docs/showcases/cases/0624-pr-issue-auto-fix.html @@ -56,6 +56,8 @@ .metric{border:1px solid rgba(255,255,255,.1);border-radius:10px;padding:17px 19px;background:#0c0c0e} .metric strong{display:block;font-size:24px;letter-spacing:-.02em;color:#fafafa} .metric span{display:block;margin-top:5px;font-family:'Geist Mono',ui-monospace,monospace;font-size:10.5px;color:#62666d;text-transform:uppercase} + .mini-metrics{display:flex;flex-wrap:wrap;gap:8px;margin-top:13px} + .mini-metrics span{font-family:'Geist Mono',ui-monospace,monospace;font-size:10.5px;color:#aeb3ba;border:1px solid rgba(255,255,255,.1);border-radius:5px;padding:4px 8px;background:#0b0b0c} .search{width:100%;height:42px;margin:22px 0 16px;border:1px solid rgba(255,255,255,.12);border-radius:8px;background:#0e0e10;color:#f1f2f3;padding:0 12px;font:14px 'Geist',system-ui,sans-serif} .cards{display:flex;flex-direction:column;gap:12px} .card{display:block;text-decoration:none;border:1px solid rgba(255,255,255,.1);border-radius:10px;background:#0e0e10;padding:18px 20px} @@ -63,8 +65,16 @@ .card .meta{font-family:'Geist Mono',ui-monospace,monospace;font-size:10.5px;color:#62666d;margin-bottom:8px} .card p{font-size:14px;margin-top:9px} .hide{display:none} + .experiment{margin-top:24px;border:1px solid rgba(255,255,255,.1);border-radius:10px;background:#0e0e10;padding:20px} + .experiment h3{font-size:18px} + .experiment p{font-size:14px;margin-top:8px} + .experiment-grid{display:grid;grid-template-columns:repeat(3,minmax(0,1fr));gap:12px;margin-top:16px} + .experiment-card{border:1px solid rgba(255,255,255,.1);border-radius:8px;background:#0b0b0c;padding:14px} + .experiment-card strong{display:block;font-size:14px;line-height:1.45} + .experiment-card span{display:block;margin-top:9px;font-size:13px;line-height:1.55;color:#9ea3aa} + .experiment-card em{display:block;margin-top:10px;font-style:normal;font-family:'Geist Mono',ui-monospace,monospace;font-size:10.5px;line-height:1.55;color:var(--accent)} footer{margin-top:76px;color:#62666d;font-family:'Geist Mono',ui-monospace,monospace;font-size:10.5px;line-height:1.7} - @media(max-width:720px){article{padding:52px 18px 90px}.panel-row{grid-template-columns:1fr}.panel-key{border-right:0;border-bottom:1px solid rgba(255,255,255,.07)}.metric-grid,.evidence-grid{grid-template-columns:1fr}.source-ref{grid-template-columns:1fr;gap:6px}} + @media(max-width:720px){article{padding:52px 18px 90px}.panel-row{grid-template-columns:1fr}.panel-key{border-right:0;border-bottom:1px solid rgba(255,255,255,.07)}.metric-grid,.evidence-grid,.experiment-grid{grid-template-columns:1fr}.source-ref{grid-template-columns:1fr;gap:6px}} @@ -110,15 +120,15 @@

    PR Issue 自动 Fix

    01

    案例背景

    -

    这个案例把 PR issue、review comment 或 issue text 转成可执行修复闭环。它不是“读一段评论就改代码”，而是先做 metadata/intake、repro、branch-local patch、validation 和 review packet。

    仓库已经有 issue-fix capability：模块在 `loopx/capabilities/issue_fix/`，CLI 入口是 `loopx issue-fix ...`，协议文档和 smoke 都在公开仓库内。

    +

    这个案例把 PR issue、review comment 或 issue text 转成可执行修复闭环。它不是“读一段评论就改代码”，而是先做 metadata/intake、repro、branch-local patch、validation 和 review packet。

    对用户有价值的是反馈不再停在评论区：它会进入一个有 owner、有复现、有分支修复、有验证、有 review handoff 的闭环。公开证据只证明这条闭环和边界，不把 raw issue body、私有 timeline 或本地路径带进页面。

    02

    仓库证据

    证明点

    Issue 和 review 反馈可以变成受控、可执行的修复循环。

    LoopX 介入

    issue-fix workflow、command pack、repro smoke、PR review feedback

    - -
    产品入口

    `docs/capabilities/issue-fix/README.md` 声明 `loopx issue-fix ...`、content-ops bridge、protocol docs 和 smoke。

    工作流协议

    `issue_fix_workflow_contract_v0` 明确 metadata preview、intake classification、workflow plan、todo writeback、caller repo branch、validation、PR review packet 和 gate handling。

    可执行闭环

    `issue_fix_acceptance_loop_v0` 包含 acceptance fixture、repo-branch fixture 和 caller-approved repo branch mode。

    验证面

    `examples/issue-fix-workflow-plan-smoke.py`、`examples/issue-fix-workflow-contract-smoke.py`、`examples/issue-fix-acceptance-loop-smoke.py` 等 smoke 覆盖这个路径。

    +
    5修复闭环阶段
    1branch-ready 修复包
    1review handoff 保留
    0公开 artifact 中 raw issue body
    +
    产品入口

    `docs/capabilities/issue-fix/README.md` 声明 `loopx issue-fix ...`、content-ops bridge、protocol docs 和 smoke；入口面向维护者的真实动作是把反馈转成可执行修复包。

    工作流协议

    `issue_fix_workflow_contract_v0` 明确 metadata preview、intake classification、workflow plan、todo writeback、caller repo branch、validation、PR review packet 和 gate handling。

    可执行闭环

    `issue_fix_acceptance_loop_v0` 包含 acceptance fixture、repo-branch fixture 和 caller-approved repo branch mode。

    验证面

    focused smokes 保护 metadata preview、content-ops intake、workflow plan、workflow contract、acceptance loop 和端到端 workflow 的关键边界。

    03

    LoopX 行为

    • 1
      public metadata 可以进入 packet；raw issue body、comment body、timeline、provider payload 都是 gated source。
    • 2
      accepted candidates 写成有序 LoopX todos：repro smoke、code-context route、branch-local patch、validation、review-packet readiness。
    • 3
      caller-approved repo 模式只在 `--execute` 时检查本地 repo，并且输出 repo-relative changed files、validation pass/fail 和 PR-readiness。
    • 4
      外部 comment、PR creation、merge、publish、destructive git 和 production action 都保留为显式 gate。
    diff --git a/docs/showcases/cases/0627-overnight-pr-batch.en.html b/docs/showcases/cases/0627-overnight-pr-batch.en.html index 14aa1760..292cb490 100644 --- a/docs/showcases/cases/0627-overnight-pr-batch.en.html +++ b/docs/showcases/cases/0627-overnight-pr-batch.en.html @@ -56,6 +56,8 @@ .metric{border:1px solid rgba(255,255,255,.1);border-radius:10px;padding:17px 19px;background:#0c0c0e} .metric strong{display:block;font-size:24px;letter-spacing:-.02em;color:#fafafa} .metric span{display:block;margin-top:5px;font-family:'Geist Mono',ui-monospace,monospace;font-size:10.5px;color:#62666d;text-transform:uppercase} + .mini-metrics{display:flex;flex-wrap:wrap;gap:8px;margin-top:13px} + .mini-metrics span{font-family:'Geist Mono',ui-monospace,monospace;font-size:10.5px;color:#aeb3ba;border:1px solid rgba(255,255,255,.1);border-radius:5px;padding:4px 8px;background:#0b0b0c} .search{width:100%;height:42px;margin:22px 0 16px;border:1px solid rgba(255,255,255,.12);border-radius:8px;background:#0e0e10;color:#f1f2f3;padding:0 12px;font:14px 'Geist',system-ui,sans-serif} .cards{display:flex;flex-direction:column;gap:12px} .card{display:block;text-decoration:none;border:1px solid rgba(255,255,255,.1);border-radius:10px;background:#0e0e10;padding:18px 20px} @@ -63,8 +65,16 @@ .card .meta{font-family:'Geist Mono',ui-monospace,monospace;font-size:10.5px;color:#62666d;margin-bottom:8px} .card p{font-size:14px;margin-top:9px} .hide{display:none} + .experiment{margin-top:24px;border:1px solid rgba(255,255,255,.1);border-radius:10px;background:#0e0e10;padding:20px} + .experiment h3{font-size:18px} + .experiment p{font-size:14px;margin-top:8px} + .experiment-grid{display:grid;grid-template-columns:repeat(3,minmax(0,1fr));gap:12px;margin-top:16px} + .experiment-card{border:1px solid rgba(255,255,255,.1);border-radius:8px;background:#0b0b0c;padding:14px} + .experiment-card strong{display:block;font-size:14px;line-height:1.45} + .experiment-card span{display:block;margin-top:9px;font-size:13px;line-height:1.55;color:#9ea3aa} + .experiment-card em{display:block;margin-top:10px;font-style:normal;font-family:'Geist Mono',ui-monospace,monospace;font-size:10.5px;line-height:1.55;color:var(--accent)} footer{margin-top:76px;color:#62666d;font-family:'Geist Mono',ui-monospace,monospace;font-size:10.5px;line-height:1.7} - @media(max-width:720px){article{padding:52px 18px 90px}.panel-row{grid-template-columns:1fr}.panel-key{border-right:0;border-bottom:1px solid rgba(255,255,255,.07)}.metric-grid,.evidence-grid{grid-template-columns:1fr}.source-ref{grid-template-columns:1fr;gap:6px}} + @media(max-width:720px){article{padding:52px 18px 90px}.panel-row{grid-template-columns:1fr}.panel-key{border-right:0;border-bottom:1px solid rgba(255,255,255,.07)}.metric-grid,.evidence-grid,.experiment-grid{grid-template-columns:1fr}.source-ref{grid-template-columns:1fr;gap:6px}} @@ -117,7 +127,7 @@

    Overnight PR batch with reviewable control

    Proof

    High-throughput multi-lane work can remain PR-sized, reviewable, and merge-safe.

