feat(issues): org-wide issue forms with triage automation (issue-template-kit v1.0.0)

.github/ISSUE_TEMPLATE/1-bug.yml

@@ -0,0 +1,105 @@
+name: 🐛 Bug
+description: Something behaves incorrectly and you can show it.
+title: "bug: "
+labels: ["type:bug", "needs:triage"]
+body:
+  - type: markdown
+    attributes:
+      value: |
+        Lead with the error. A worked example lives in
+        [EXAMPLE.md](../blob/HEAD/.github/ISSUE_TEMPLATE/EXAMPLE.md).
+
+  - type: textarea
+    id: problem
+    attributes:
+      label: Problem
+      description: The symptom you observed, in one or two sentences. Not your theory.
+      placeholder: Agent runs abort at ~70% of the context limit with BudgetExceeded.
+    validations:
+      required: true
+
+  - type: textarea
+    id: evidence
+    attributes:
+      label: Error output
+      description: Full traceback, failing assertion, alert, or log line. Redact secrets.
+      render: shell
+    validations:
+      required: true
+
+  - type: textarea
+    id: reproduce
+    attributes:
+      label: Reproduction
+      description: Exact commands from a clean checkout. A failing test is ideal.
+      value: |
+        1. `git clone … && cd …`
+        2. `uv sync`
+        3. `pytest tests/test_x.py::test_y`
+      render: shell
+    validations:
+      required: true
+
+  - type: input
+    id: expected
+    attributes:
+      label: Expected behavior
+      placeholder: Budget reports the actual measured token count.
+    validations:
+      required: true
+
+  - type: input
+    id: version
+    attributes:
+      label: Version / commit
+      description: Release tag or full commit SHA. "latest" is not a version.
+      placeholder: v0.9.3 or 4f2a1c9
+    validations:
+      required: true
+
+  - type: dropdown
+    id: severity
+    attributes:
+      label: Severity
+      description: This routes triage. Be honest; inflation gets ignored.
+      options:
+        - S4 — cosmetic or minor annoyance
+        - S3 — degraded, workaround exists
+        - S2 — major function broken, no workaround
+        - S1 — production down, data loss, or security exposure
+    validations:
+      required: true
+
+  - type: dropdown
+    id: env
+    attributes:
+      label: Environment
+      multiple: true
+      options: [local (macOS), local (Linux), CI (GitHub Actions), dev, staging, prod]
+    validations:
+      required: true
+
+  - type: input
+    id: regression
+    attributes:
+      label: Last known good version
+      description: Leave blank if it never worked. If set, a bisect range is implied.
+      placeholder: v0.9.1
+
+  - type: textarea
+    id: context
+    attributes:
+      label: Anything else
+      description: Suspected cause, related issues, screenshots, workaround in use.
+
+  - type: checkboxes
+    id: gates
+    attributes:
+      label: Before submitting
+      options:
+        - label: I searched open and closed issues for a duplicate.
+          required: true
+        - label: I reproduced this on the latest release or main.
+          required: true
+        - label: No secrets, tokens, or customer data appear anywhere above.
+          required: true

.github/ISSUE_TEMPLATE/2-feature.yml

@@ -0,0 +1,79 @@
+name: ✨ Feature
+description: Propose a capability that does not exist yet.
+title: "feat: "
+labels: ["type:feature", "needs:triage"]
+body:
+  - type: textarea
+    id: problem
+    attributes:
+      label: Problem
+      description: The user-facing pain, not the solution. Who is blocked, and on what?
+      placeholder: Operators cannot tell which agent consumed a token budget, so cost overruns are diagnosed by guesswork.
+    validations:
+      required: true
+
+  - type: textarea
+    id: today
+    attributes:
+      label: What you do today
+      description: The current workaround and its cost in time, money, or risk.
+    validations:
+      required: true
+
+  - type: textarea
+    id: proposal
+    attributes:
+      label: Proposed behavior
+      description: What should exist. Include the interface — CLI flag, API shape, config key.
+      render: shell
+    validations:
+      required: true
+
+  - type: textarea
+    id: alternatives
+    attributes:
+      label: Alternatives considered
+      description: Options you rejected and why. "None" usually means the problem is underexplored.
+    validations:
+      required: true
+
+  - type: textarea
+    id: done
+    attributes:
+      label: Definition of done
+      description: Observable, checkable outcomes. This becomes the acceptance criteria.
+      value: |
+        - [ ]
+        - [ ]
+    validations:
+      required: true
+
+  - type: dropdown
+    id: scope
+    attributes:
+      label: Scope
+      options:
+        - S — under a day, one file or module
+        - M — a few days, one repo
+        - L — multi-repo or migration required
+        - XL — needs a design doc first
+    validations:
+      required: true
+
+  - type: dropdown
+    id: breaking
+    attributes:
+      label: Breaking change?
+      options: ["No", "Yes — behind a flag", "Yes — requires a major bump"]
+    validations:
+      required: true
+
+  - type: checkboxes
+    id: gates
+    attributes:
+      label: Before submitting
+      options:
+        - label: I searched for an existing request covering this.
+          required: true
+        - label: This is a concrete capability, not an open-ended question (those go to Discussions).
+          required: true

.github/ISSUE_TEMPLATE/3-task.yml

@@ -0,0 +1,45 @@
+name: 🔧 Task
+description: Known work with a known solution — chore, refactor, infra, docs.
+title: "chore: "
+labels: ["type:task"]
+body:
+  - type: textarea
+    id: problem
+    attributes:
+      label: Problem
+      description: Why this work exists. What breaks or degrades if it is never done?
+    validations:
+      required: true
+
+  - type: textarea
+    id: work
+    attributes:
+      label: Work
+      description: The steps. Each line should be independently verifiable.
+      value: |
+        - [ ]
+        - [ ]
+    validations:
+      required: true
+
+  - type: dropdown
+    id: kind
+    attributes:
+      label: Kind
+      options: [chore, refactor, infra / IaC, CI / CD, docs, dependency bump, test coverage]
+    validations:
+      required: true
+
+  - type: textarea
+    id: risk
+    attributes:
+      label: Risk and rollback
+      description: Blast radius if this goes wrong, and how to undo it.
+    validations:
+      required: true
+
+  - type: input
+    id: blocks
+    attributes:
+      label: Blocks / blocked by
+      placeholder: "blocked by #1191, blocks #1204"

.github/ISSUE_TEMPLATE/4-incident.yml

@@ -0,0 +1,90 @@
+name: 🚨 Incident
+description: Production is degraded or down. File first, investigate second.
+title: "incident: "
+labels: ["type:incident", "sev:untriaged", "priority:P0"]
+assignees: []
+body:
+  - type: markdown
+    attributes:
+      value: |
+        **Page first, file second.** If customers are affected and nobody is paged,
+        stop and page on-call. This issue is the written record, not the alarm.
+        If this is a security compromise, use a private advisory instead — not this form.
+
+  - type: input
+    id: started
+    attributes:
+      label: Detected at (UTC)
+      placeholder: "2026-07-28T15:04:00Z"
+    validations:
+      required: true
+
+  - type: textarea
+    id: problem
+    attributes:
+      label: Problem
+      description: Customer-visible symptom. What is broken, for whom, since when?
+    validations:
+      required: true
+
+  - type: dropdown
+    id: severity
+    attributes:
+      label: Severity
+      options:
+        - S1 — full outage, data loss, or security exposure
+        - S2 — major degradation, no workaround
+        - S3 — partial degradation, workaround in place
+    validations:
+      required: true
+
+  - type: textarea
+    id: evidence
+    attributes:
+      label: Signal
+      description: Alert name, dashboard link, error rate, sample traceback.
+      render: shell
+    validations:
+      required: true
+
+  - type: textarea
+    id: impact
+    attributes:
+      label: Blast radius
+      description: Services, regions, tenants, and estimated request or user count.
+    validations:
+      required: true
+
+  - type: textarea
+    id: mitigation
+    attributes:
+      label: Mitigation in progress
+      description: What is being done right now. Update this field as it evolves.
+    validations:
+      required: true
+
+  - type: input
+    id: suspect
+    attributes:
+      label: Suspected trigger
+      description: Deploy, PR, config change, or upstream provider. Link it.
+      placeholder: "#1184, deployed 14:52Z"
+
+  - type: input
+    id: commander
+    attributes:
+      label: Incident commander
+      placeholder: "@handle"
+    validations:
+      required: true
+
+  - type: checkboxes
+    id: gates
+    attributes:
+      label: Response checklist
+      options:
+        - label: On-call paged.
+          required: true
+        - label: Status page or stakeholder comms updated, or explicitly not needed.
+        - label: Timeline is being kept in the comments below.
+        - label: Postmortem issue will be opened within 48h of resolution.