    LoopX intervention

    todo claim, review packet, self-merge boundary, focused smoke, public-boundary scan

    -
    22merged PR commits
    10hpublic window
    +
    22reviewable merged commits
    10PR-numbered commits
    10hpublic evidence window
    0raw-agent-log dependency
    Public window

    2026-06-27 01:29 to 11:29 +08:00, a 10-hour Git evidence window.

    Change shape

    22 merged commits across docs, runtime, status/quota, benchmark contracts, smokes, and release/runtime guardrails.

    Review boundary

    Work landed as PR-sized slices instead of one broad diff that maintainers would have to trust blindly.

    Reproduction

    The case document gives `git log --since ... --until ...` and `git log --numstat` commands.

    03

    LoopX behavior

    diff --git a/docs/showcases/cases/0627-overnight-pr-batch.html b/docs/showcases/cases/0627-overnight-pr-batch.html index a6eb4024..11d85d62 100644 --- a/docs/showcases/cases/0627-overnight-pr-batch.html +++ b/docs/showcases/cases/0627-overnight-pr-batch.html @@ -56,6 +56,8 @@ .metric{border:1px solid rgba(255,255,255,.1);border-radius:10px;padding:17px 19px;background:#0c0c0e} .metric strong{display:block;font-size:24px;letter-spacing:-.02em;color:#fafafa} .metric span{display:block;margin-top:5px;font-family:'Geist Mono',ui-monospace,monospace;font-size:10.5px;color:#62666d;text-transform:uppercase} + .mini-metrics{display:flex;flex-wrap:wrap;gap:8px;margin-top:13px} + .mini-metrics span{font-family:'Geist Mono',ui-monospace,monospace;font-size:10.5px;color:#aeb3ba;border:1px solid rgba(255,255,255,.1);border-radius:5px;padding:4px 8px;background:#0b0b0c} .search{width:100%;height:42px;margin:22px 0 16px;border:1px solid rgba(255,255,255,.12);border-radius:8px;background:#0e0e10;color:#f1f2f3;padding:0 12px;font:14px 'Geist',system-ui,sans-serif} .cards{display:flex;flex-direction:column;gap:12px} .card{display:block;text-decoration:none;border:1px solid rgba(255,255,255,.1);border-radius:10px;background:#0e0e10;padding:18px 20px} @@ -63,8 +65,16 @@ .card .meta{font-family:'Geist Mono',ui-monospace,monospace;font-size:10.5px;color:#62666d;margin-bottom:8px} .card p{font-size:14px;margin-top:9px} .hide{display:none} + .experiment{margin-top:24px;border:1px solid rgba(255,255,255,.1);border-radius:10px;background:#0e0e10;padding:20px} + .experiment h3{font-size:18px} + .experiment p{font-size:14px;margin-top:8px} + .experiment-grid{display:grid;grid-template-columns:repeat(3,minmax(0,1fr));gap:12px;margin-top:16px} + .experiment-card{border:1px solid rgba(255,255,255,.1);border-radius:8px;background:#0b0b0c;padding:14px} + .experiment-card strong{display:block;font-size:14px;line-height:1.45} + .experiment-card span{display:block;margin-top:9px;font-size:13px;line-height:1.55;color:#9ea3aa} + .experiment-card em{display:block;margin-top:10px;font-style:normal;font-family:'Geist Mono',ui-monospace,monospace;font-size:10.5px;line-height:1.55;color:var(--accent)} footer{margin-top:76px;color:#62666d;font-family:'Geist Mono',ui-monospace,monospace;font-size:10.5px;line-height:1.7} - @media(max-width:720px){article{padding:52px 18px 90px}.panel-row{grid-template-columns:1fr}.panel-key{border-right:0;border-bottom:1px solid rgba(255,255,255,.07)}.metric-grid,.evidence-grid{grid-template-columns:1fr}.source-ref{grid-template-columns:1fr;gap:6px}} + @media(max-width:720px){article{padding:52px 18px 90px}.panel-row{grid-template-columns:1fr}.panel-key{border-right:0;border-bottom:1px solid rgba(255,255,255,.07)}.metric-grid,.evidence-grid,.experiment-grid{grid-template-columns:1fr}.source-ref{grid-template-columns:1fr;gap:6px}} @@ -117,7 +127,7 @@

    一晚 30 个高价值 PR

    证明点

    高吞吐多 lane 工作也可以保持 PR 粒度、可审阅、可合并。

    LoopX 介入

    todo claim、review packet、自合并边界、focused smoke、public-boundary scan

    -
    22merged PR commits
    10hpublic window
    +
    22可审阅 merged commits
    10带 PR 编号的 commits
    10h公开证据窗口
    0raw agent logs 依赖
    公开窗口

    2026-06-27 01:29 到 11:29 +08:00，10 小时 Git 证据窗口。

    变更粒度

    22 个 merged commits 覆盖 docs、runtime、status/quota、benchmark contracts、smokes 和 release/runtime guardrails。

    审阅边界

    工作以 PR-sized slices 落地，而不是一个需要 maintainer 盲信的巨大 diff。

    复现方式

    case 文档给出 `git log --since ... --until ...` 和 `git log --numstat` 命令。

    03

    LoopX 行为

    diff --git a/docs/showcases/frontend-surface.md b/docs/showcases/frontend-surface.md index da20d3c9..337cb827 100644 --- a/docs/showcases/frontend-surface.md +++ b/docs/showcases/frontend-surface.md @@ -36,9 +36,10 @@ Use these catalog fields directly: | `user_value` | Outcome in plain language. | | `evidence_boundary` | Redaction and claim boundary drawer. | | `frontend_card.visual_metaphor` | Suggested visual treatment. | -| `frontend_card.primary_metric_hint` | Lightweight signal, not a hard claim. | +| `frontend_card.primary_metric_hint` | Lightweight value signal for leaders/users, not implementation trivia. | | `frontend_card.badges` | Compact chips. | | `frontend_card.story_beats` | Backward-compatible field for the case detail evidence sequence. New copy should render it as evidence, not as author notes. | +| `evidence_metrics` | Optional compact value metrics. Use outcome or boundary signals such as reviewable commits, user wait avoided, gated action prevented, compression range, or public evidence window. Do not use raw file counts, smoke counts, panel counts, or other implementation-surface trivia as the main proof. | | `workload_signal.efficiency_model` | Optional evidence panel for conservative baseline-vs-actual efficiency modeling. | ## First Screen diff --git a/docs/showcases/index.en.html b/docs/showcases/index.en.html index 4bd0c0c5..356b31d5 100644 --- a/docs/showcases/index.en.html +++ b/docs/showcases/index.en.html @@ -56,6 +56,8 @@ .metric{border:1px solid rgba(255,255,255,.1);border-radius:10px;padding:17px 19px;background:#0c0c0e} .metric strong{display:block;font-size:24px;letter-spacing:-.02em;color:#fafafa} .metric span{display:block;margin-top:5px;font-family:'Geist Mono',ui-monospace,monospace;font-size:10.5px;color:#62666d;text-transform:uppercase} + .mini-metrics{display:flex;flex-wrap:wrap;gap:8px;margin-top:13px} + .mini-metrics span{font-family:'Geist Mono',ui-monospace,monospace;font-size:10.5px;color:#aeb3ba;border:1px solid rgba(255,255,255,.1);border-radius:5px;padding:4px 8px;background:#0b0b0c} .search{width:100%;height:42px;margin:22px 0 16px;border:1px solid rgba(255,255,255,.12);border-radius:8px;background:#0e0e10;color:#f1f2f3;padding:0 12px;font:14px 'Geist',system-ui,sans-serif} .cards{display:flex;flex-direction:column;gap:12px} .card{display:block;text-decoration:none;border:1px solid rgba(255,255,255,.1);border-radius:10px;background:#0e0e10;padding:18px 20px} @@ -63,8 +65,16 @@ .card .meta{font-family:'Geist Mono',ui-monospace,monospace;font-size:10.5px;color:#62666d;margin-bottom:8px} .card p{font-size:14px;margin-top:9px} .hide{display:none} + .experiment{margin-top:24px;border:1px solid rgba(255,255,255,.1);border-radius:10px;background:#0e0e10;padding:20px} + .experiment h3{font-size:18px} + .experiment p{font-size:14px;margin-top:8px} + .experiment-grid{display:grid;grid-template-columns:repeat(3,minmax(0,1fr));gap:12px;margin-top:16px} + .experiment-card{border:1px solid rgba(255,255,255,.1);border-radius:8px;background:#0b0b0c;padding:14px} + .experiment-card strong{display:block;font-size:14px;line-height:1.45} + .experiment-card span{display:block;margin-top:9px;font-size:13px;line-height:1.55;color:#9ea3aa} + .experiment-card em{display:block;margin-top:10px;font-style:normal;font-family:'Geist Mono',ui-monospace,monospace;font-size:10.5px;line-height:1.55;color:var(--accent)} footer{margin-top:76px;color:#62666d;font-family:'Geist Mono',ui-monospace,monospace;font-size:10.5px;line-height:1.7} - @media(max-width:720px){article{padding:52px 18px 90px}.panel-row{grid-template-columns:1fr}.panel-key{border-right:0;border-bottom:1px solid rgba(255,255,255,.07)}.metric-grid,.evidence-grid{grid-template-columns:1fr}.source-ref{grid-template-columns:1fr;gap:6px}} + @media(max-width:720px){article{padding:52px 18px 90px}.panel-row{grid-template-columns:1fr}.panel-key{border-right:0;border-bottom:1px solid rgba(255,255,255,.07)}.metric-grid,.evidence-grid,.experiment-grid{grid-template-columns:1fr}.source-ref{grid-template-columns:1fr;gap:6px}} @@ -111,76 +121,108 @@

    Showcase & Good Case

    01

    Top showcase cases

    - +
    2026-06-27 · public_evidence_case

    Overnight PR batch with reviewable control

    An overnight LoopX run can produce many PR-sized slices while keeping review, validation, and public evidence boundaries visible.

    Proof: High-throughput multi-lane work can remain PR-sized, reviewable, and merge-safe.

    +
    22 · reviewable merged commits10 · PR-numbered commits10h · public evidence window
    high_throughput_reviewable_workpr_sized_slicesself_merge_policyvalidation_writeback
     - +
    2026-06-24 · public_safe_pattern_case

    PR issue automatic fix loop

    Review feedback should become an ordered repair workflow with repro, fix, validation, and reviewer handoff.

    Proof: Issue and review feedback can enter an executable repair loop.

    +
    5 · repair-loop stages1 · branch-ready fix packet1 · review handoff preserved
    issue_fix_workflowreview_feedbackrepro_smokecommand_pack
     - +
    2026-06-23 · public_safe_pattern_case

    Agent-to-agent PR comment and fix loop

    PR review feedback can become an owned agent todo with fix evidence instead of a loose chat reminder.

    Proof: Multiple agents can coordinate around PR review comments without losing owner review.

    +
    1 · owner per feedback item3 · owner/fix/review questions answered1 · review-packet handoff
    agent_to_agent_handoffpr_comment_loopreview_packetclaimed_todo
     - +
    2026-06-23 · public_safe_pattern_case

    Overnight project refactor as PR-sized slices

    A broad refactor can run overnight while staying split into human-sized review units.

    Proof: Long unattended refactors can split into moderate PR slices instead of one unreviewable diff.

    +
    1 · PR-sized slice at a time3 · successor/supersede/handoff routes1 · review gate for broad risk
    long_unattended_goalpr_sized_slicestodo_follow_upsupersede
     - +
    2026-06-19 · public_safe_interactive_case

    Dynamic workflow for hardware-agent development

    A fuzzy long-running engineering goal needs a shared control plane when multiple worker agents participate.

    Proof: Fuzzy goals, multiple workers, and long unattended runs can still converge.

    +
    5 · public-safe hardware workflows3 · LoopX/orchestrator/worker role split1 · canonical artifact preserved
    dynamic_workflowmulti_agent_coordinationshared_control_planelong_unattended_goal
     - +
    2026-06-19 · public_evidence_case

    LoopX self-iteration loop

    A high-churn LoopX repo stayed legible while benchmark, product, docs, planning, and side-agent lanes moved in parallel.

    Proof: A high-churn multi-lane project can keep state, boundaries, and evidence coherent.

    +
    801 · public commits244 · commits since Jun 1859-92d · AI-assisted baseline
    self_iterationside_agent_scopetodo_claim_ownershipidentity_aware_prompt
     - +
    2026-06-17 · reproducible_synthetic_demo

    Blocked P0 with safe P1/P2 rotation

    A gated P0 lane should not stall a whole long-running goal when safe fallback work exists.

    Proof: A user decision should not block all safe work.

    +
    1 · concrete P0 user decision1 · safe fallback lane0 · gated-lane auto-progress
    blocked_priority_fallbackconcrete_user_gatesafe_fallback_workquota_discipline
    +
    +

    Experimental today-value path

    +

    A lower-priority entry point for users who want to pick one useful LoopX capability today without replacing the showcase first screen.

    +
    +
    + PR review/comment -> fix loop + Branch-ready fix packet with repro, smoke result, and remaining review owner. + Fewer dropped review threads. +
    + +
    + Overnight PR-sized refactor + Reviewable slice list, validation notes, successor todo, and merge boundary. + More merged commits without a giant diff audit. +
    + +
    + P0 blocked -> safe fallback + Kernel projection of the exact user gate, safe fallback todo, quota decision, and evidence boundary. + Less idle agent time while preserving human judgment. +
    +
    +
    +
    02

    Appendix case

    - +
    2026-06-20 · public_safe_case_spec

    Creator-operator long-running agent case

    A creator-operator needs a long-running agent loop that keeps research moving while publishing decisions stay gated.

    Proof: A creator-operator needs a long-running agent loop that keeps research moving while publishing decisions stay gated.

    +
    1 · publish hard gate1 · safe research side path0 · autopublish actions
    creator_operator_workflowgate_aware_continuationfeedback_capturesafe_side_path
    diff --git a/docs/showcases/index.html b/docs/showcases/index.html index a92b4582..35b56f52 100644 --- a/docs/showcases/index.html +++ b/docs/showcases/index.html @@ -56,6 +56,8 @@ .metric{border:1px solid rgba(255,255,255,.1);border-radius:10px;padding:17px 19px;background:#0c0c0e} .metric strong{display:block;font-size:24px;letter-spacing:-.02em;color:#fafafa} .metric span{display:block;margin-top:5px;font-family:'Geist Mono',ui-monospace,monospace;font-size:10.5px;color:#62666d;text-transform:uppercase} + .mini-metrics{display:flex;flex-wrap:wrap;gap:8px;margin-top:13px} + .mini-metrics span{font-family:'Geist Mono',ui-monospace,monospace;font-size:10.5px;color:#aeb3ba;border:1px solid rgba(255,255,255,.1);border-radius:5px;padding:4px 8px;background:#0b0b0c} .search{width:100%;height:42px;margin:22px 0 16px;border:1px solid rgba(255,255,255,.12);border-radius:8px;background:#0e0e10;color:#f1f2f3;padding:0 12px;font:14px 'Geist',system-ui,sans-serif} .cards{display:flex;flex-direction:column;gap:12px} .card{display:block;text-decoration:none;border:1px solid rgba(255,255,255,.1);border-radius:10px;background:#0e0e10;padding:18px 20px} @@ -63,8 +65,16 @@ .card .meta{font-family:'Geist Mono',ui-monospace,monospace;font-size:10.5px;color:#62666d;margin-bottom:8px} .card p{font-size:14px;margin-top:9px} .hide{display:none} + .experiment{margin-top:24px;border:1px solid rgba(255,255,255,.1);border-radius:10px;background:#0e0e10;padding:20px} + .experiment h3{font-size:18px} + .experiment p{font-size:14px;margin-top:8px} + .experiment-grid{display:grid;grid-template-columns:repeat(3,minmax(0,1fr));gap:12px;margin-top:16px} + .experiment-card{border:1px solid rgba(255,255,255,.1);border-radius:8px;background:#0b0b0c;padding:14px} + .experiment-card strong{display:block;font-size:14px;line-height:1.45} + .experiment-card span{display:block;margin-top:9px;font-size:13px;line-height:1.55;color:#9ea3aa} + .experiment-card em{display:block;margin-top:10px;font-style:normal;font-family:'Geist Mono',ui-monospace,monospace;font-size:10.5px;line-height:1.55;color:var(--accent)} footer{margin-top:76px;color:#62666d;font-family:'Geist Mono',ui-monospace,monospace;font-size:10.5px;line-height:1.7} - @media(max-width:720px){article{padding:52px 18px 90px}.panel-row{grid-template-columns:1fr}.panel-key{border-right:0;border-bottom:1px solid rgba(255,255,255,.07)}.metric-grid,.evidence-grid{grid-template-columns:1fr}.source-ref{grid-template-columns:1fr;gap:6px}} + @media(max-width:720px){article{padding:52px 18px 90px}.panel-row{grid-template-columns:1fr}.panel-key{border-right:0;border-bottom:1px solid rgba(255,255,255,.07)}.metric-grid,.evidence-grid,.experiment-grid{grid-template-columns:1fr}.source-ref{grid-template-columns:1fr;gap:6px}} @@ -111,76 +121,108 @@