.github/ISSUE_TEMPLATE/EXAMPLE.md

@@ -0,0 +1,54 @@
+<!-- WORKED EXAMPLE — the standard for a bug report in this org.
+     This is what the 🐛 Bug form looks like when filled in well. Reference it. -->
+
+# bug: token budget aborts runs at ~70% of the real context limit
+
+**Problem**
+Long agent runs abort well before the context limit. Operators see `BudgetExceeded`
+at roughly 122k actual tokens against a 200k limit, so ~40% of usable context is
+unreachable.
+
+**Error output**
+```
+agentkit.budget.BudgetExceeded: 198,004 / 200,000 tokens at turn 42
+  at agentkit/budget.py:88 in BudgetTracker.check
+  (independently measured context via tiktoken: 122,311 tokens)
+```
+
+**Reproduction**
+```
+1. git clone git@github.com:acme/agentkit && cd agentkit && git checkout v0.9.3
+2. uv sync
+3. pytest tests/replay/test_long_run.py::test_42_turns  -> fails at turn 42
+```
+
+**Expected behavior**
+The tracker reports the actual measured token count, so runs continue to the real limit.
+
+**Version / commit** — v0.9.3 (`4f2a1c9`)
+**Severity** — S2 — major function broken, no workaround
+**Environment** — local (macOS), CI (GitHub Actions), prod
+**Last known good version** — v0.8.7, so the regression is in the 0.9.0 reducer rewrite
+
+**Anything else**
+Reported usage is consistently ~1.45x measured, and the ratio tracks the count of
+tool-result messages — suggesting those are counted twice. `BudgetTracker.add()`
+appears to be called from both `transport.py:212` and `reducer.py:96`. Workaround in
+use: `AGENTKIT_BUDGET_LIMIT=290000` on the four affected agents, which is unsafe
+because the real ceiling is then unenforced.
+
+**Before submitting** — searched (#902 is related but closed as fixed), reproduced on
+main at `4f2a1c9`, trace excerpts scrubbed by `scripts/scrub_trace.py`.
+
+---
+
+## Why this is a good report
+
+- The **problem** is the symptom an operator saw, not a theory about the cause.
+- The traceback is complete, and it includes an **independent measurement** that
+  proves the number is wrong rather than merely surprising.
+- Reproduction starts from a clean clone at a pinned SHA.
+- **Last known good** turns a vague bug into a bisect range.
+- The theory is present but quarantined in "anything else", below the facts.
+- The workaround is stated **along with why it is unsafe**, which is what makes this
+  urgent rather than merely annoying.

.github/ISSUE_TEMPLATE/config.yml

@@ -0,0 +1,11 @@
+blank_issues_enabled: false
+contact_links:
+  - name: 🔐 Report a security vulnerability
+    url: https://github.com/Quantum-L9/.github/security/advisories/new
+    about: Do NOT open a public issue. Use a private security advisory.
+  - name: 💬 Question or design discussion
+    url: https://github.com/orgs/Quantum-L9/discussions
+    about: Open-ended questions, architecture debate, and "how do I…" belong in Discussions.
+  - name: 📕 Documentation
+    url: https://github.com/Quantum-L9/Cursor-Governance/blob/main/CANONICAL_LAW.md
+    about: Check the runbooks and reference before filing.