    Showcase & Good Case

    01

    顶部 Showcase 案例

    - +
    2026-06-27 · public_evidence_case

    一晚 30 个高价值 PR

    高吞吐、多车道推进仍保持 PR 级可 review、可合并。

    证明点: 高吞吐多 lane 工作也可以保持 PR 粒度、可审阅、可合并。

    +
    22 · 可审阅 merged commits10 · 带 PR 编号的 commits10h · 公开证据窗口
    high_throughput_reviewable_workpr_sized_slicesself_merge_policyvalidation_writeback
     - +
    2026-06-24 · public_safe_pattern_case

    PR Issue 自动 Fix

    Issue / Review comment 可以进入可执行修复闭环。

    证明点: Issue 和 review 反馈可以变成受控、可执行的修复循环。

    +
    5 · 修复闭环阶段1 · branch-ready 修复包1 · review handoff 保留
    issue_fix_workflowreview_feedbackrepro_smokecommand_pack
     - +
    2026-06-23 · public_safe_pattern_case

    Agent to agent 回复 PR comment 和 PR Fix

    多 agent 可围绕 PR review 协同发现、评论、修复。

    证明点: 多 agent lane 可以围绕 PR comment 协作，同时保留 owner review。

    +
    1 · 每条反馈一个 owner3 · owner/fix/review 可回答问题1 · review packet handoff
    agent_to_agent_handoffpr_comment_loopreview_packetclaimed_todo
     - +
    2026-06-23 · public_safe_pattern_case

    一晚上自主重构项目

    长时间无值守重构能拆成适中 PR，而不是一个不可 review 大改。

    证明点: 无人值守 refactor 可以拆成 PR-sized slice。

    +
    1 · 一次一个 PR-sized slice3 · successor/supersede/handoff 路线1 · 宽风险 review gate
    long_unattended_goalpr_sized_slicestodo_follow_upsupersede
     - +
    2026-06-19 · public_safe_interactive_case

    外部芯片 agent workflow

    模糊目标、多 worker、长时间无值守下仍可收敛。

    证明点: 模糊目标、多 worker、长时间无值守下仍可收敛。

    +
    5 · 公开安全硬件 workflow3 · LoopX/编排/worker 角色分离1 · canonical artifact 保留
    dynamic_workflowmulti_agent_coordinationshared_control_planelong_unattended_goalCanonical 原页面
     - +
    2026-06-19 · public_evidence_case

    LoopX Meta Agent 自迭代

    高变更、多车道项目可保持状态、边界和证据。

    证明点: 高 churn 多 lane agent repo 可以保持状态、证据和边界一致。

    +
    801 · public commits244 · 06-18 后 commits59-92d · AI-assisted baseline
    self_iterationside_agent_scopetodo_claim_ownershipidentity_aware_prompt
     - +
    2026-06-17 · reproducible_synthetic_demo

    P0 block 后推进 P1/P2

    用户决策不应阻塞全部安全工作。

    证明点: 被阻塞的 P0 决策不应该阻止安全的 P1/P2 工作继续。

    +
    1 · 具体 P0 用户决策1 · safe fallback lane0 · gated lane 自动推进
    blocked_priority_fallbackconcrete_user_gatesafe_fallback_workquota_discipline
    +
    +

    Experimental today-value path

    +

    一个放在首屏下方的实验性入口：帮助用户从三个 LoopX 能力里选择今天就能产生价值的一项。

    +
    +
    + PR review/comment -> fix loop + 可复核的修复包：repro、smoke 结果、剩余 review owner。 + 更少遗漏 review 线程。 +
    + +
    + Overnight PR-sized refactor + 可 review 的 slice 列表、验证记录、后续 todo、merge 边界。 + 增加可合并 commit，避免巨型 diff。 +
    + +
    + P0 blocked -> safe fallback + 已有 goal 内由 kernel 投影具体 user gate、安全 fallback todo、quota 决策和证据边界。 + 减少 agent 空转，同时保留人类判断。 +
    +
    +
    +
    02

    附录案例

    - +
    2026-06-20 · public_safe_case_spec

    创作者-运营者长跑 Agent 案例

    研究可以继续，发布决策仍然 gated。

    证明点: 创作与运营工作可以共享一个 gate-aware 的长期 agent loop。

    +
    1 · publish hard gate1 · safe research side path0 · autopublish 动作
    creator_operator_workflowgate_aware_continuationfeedback_capturesafe_side_path
    diff --git a/docs/showcases/showcase-catalog.json b/docs/showcases/showcase-catalog.json index 960fd34a..8b82c946 100644 --- a/docs/showcases/showcase-catalog.json +++ b/docs/showcases/showcase-catalog.json @@ -81,7 +81,7 @@ }, "frontend_card": { "visual_metaphor": "parallel PR lanes converge into a reviewable merge rail", - "primary_metric_hint": "High-throughput multi-lane work can remain PR-sized, reviewable, and merge-safe.", + "primary_metric_hint": "22 reviewable merged commits, including 10 PR-numbered commits, landed inside a 10h public evidence window.", "badges": [ "PR batch", "review packet", @@ -93,7 +93,37 @@ "limit self-merge to narrow validated changes while preserving broader review gates", "record public evidence from Git history instead of raw agent logs or private screenshots" ] - } + }, + "evidence_metrics": [ + { + "value": "22", + "labels": { + "zh": "可审阅 merged commits", + "en": "reviewable merged commits" + } + }, + { + "value": "10", + "labels": { + "zh": "带 PR 编号的 commits", + "en": "PR-numbered commits" + } + }, + { + "value": "10h", + "labels": { + "zh": "公开证据窗口", + "en": "public evidence window" + } + }, + { + "value": "0", + "labels": { + "zh": "raw agent logs 依赖", + "en": "raw-agent-log dependency" + } + } + ] }, { "id": "2026-06-24-pr-issue-auto-fix", @@ -139,7 +169,7 @@ }, "frontend_card": { "visual_metaphor": "issue feedback closes through repro, fix, validation, and review handoff", - "primary_metric_hint": "Issue and review feedback can enter an executable repair loop.", + "primary_metric_hint": "Feedback becomes a five-stage repair loop with branch-ready fix evidence and reviewer handoff.", "badges": [ "issue fix", "repro smoke", @@ -151,7 +181,37 @@ "keep gated source reads explicit", "separate repro, implementation, validation, and reviewer handoff" ] - } + }, + "evidence_metrics": [ + { + "value": "5", + "labels": { + "zh": "修复闭环阶段", + "en": "repair-loop stages" + } + }, + { + "value": "1", + "labels": { + "zh": "branch-ready 修复包", + "en": "branch-ready fix packet" + } + }, + { + "value": "1", + "labels": { + "zh": "review handoff 保留", + "en": "review handoff preserved" + } + }, + { + "value": "0", + "labels": { + "zh": "公开 artifact 中 raw issue body", + "en": "raw issue bodies in public artifact" + } + } + ] }, { "id": "2026-06-23-agent-to-agent-pr-comments", @@ -197,7 +257,7 @@ }, "frontend_card": { "visual_metaphor": "agent comments pass through explicit ownership instead of loose threads", - "primary_metric_hint": "Multiple agents can coordinate around PR review comments without losing owner review.", + "primary_metric_hint": "Every PR feedback item gets an owner, fix evidence, and a visible reviewer handoff instead of becoming a loose reminder.", "badges": [ "handoff", "PR comment", @@ -209,7 +269,37 @@ "record fix and validation evidence in the review packet", "keep successor work explicit after the comment is handled" ] - } + }, + "evidence_metrics": [ + { + "value": "1", + "labels": { + "zh": "每条反馈一个 owner", + "en": "owner per feedback item" + } + }, + { + "value": "3", + "labels": { + "zh": "owner/fix/review 可回答问题", + "en": "owner/fix/review questions answered" + } + }, + { + "value": "1", + "labels": { + "zh": "review packet handoff", + "en": "review-packet handoff" + } + }, + { + "value": "0", + "labels": { + "zh": "无主 PR comment reminder", + "en": "unowned PR comment reminders" + } + } + ] }, { "id": "2026-06-23-overnight-project-refactor", @@ -255,7 +345,7 @@ }, "frontend_card": { "visual_metaphor": "a large refactor moves as small reviewable packets", - "primary_metric_hint": "Long unattended refactors can split into moderate PR slices instead of one unreviewable diff.", + "primary_metric_hint": "Unattended refactors stay reviewable by landing one PR-sized slice at a time with follow-up and review gates.", "badges": [ "refactor", "PR slices", @@ -267,7 +357,37 @@ "supersede stale tasks when the route changes", "validate each slice before merge or handoff" ] - } + }, + "evidence_metrics": [ + { + "value": "1", + "labels": { + "zh": "一次一个 PR-sized slice", + "en": "PR-sized slice at a time" + } + }, + { + "value": "3", + "labels": { + "zh": "successor/supersede/handoff 路线", + "en": "successor/supersede/handoff routes" + } + }, + { + "value": "1", + "labels": { + "zh": "宽风险 review gate", + "en": "review gate for broad risk" + } + }, + { + "value": "0", + "labels": { + "zh": "巨型 diff 目标", + "en": "giant-diff target" + } + } + ] }, { "id": "2026-06-19-dynamic-workflow-hardware-agent", @@ -302,7 +422,7 @@ "evidence_boundary": "Public-safe interactive artifact; no raw chats, screenshots, proprietary design details, private repositories, local paths, task ids, credentials, or unpublished hardware artifacts.", "frontend_card": { "visual_metaphor": "multiple worker lanes converging through one shared control plane", - "primary_metric_hint": "five public hardware-agent cases under one control plane", + "primary_metric_hint": "5 public-safe hardware workflows demonstrate multi-worker convergence while proprietary details stay out of the artifact.", "badges": [ "interactive-html", "multi-agent", @@ -325,7 +445,37 @@ "showcase_table": { "proof_point": "Fuzzy goals, multiple workers, and long unattended runs can still converge.", "loopx_intervention": "goal state, worker handoff, dynamic workflow" - } + }, + "evidence_metrics": [ + { + "value": "5", + "labels": { + "zh": "公开安全硬件 workflow", + "en": "public-safe hardware workflows" + } + }, + { + "value": "3", + "labels": { + "zh": "LoopX/编排/worker 角色分离", + "en": "LoopX/orchestrator/worker role split" + } + }, + { + "value": "1", + "labels": { + "zh": "canonical artifact 保留", + "en": "canonical artifact preserved" + } + }, + { + "value": "0", + "labels": { + "zh": "专有设计细节暴露", + "en": "proprietary design details exposed" + } + } + ] }, { "id": "2026-06-19-loopx-self-iteration", @@ -425,7 +575,7 @@ "evidence_boundary": "Public Git evidence only; no private thread text, local active-state bodies, internal document links, screenshots, raw benchmark material, credentials, or machine-specific paths.", "frontend_card": { "visual_metaphor": "many repo lanes flowing through one control plane, with a side-agent lane branching safely beside the primary benchmark lane", - "primary_metric_hint": "801 public commits mapped to a conservative 59-92 AI-assisted developer-day baseline over a 19.6-day public Git window", + "primary_metric_hint": "801 public commits, 244 recent commits, and a conservative 59-92 AI-assisted developer-day baseline show self-iteration throughput.", "badges": [ "self-iteration", "commit-backed", @@ -452,7 +602,37 @@ "showcase_table": { "proof_point": "A high-churn multi-lane project can keep state, boundaries, and evidence coherent.", "loopx_intervention": "todo, quota, gate, evidence, review packet, frontstage" - } + }, + "evidence_metrics": [ + { + "value": "801", + "labels": { + "zh": "public commits", + "en": "public commits" + } + }, + { + "value": "244", + "labels": { + "zh": "06-18 后 commits", + "en": "commits since Jun 18" + } + }, + { + "value": "59-92d", + "labels": { + "zh": "AI-assisted baseline", + "en": "AI-assisted baseline" + } + }, + { + "value": "3.0-4.7x", + "labels": { + "zh": "calendar compression", + "en": "calendar compression" + } + } + ] }, { "id": "2026-06-17-blocked-p0-safe-rotation", @@ -486,7 +666,7 @@ "evidence_boundary": "Synthetic public fixture only; no private screenshots, raw tasks, internal links, local image names, or raw run logs.", "frontend_card": { "visual_metaphor": "priority lanes with one gated lane and one active fallback lane", - "primary_metric_hint": "attention reduction: user sees one concrete decision instead of reading the whole run log", + "primary_metric_hint": "One concrete P0 user decision is isolated while one safe fallback lane can continue with zero gated-lane spend.", "badges": [ "reproducible", "user-gate", @@ -510,7 +690,37 @@ "showcase_table": { "proof_point": "A user decision should not block all safe work.", "loopx_intervention": "concrete user todo, safe fallback, quota control" - } + }, + "evidence_metrics": [ + { + "value": "1", + "labels": { + "zh": "具体 P0 用户决策", + "en": "concrete P0 user decision" + } + }, + { + "value": "1", + "labels": { + "zh": "safe fallback lane", + "en": "safe fallback lane" + } + }, + { + "value": "0", + "labels": { + "zh": "gated lane 自动推进", + "en": "gated-lane auto-progress" + } + }, + { + "value": "0", + "labels": { + "zh": "private upload 依赖", + "en": "private upload dependency" + } + } + ] }, { "id": "2026-06-20-creator-operator-case-spec", @@ -561,7 +771,37 @@ "localized_pages": { "zh": "docs/showcases/cases/0620-creator-operator-case-spec.html", "en": "docs/showcases/cases/0620-creator-operator-case-spec.en.html" - } + }, + "evidence_metrics": [ + { + "value": "1", + "labels": { + "zh": "publish hard gate", + "en": "publish hard gate" + } + }, + { + "value": "1", + "labels": { + "zh": "safe research side path", + "en": "safe research side path" + } + }, + { + "value": "0", + "labels": { + "zh": "autopublish 动作", + "en": "autopublish actions" + } + }, + { + "value": "0", + "labels": { + "zh": "真实运营数据暴露", + "en": "real operations data exposed" + } + } + ] } ] } diff --git a/examples/showcase-catalog-smoke.py b/examples/showcase-catalog-smoke.py index 3c8cadfe..82211da3 100644 --- a/examples/showcase-catalog-smoke.py +++ b/examples/showcase-catalog-smoke.py @@ -39,6 +39,18 @@ "Story " + "beats", "Website Story " + "Beats", ) +FORBIDDEN_TRIVIAL_METRIC_COPY = ( + "issue-fix 公开文件", + "issue-fix public files", + "smoke assertions", + "storyboard panels", + "source statuses", + "coverage points", + "render refs", + "handoff 回归引用", + "synthetic todos", + "todo lifecycle smokes", +) def read(path: Path) -> str: @@ -78,6 +90,13 @@ def main() -> int: assert case.get("evidence_boundary"), case assert case.get("user_value"), case assert isinstance(case.get("pattern_tags"), list) and case["pattern_tags"], case + evidence_metrics = case.get("evidence_metrics") + assert isinstance(evidence_metrics, list) and len(evidence_metrics) >= 2, case + for metric in evidence_metrics: + assert metric.get("value"), case + labels = metric.get("labels") + assert isinstance(labels, dict), case + assert labels.get("zh") and labels.get("en"), case frontend = case.get("frontend_card") appendix = case.get("appendix_surface") assert isinstance(frontend, dict) or isinstance(appendix, dict), case @@ -100,6 +119,8 @@ def main() -> int: localized_text = read(localized_path) for phrase in FORBIDDEN_SHOWCASE_COPY: assert phrase not in localized_text, f"{localized_page}: forbidden copy {phrase!r}" + for phrase in FORBIDDEN_TRIVIAL_METRIC_COPY: + assert phrase not in localized_text, f"{localized_page}: trivial metric copy {phrase!r}" if case_id != "2026-06-19-dynamic-workflow-hardware-agent" or lang == "en": assert "Repository evidence" in localized_text or "仓库证据" in localized_text, localized_page assert "Repository sources" in localized_text or "仓库来源" in localized_text, localized_page diff --git a/examples/showcase-html-pages.py b/examples/showcase-html-pages.py index b6431daf..b9e31db7 100644 --- a/examples/showcase-html-pages.py +++ b/examples/showcase-html-pages.py @@ -185,13 +185,13 @@ "zh": { "context": [ "这个案例把 PR issue、review comment 或 issue text 转成可执行修复闭环。它不是“读一段评论就改代码”，而是先做 metadata/intake、repro、branch-local patch、validation 和 review packet。", - "仓库已经有 issue-fix capability：模块在 `loopx/capabilities/issue_fix/`，CLI 入口是 `loopx issue-fix ...`，协议文档和 smoke 都在公开仓库内。", + "对用户有价值的是反馈不再停在评论区：它会进入一个有 owner、有复现、有分支修复、有验证、有 review handoff 的闭环。公开证据只证明这条闭环和边界，不把 raw issue body、私有 timeline 或本地路径带进页面。", ], "evidence": [ - ("产品入口", "`docs/capabilities/issue-fix/README.md` 声明 `loopx issue-fix ...`、content-ops bridge、protocol docs 和 smoke。"), + ("产品入口", "`docs/capabilities/issue-fix/README.md` 声明 `loopx issue-fix ...`、content-ops bridge、protocol docs 和 smoke；入口面向维护者的真实动作是把反馈转成可执行修复包。"), ("工作流协议", "`issue_fix_workflow_contract_v0` 明确 metadata preview、intake classification、workflow plan、todo writeback、caller repo branch、validation、PR review packet 和 gate handling。"), ("可执行闭环", "`issue_fix_acceptance_loop_v0` 包含 acceptance fixture、repo-branch fixture 和 caller-approved repo branch mode。"), - ("验证面", "`examples/issue-fix-workflow-plan-smoke.py`、`examples/issue-fix-workflow-contract-smoke.py`、`examples/issue-fix-acceptance-loop-smoke.py` 等 smoke 覆盖这个路径。"), + ("验证面", "focused smokes 保护 metadata preview、content-ops intake、workflow plan、workflow contract、acceptance loop 和端到端 workflow 的关键边界。"), ], "mechanism": [ "public metadata 可以进入 packet；raw issue body、comment body、timeline、provider payload 都是 gated source。", @@ -214,13 +214,13 @@ "en": { "context": [ "This case turns a PR issue, review comment, or issue text into an executable repair loop. It is not simply reading a comment and editing code; the path goes through metadata/intake, repro, branch-local patch, validation, and review packet.", - "The repository already contains the issue-fix capability: `loopx/capabilities/issue_fix/`, the `loopx issue-fix ...` CLI entry, protocol docs, and public smokes.", + "The user-facing value is that feedback no longer sits in a comment thread: it enters a loop with an owner, repro path, branch-local fix, validation, and review handoff. The public evidence proves the loop and its boundaries without exposing raw issue bodies, private timelines, or local paths.", ], "evidence": [ - ("Product entry", "`docs/capabilities/issue-fix/README.md` names `loopx issue-fix ...`, the content-ops bridge, protocol docs, and smokes."), + ("Product entry", "`docs/capabilities/issue-fix/README.md` names `loopx issue-fix ...`, the content-ops bridge, protocol docs, and smokes; the maintainer-facing action is turning feedback into an executable fix packet."), ("Workflow contract", "`issue_fix_workflow_contract_v0` defines metadata preview, intake classification, workflow plan, todo writeback, caller repo branch, validation, PR review packet, and gate handling."), ("Executable loop", "`issue_fix_acceptance_loop_v0` includes the acceptance fixture, repo-branch fixture, and caller-approved repo branch mode."), - ("Validation surface", "Smokes cover workflow planning, workflow contract, metadata/intake, and acceptance-loop behavior."), + ("Validation surface", "Focused smokes protect the key boundaries for metadata preview, content-ops intake, workflow plan, workflow contract, acceptance loop, and end-to-end workflow behavior."), ], "mechanism": [ "Public metadata can enter packets; raw issue bodies, comment bodies, timelines, and provider payloads remain gated sources.", @@ -245,13 +245,13 @@ "zh": { "context": [ "这个案例描述的是 PR review feedback 在多 agent 之间流转时如何不丢 ownership。重点不是聊天记录，而是 comment、claim、handoff、fix、validation、review packet 这条链。", - "公开仓库里可以看到相关控制面已经被产品化：`claimed_by` 出现在 todo projection、review packet、event-sourced state 和多 agent/side-agent prompt 合约里。", + "对用户有价值的是每条反馈都能回答三个问题：谁负责、修复证据在哪里、还需要谁 review。公开仓库里的 event contract、review packet、heartbeat prompt 和 validation fixtures 共同证明这条 handoff 链。", ], "evidence": [ ("todo ownership", "`event_sourced_state_contract_v0` 把 `todo_claimed` 定义为 canonical event，记录 ownership、lease 或 `claimed_by`。"), - ("review packet", "`loopx/review_packet.py` 会把 open todo 的 `claimed_by` 显示进 handoff/review 文本。"), + ("review packet", "`loopx/review_packet.py` 在 open-todo rendering 和 handoff ranking 路径中保留 `claimed_by`，让 review packet 能显示 owner。"), ("side-agent contract", "`docs/heartbeat-automation-prompt.md` 规定 side-agent 小变更可带 evidence 自合并，否则创建 claimed-by handoff todo。"), - ("CLI smokes", "`examples/todo-lifecycle-cli-smoke.py`、`examples/todo-cli-smoke.py` 覆盖 claim、handoff successor、side-agent self-merge 和 review handoff 规则。"), + ("CLI smokes", "`examples/todo-lifecycle-cli-smoke.py` 和 `examples/todo-cli-smoke.py` 覆盖 claim、handoff successor、side-agent self-merge、review handoff 和 same-agent handoff rejection。"), ], "mechanism": [ "PR feedback 先成为一个有 owner 的 todo，而不是 chat reminder。", @@ -273,13 +273,13 @@ "en": { "context": [ "This case shows how PR review feedback can move across multiple agents without losing ownership. The important chain is comment, claim, handoff, fix, validation, and review packet rather than the chat transcript.", - "The public repository contains the control-plane pieces: `claimed_by` appears in todo projection, review packets, event-sourced state, and multi-agent or side-agent prompt contracts.", + "The user-facing value is that every feedback item can answer three questions: who owns it, where the fix evidence is, and who still needs to review it. Public evidence spans the event contract, review packet, heartbeat prompt, and validation fixtures.", ], "evidence": [ ("Todo ownership", "`event_sourced_state_contract_v0` defines `todo_claimed` as a canonical event for ownership, lease, or `claimed_by`."), - ("Review packet", "`loopx/review_packet.py` renders open todo `claimed_by` values into handoff and review text."), + ("Review packet", "`loopx/review_packet.py` preserves `claimed_by` in open-todo rendering and handoff ranking so review packets can show ownership."), ("Side-agent contract", "`docs/heartbeat-automation-prompt.md` requires side agents to self-merge only small validated evidence-backed work or create a claimed handoff todo."), - ("CLI smokes", "`examples/todo-lifecycle-cli-smoke.py` and `examples/todo-cli-smoke.py` cover claim, handoff successor, side-agent self-merge, and review handoff rules."), + ("CLI smokes", "`examples/todo-lifecycle-cli-smoke.py` and `examples/todo-cli-smoke.py` cover claim, handoff successor, side-agent self-merge, review handoff, and same-agent handoff rejection."), ], "mechanism": [ "PR feedback becomes an owned todo rather than a chat reminder.", @@ -303,12 +303,12 @@ "zh": { "context": [ "这个案例解决的是无人值守重构的风险：长时间 agent 容易把 cleanup、行为改变、发现的新问题和过期计划混成一个大 diff。", - "LoopX 的公开证据不是某个私有夜间截图，而是 todo lifecycle、successor/supersede、validation writeback 和 review-packet 这些已经在仓库里有文档和 smoke 的控制面。", + "LoopX 的公开证据不是某个私有夜间截图，而是 todo lifecycle、successor/supersede、validation writeback 和 review-packet 这些已经在仓库里有文档和 smoke 的控制面。`todo-lifecycle-cli-smoke.py` 里能看到 successor、supersede、handoff 和 self-merge 的完整回归覆盖。", ], "evidence": [ ("successor path", "`docs/lark-kanban-control-plane-adapter.md` 明确 real successor 使用 `todo complete --next-*`，replacement 或 narrower split 使用 `todo supersede --next-agent-todo`。"), ("side-agent completion", "`docs/heartbeat-automation-prompt.md` 要求非平凡完成创建 successor todo 或写 no-follow-up rationale。"), - ("CLI validation", "`examples/todo-lifecycle-cli-smoke.py` 覆盖 `--next-agent-todo`、`todo supersede`、claim 继承和 handoff successor。"), + ("CLI validation", "`examples/todo-lifecycle-cli-smoke.py` 覆盖 `--next-agent-todo` successor、`todo supersede`、claim 继承、handoff successor、same-agent handoff rejection 和 side-agent self-merge evidence。"), ("review shape", "`loopx review-packet` 把当前 open todo、claimed_by 和 handoff 状态打包成 reviewer 可读的 packet。"), ], "mechanism": [ @@ -331,12 +331,12 @@ "en": { "context": [ "This case addresses the risk of unattended refactoring: a long-running agent can mix cleanup, behavior change, discoveries, and stale plans into one broad diff.", - "The public evidence is not a private overnight screenshot. It is the control-plane behavior already documented and smoke-tested in the repository: todo lifecycle, successor/supersede, validation writeback, and review packets.", + "The public evidence is not a private overnight screenshot. It is the control-plane behavior already documented and smoke-tested in the repository: todo lifecycle, successor/supersede, validation writeback, and review packets. `todo-lifecycle-cli-smoke.py` carries regression coverage for successors, supersede, handoff, and self-merge.", ], "evidence": [ ("Successor path", "`docs/lark-kanban-control-plane-adapter.md` says real successors use `todo complete --next-*`, while replacements or narrower splits use `todo supersede --next-agent-todo`."), ("Side-agent completion", "`docs/heartbeat-automation-prompt.md` requires nontrivial completion to create a successor todo or a no-follow-up rationale."), - ("CLI validation", "`examples/todo-lifecycle-cli-smoke.py` covers `--next-agent-todo`, `todo supersede`, claim inheritance, and handoff successors."), + ("CLI validation", "`examples/todo-lifecycle-cli-smoke.py` covers `--next-agent-todo` successors, `todo supersede`, claim inheritance, handoff successors, same-agent handoff rejection, and side-agent self-merge evidence."), ("Review shape", "`loopx review-packet` packages open todos, claimed_by, and handoff state for reviewer consumption."), ], "mechanism": [ @@ -454,11 +454,11 @@ "zh": { "context": [ "这个案例展示 P0 被用户决策卡住时，系统不应该继续硬跑，也不应该让整个目标停摆。原场景是 benchmark rotation：一个 lane 需要大型本地 image，其他 no-upload benchmark work 仍然安全。", - "公开仓库没有暴露原始 benchmark task 或本地 image 名，而是用 synthetic smoke 复现控制面行为。", + "公开仓库没有暴露原始 benchmark task 或本地 image 名，而是用 synthetic smoke 复现控制面行为。用户价值是明确看到一个需要决策的 P0，同时安全 fallback 可以继续，且 gated lane 不消耗额外自动推进预算。", ], "evidence": [ - ("synthetic fixture", "`examples/showcase-0617-blocked-p0-safe-rotation-smoke.py` 构造 P0 user gate、被 gate 阻塞的 P0 agent todo 和 P1 no-upload fallback。"), - ("quota contract", "smoke 断言 `should_run=True`、`requires_user_action=True`、`safe_bypass_allowed=True`、`safe_bypass_kind=scoped_user_gate_fallback`。"), + ("synthetic fixture", "`examples/showcase-0617-blocked-p0-safe-rotation-smoke.py` 复现 P0 user gate、被 gate 阻塞的 P0 agent lane 和 P1 no-upload fallback。"), + ("quota contract", "smoke 固定 `should_run=True`、`requires_user_action=True`、`safe_bypass_allowed=True`、`safe_bypass_kind=scoped_user_gate_fallback` 等关键 contract。"), ("selected fallback", "fixture 选择 `terminal_bench_no_upload`，同时保留 `ale_image` gate 的 user-visible blocker。"), ("rendered evidence", "smoke 检查 markdown 中包含 `scoped_user_gate_fallback` 和 safe no-upload Terminal-Bench rotation。"), ], @@ -479,11 +479,11 @@ "en": { "context": [ "This case shows what should happen when a P0 route is blocked by a user decision: the system should neither keep forcing that lane nor stop the whole goal. The original shape was a benchmark rotation where one lane needed a large local image while other no-upload benchmark work remained safe.", - "The public repository does not expose raw benchmark tasks or local image names. It reproduces the control-plane behavior with a synthetic smoke.", + "The public repository does not expose raw benchmark tasks or local image names. It reproduces the control-plane behavior with a synthetic smoke. The user-facing value is seeing one concrete P0 decision while safe fallback work can continue and the gated lane does not burn automated progress budget.", ], "evidence": [ - ("Synthetic fixture", "`examples/showcase-0617-blocked-p0-safe-rotation-smoke.py` creates a P0 user gate, a P0 agent todo blocked by that gate, and a P1 no-upload fallback."), - ("Quota contract", "The smoke asserts `should_run=True`, `requires_user_action=True`, `safe_bypass_allowed=True`, and `safe_bypass_kind=scoped_user_gate_fallback`."), + ("Synthetic fixture", "`examples/showcase-0617-blocked-p0-safe-rotation-smoke.py` reproduces a P0 user gate, a P0 agent lane blocked by that gate, and a P1 no-upload fallback."), + ("Quota contract", "The smoke pins `should_run=True`, `requires_user_action=True`, `safe_bypass_allowed=True`, `safe_bypass_kind=scoped_user_gate_fallback`, and related fallback evidence."), ("Selected fallback", "The fixture selects `terminal_bench_no_upload` while preserving the `ale_image` gate as the user-visible blocker."), ("Rendered evidence", "The smoke checks markdown for `scoped_user_gate_fallback` and safe no-upload Terminal-Bench rotation."), ], @@ -506,12 +506,12 @@ "zh": { "context": [ "这是 appendix case：它展示非技术创作者/运营者如何用 LoopX 管一个长期 research + planning loop，但还不是 top-card proof，因为没有真实用户公开证据。", - "公开材料完全使用 synthetic data，重点是产品形态：趋势候选、偏好映射、insight board、draft queue、material library、人类反馈和 controlled replan。", + "公开材料完全使用 synthetic data，重点不是证明增长，而是证明产品边界：研究和素材整理可以继续，发布、品牌判断和对外动作必须停在人类 gate。", ], "evidence": [ - ("storyboard", "`creator-ops-fake-data-storyboard.md` 定义七个面板和完整 fake fixture，不需要 live platform access。"), - ("feedback contract", "`creator-ops-feedback-boundary-contract.md` 把 gate decision、preference hint、todo update、boundary correction、reward signal、product improvement note 分开。"), - ("source status", "contract 要求 topic、insight、draft、material item 都有 source status，public repo 默认 `synthetic_demo`。"), + ("storyboard", "`creator-ops-fake-data-storyboard.md` 给出完整 fake fixture，不需要 live platform access，适合公开展示用户旅程而不泄漏真实运营数据。"), + ("feedback contract", "`creator-ops-feedback-boundary-contract.md` 把 gate decision、preference hint、todo update、boundary correction、reward signal 和 product improvement note 分开，避免把偏好误当发布授权。"), + ("source status", "contract 要求 topic、insight、draft、material item 都携带 source status；public repo 默认 `synthetic_demo`，不伪装成真实增长证据。"), ("no autopublish", "publishing 是 hard user gate；safe side work 可以继续，但不能把偏好或 reward 当成发布授权。"), ], "mechanism": [ @@ -533,12 +533,12 @@ "en": { "context": [ "This is an appendix case: it shows how a non-technical creator-operator might use LoopX to manage a long-running research and planning loop, but it is not a top-card proof until real public user evidence exists.", - "The public material uses only synthetic data. The product shape is trend candidates, preference map, insight board, draft queue, material library, human feedback, and controlled replan.", + "The public material uses only synthetic data. The point is not proving growth; it is proving the product boundary: research and material organization can continue, while publishing, brand judgment, and external action remain human gates.", ], "evidence": [ - ("Storyboard", "`creator-ops-fake-data-storyboard.md` defines seven panels and a full fake fixture with no live platform access."), - ("Feedback contract", "`creator-ops-feedback-boundary-contract.md` separates gate decision, preference hint, todo update, boundary correction, reward signal, and product improvement note."), - ("Source status", "The contract requires every topic, insight, draft, and material item to carry source status; public repo defaults to `synthetic_demo`."), + ("Storyboard", "`creator-ops-fake-data-storyboard.md` provides a complete fake fixture with no live platform access, suitable for showing the user journey without exposing real operations data."), + ("Feedback contract", "`creator-ops-feedback-boundary-contract.md` separates gate decisions, preference hints, todo updates, boundary corrections, reward signals, and product improvement notes so preference is not mistaken for publish approval."), + ("Source status", "The contract requires every topic, insight, draft, and material item to carry source status; the public repo defaults to `synthetic_demo` instead of pretending to show real growth evidence."), ("No autopublish", "Publishing is a hard user gate; safe side work may continue, but preference or reward is not publication approval."), ], "mechanism": [ @@ -583,6 +583,13 @@ "open": "Open", "demo": "Demo", "search": "Search showcase cases", + "experimental_title": "Experimental today-value path", + "experimental_intro": "A lower-priority entry point for users who want to pick one useful LoopX capability today without replacing the showcase first screen.", + "experimental_rows": [ + ("PR review/comment -> fix loop", "Branch-ready fix packet with repro, smoke result, and remaining review owner.", "Fewer dropped review threads."), + ("Overnight PR-sized refactor", "Reviewable slice list, validation notes, successor todo, and merge boundary.", "More merged commits without a giant diff audit."), + ("P0 blocked -> safe fallback", "Kernel projection of the exact user gate, safe fallback todo, quota decision, and evidence boundary.", "Less idle agent time while preserving human judgment."), + ], "footer": "Generated from docs/showcases/showcase-catalog.json. Private links, raw chats, local state, and internal media are excluded.", }, "zh": { @@ -607,6 +614,13 @@ "open": "打开", "demo": "Demo", "search": "搜索 showcase 案例", + "experimental_title": "Experimental today-value path", + "experimental_intro": "一个放在首屏下方的实验性入口：帮助用户从三个 LoopX 能力里选择今天就能产生价值的一项。", + "experimental_rows": [ + ("PR review/comment -> fix loop", "可复核的修复包：repro、smoke 结果、剩余 review owner。", "更少遗漏 review 线程。"), + ("Overnight PR-sized refactor", "可 review 的 slice 列表、验证记录、后续 todo、merge 边界。", "增加可合并 commit，避免巨型 diff。"), + ("P0 blocked -> safe fallback", "已有 goal 内由 kernel 投影具体 user gate、安全 fallback todo、quota 决策和证据边界。", "减少 agent 空转，同时保留人类判断。"), + ], "footer": "由 docs/showcases/showcase-catalog.json 生成。不包含私有链接、原始聊天、本地状态或内部媒体。", }, } @@ -816,6 +830,8 @@ def css() -> str: .metric{border:1px solid rgba(255,255,255,.1);border-radius:10px;padding:17px 19px;background:#0c0c0e} .metric strong{display:block;font-size:24px;letter-spacing:-.02em;color:#fafafa} .metric span{display:block;margin-top:5px;font-family:'Geist Mono',ui-monospace,monospace;font-size:10.5px;color:#62666d;text-transform:uppercase} + .mini-metrics{display:flex;flex-wrap:wrap;gap:8px;margin-top:13px} + .mini-metrics span{font-family:'Geist Mono',ui-monospace,monospace;font-size:10.5px;color:#aeb3ba;border:1px solid rgba(255,255,255,.1);border-radius:5px;padding:4px 8px;background:#0b0b0c} .search{width:100%;height:42px;margin:22px 0 16px;border:1px solid rgba(255,255,255,.12);border-radius:8px;background:#0e0e10;color:#f1f2f3;padding:0 12px;font:14px 'Geist',system-ui,sans-serif} .cards{display:flex;flex-direction:column;gap:12px} .card{display:block;text-decoration:none;border:1px solid rgba(255,255,255,.1);border-radius:10px;background:#0e0e10;padding:18px 20px} @@ -823,8 +839,16 @@ def css() -> str: .card .meta{font-family:'Geist Mono',ui-monospace,monospace;font-size:10.5px;color:#62666d;margin-bottom:8px} .card p{font-size:14px;margin-top:9px} .hide{display:none} + .experiment{margin-top:24px;border:1px solid rgba(255,255,255,.1);border-radius:10px;background:#0e0e10;padding:20px} + .experiment h3{font-size:18px} + .experiment p{font-size:14px;margin-top:8px} + .experiment-grid{display:grid;grid-template-columns:repeat(3,minmax(0,1fr));gap:12px;margin-top:16px} + .experiment-card{border:1px solid rgba(255,255,255,.1);border-radius:8px;background:#0b0b0c;padding:14px} + .experiment-card strong{display:block;font-size:14px;line-height:1.45} + .experiment-card span{display:block;margin-top:9px;font-size:13px;line-height:1.55;color:#9ea3aa} + .experiment-card em{display:block;margin-top:10px;font-style:normal;font-family:'Geist Mono',ui-monospace,monospace;font-size:10.5px;line-height:1.55;color:var(--accent)} footer{margin-top:76px;color:#62666d;font-family:'Geist Mono',ui-monospace,monospace;font-size:10.5px;line-height:1.7} - @media(max-width:720px){article{padding:52px 18px 90px}.panel-row{grid-template-columns:1fr}.panel-key{border-right:0;border-bottom:1px solid rgba(255,255,255,.07)}.metric-grid,.evidence-grid{grid-template-columns:1fr}.source-ref{grid-template-columns:1fr;gap:6px}} + @media(max-width:720px){article{padding:52px 18px 90px}.panel-row{grid-template-columns:1fr}.panel-key{border-right:0;border-bottom:1px solid rgba(255,255,255,.07)}.metric-grid,.evidence-grid,.experiment-grid{grid-template-columns:1fr}.source-ref{grid-template-columns:1fr;gap:6px}} """ @@ -880,11 +904,37 @@ def control_diagram(case: dict[str, Any], lang: str) -> str: """ -def metrics(case: dict[str, Any]) -> str: +def evidence_metric_cards(case: dict[str, Any], lang: str) -> list[tuple[str, str]]: + raw_metrics = case.get("evidence_metrics") + cards: list[tuple[str, str]] = [] + if not isinstance(raw_metrics, list): + return cards + for item in raw_metrics: + if not isinstance(item, dict): + continue + value = item.get("value") + labels = item.get("labels") + label = "" + if isinstance(labels, dict): + label = str(labels.get(lang) or labels.get("en") or "") + if not label: + label = str(item.get("label") or "") + if value is None or not label: + continue + cards.append((str(value), label)) + return cards + + +def metrics(case: dict[str, Any], lang: str) -> str: + cards = evidence_metric_cards(case, lang) + if cards: + return '
    ' + "".join( + f'
    {esc(value)}{esc(label)}
    ' + for value, label in cards[:4] + ) + "
    " workload = case.get("workload_signal") if not isinstance(workload, dict): return "" - cards: list[tuple[str, str]] = [] public_git = workload.get("public_git") if isinstance(workload.get("public_git"), dict) else {} whole = workload.get("whole_repository") if isinstance(workload.get("whole_repository"), dict) else {} window = workload.get("window") if isinstance(workload.get("window"), dict) else {} @@ -892,15 +942,15 @@ def metrics(case: dict[str, Any]) -> str: estimated = model.get("estimated_developer_days") if isinstance(model.get("estimated_developer_days"), dict) else {} compression = model.get("single_engineer_calendar_compression") if isinstance(model.get("single_engineer_calendar_compression"), dict) else {} if public_git.get("merged_commits") is not None: - cards.append((str(public_git["merged_commits"]), "merged PR commits")) + cards.append((str(public_git["merged_commits"]), "merged PR commits" if lang == "en" else "merged PR commits")) if window.get("hours") is not None: - cards.append((f"{window['hours']}h", "public window")) + cards.append((f"{window['hours']}h", "public window" if lang == "en" else "public window")) if whole.get("commit_count") is not None: - cards.append((str(whole["commit_count"]), "public commits")) + cards.append((str(whole["commit_count"]), "public commits" if lang == "en" else "public commits")) if estimated.get("low") and estimated.get("high"): - cards.append((f"{estimated['low']}-{estimated['high']}d", "AI-assisted baseline")) + cards.append((f"{estimated['low']}-{estimated['high']}d", "AI-assisted baseline" if lang == "en" else "AI-assisted baseline")) if compression.get("low") and compression.get("high"): - cards.append((f"{compression['low']}-{compression['high']}x", "calendar compression")) + cards.append((f"{compression['low']}-{compression['high']}x", "calendar compression" if lang == "en" else "calendar compression")) if not cards: return "" return '
    ' + "".join( @@ -961,7 +1011,7 @@ def render_case_page(case: dict[str, Any], lang: str, primary: bool) -> str:
    {esc(ui(lang, "proof"))}

    {esc(table["proof_point"])}

    {esc(ui(lang, "intervention"))}

    {esc(table["loopx_intervention"])}

    - {metrics(case)} + {metrics(case, lang)} {evidence_block}
    03

    {esc(ui(lang, "behavior"))}

    @@ -989,19 +1039,46 @@ def index_card(case: dict[str, Any], current: Path, lang: str) -> str: headline = localized(case, lang, "headline") table = table_for(case, lang) tags = first_items(case.get("pattern_tags"), 4) - search = " ".join([title, headline, table["proof_point"], table["loopx_intervention"], *tags]).lower() + metric_cards = evidence_metric_cards(case, lang) + metric_terms = [f"{value} {label}" for value, label in metric_cards] + search = " ".join([title, headline, table["proof_point"], table["loopx_intervention"], *tags, *metric_terms]).lower() canonical = f'{esc(ui(lang, "canonical"))}' if str(case.get("id")) == HARDWARE_CASE_ID and lang == "zh" else "" + metric_line = "" + if metric_cards: + metric_line = '
    ' + "".join( + f"{esc(value)} · {esc(label)}" for value, label in metric_cards[:3] + ) + "
    " return f"""
    {esc(case.get("date") or "")} · {esc(case.get("status") or "")}

    {esc(title)}

    {esc(headline)}

    {esc(ui(lang, "proof"))}: {esc(table["proof_point"])}

    + {metric_line}
    {badges(tags)}{canonical}
     """ +def experimental_lane(lang: str) -> str: + cards = "\n".join( + f""" +
    + {esc(title)} + {esc(output)} + {esc(value)} +
    """ + for title, output, value in ui(lang, "experimental_rows") + ) + return f""" +
    +

    {esc(ui(lang, "experimental_title"))}

    +

    {esc(ui(lang, "experimental_intro"))}

    +
    {cards} +
    +
    """ + + def render_index(cases: list[dict[str, Any]], lang: str) -> str: primary, appendix = ordered_cases(cases) current = index_path(lang) @@ -1025,6 +1102,7 @@ def render_index(cases: list[dict[str, Any]], lang: str) -> str:
    01

    {esc(ui(lang, "top_cases"))}

    {primary_cards}
    + {experimental_lane(lang)}
    02

    {esc(ui(lang, "appendix"))}

    {appendix_cards}