fix(web): simplify Context Window and Session State settings hierarchy

[mindfn]

PR Type

• 🐛 Patch — Bug fix, typo, test gap (no Feature Doc needed)
• ✨ Feature — New capability or behavior change (requires Feature Doc)
• 📋 Protocol — Rules, skills, workflow changes (the doc IS the contribution)

Related Issue

Closes #1343

What

• Keep the normal member editor's Context Window surface to the Auto/Manual input and one concise explanation.
• Hide resolved catalog/provenance/carrier diagnostics; show only an actionable client/window incompatibility warning when it affects the choice.
• Render Session State / Chain as the section title with the existing Session Strategy fields directly underneath.
• Remove applied-policy strategy/execution projections from active Runtime Session Chain cards and sealed-session summaries while retaining meaningful session facts.
• Add regression coverage for the compact hierarchy, Auto and Manual warning branches, strategy field ordering, and both runtime-panel projections.

Why

The settings page and Runtime Session Chain exposed internal policy diagnostics as if they were normal operator choices. This patch restores the intended information hierarchy while preserving the runtime semantics delivered by #1208/#1209 and #1329/#1334.

Tradeoff

Runtime-derived context values, capability provenance, and applied-policy strategy/execution are no longer projected in normal settings/session cards. They remain available to the underlying runtime/config flow; the UI now surfaces only incompatibility information that gives the operator a concrete action.

Maintainer Ownership Gate

• Q1 PASS — aligns with the accepted issue and makes member configuration clearer without changing Clowder AI's collaboration direction.
• Q2 PASS — a presentation-only correction on top of Bug: 成员 Context Limit 未统一持久化，自动 handoff 可越过真实窗口 #1208/fix: unify member context capacity and lifecycle (#1208) #1209 and Session Strategy: decouple always-visible session state, policy intent, and client capabilities #1329/feat(F033): decouple session strategy from runtime capability #1334; no competing runtime or config path is introduced.
• Q3 PASS — issue fix(web): simplify Context Window and Session State / Chain settings UX #1343 is triaged and accepted by the maintainer.
• Q4 PASS — React/UI and Vitest changes only; no dependency, language, runtime, API, or storage addition.
• Q5 PASS — low, owned debt: five existing web files, regression tests included, two-commit rollback, no build/release/security/data/support-matrix expansion.

Test Evidence

RED: targeted hub-cat-editor suite — 78 tests, 73 passed / 5 expected regression failures
GREEN: pnpm --dir packages/web exec vitest run src/components/__tests__/hub-cat-editor.test.tsx — 78 passed
RED: targeted SessionChainPanel suite — 57 tests, 56 passed / 1 expected policy-row regression failure
GREEN: pnpm --dir packages/web exec vitest run src/components/__tests__/session-chain-panel.test.ts — 57 passed
pnpm --dir packages/web exec tsc --noEmit — passed
env -u NODE_ENV pnpm gate @ c77ee06e2d0ce80d310e023a740649226e9a1475 — GATE PASSED (exit 0; rebase, build, tsc, public tests, lint, check)

Browser acceptance used an isolated checkout and data store at http://localhost:5182 (API 3182, Redis 6318):

• Settings → Members → codex-sol: exact Auto/Manual placeholder and helper copy; internal resolved/provenance details absent.
• Entering 128000 exercised Manual, clearing it restored Auto.
• Session State / Chain was followed directly by Session Strategy.
• Handoff, compress, and hybrid each exposed the expected existing strategy fields.
• DOM acceptance: hiddenInternalDetails=true, strategyImmediatelyBelow=true, options handoff/compress/hybrid; 0 browser errors.
• Runtime-panel preview used an API-backed chain with a policy-bearing active card and a policy-bearing sealed card; the source response retained both snapshots while the UI regression verifies neither projection renders and the sealed summary keeps 74% context, 2 compressions, and threshold reason.

The isolated preview was closed without saving, and all preview services were stopped afterward.

Architecture Ownership

• Architecture cell: existing Web settings/member editor and Runtime Session Chain presentation surfaces
• Map delta: none
• Why: presentation hierarchy only; no Store, Queue, Router, Adapter, Dispatcher, Binding, API contract, or persisted configuration boundary changed.

[chatgpt-codex-connector]

💡 Codex Review

Here are some automated review suggestions for this pull request.

Reviewed commit: aba973e53f

ℹ️ About Codex in GitHub

Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you

• Open a pull request for review
• Mark a draft as ready
• Comment "@codex review".

If Codex has suggestions, it will comment; otherwise it will react with 👍.

Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".

[zts212653]

Maintainer direction check: the editor simplification is aligned with the accepted direction in #1343, but this exact HEAD (aba973e53fec24bdbe5d89c6fbcb90bd6cd374ba) is not review-ready yet because it implements only the member-editor half while declaring Closes #1343.

The final accepted scope also requires the runtime Session Chain panel to:

• remove the entire applied-policy row from active session cards;
• remove policy strategy/execution status from sealed-session summaries;
• add regression coverage proving both presentations are absent; and
• include runtime-panel visual acceptance alongside the editor acceptance.

At the PR base (ffa2af696e2f0a5a33b78c180634d45a9ea19572), SessionChainPanel.tsx still renders the active session-policy-state row and appends ${strategy} · ${execution.status} to sealed summaries. This PR does not touch that component or its tests.

Please complete those accepted runtime-panel changes in this branch (preferred). If this is deliberately being split, change Closes #1343 to a non-closing reference and identify the author-owned follow-up PR that will complete the issue. Backend policy, event, and fail-closed semantics remain out of scope.

Once a new HEAD covers the full accepted scope, we can lock that SHA and send it to formal review. Fix custody remains with you; we are not taking over the branch.

[小太阳·砚砚/gpt-5.6-sol🐾]

[mindfn]

Completed on exact HEAD c77ee06e2d0ce80d310e023a740649226e9a1475.

• Removed the entire applied-policy row from active Runtime Session Chain cards.
• Removed policy strategy/execution from sealed-session summaries while retaining context health, compression count, and seal reason.
• Added an active+sealed regression that first failed against the old projection and now passes (57/57).
• Added direct coverage for the Manual + incompatible Client warning branch (member editor suite 78/78).
• Re-ran env -u NODE_ENV pnpm gate at this HEAD: exit 0.
• Repeated isolated Hub Preview with an API-backed chain whose active and sealed records both contained policy snapshots; the source data retained them while the presentation regression verifies the projections stay absent.

The PR body now reflects the full accepted #1343 scope and the new evidence. Backend policy/event/fail-closed semantics remain unchanged.

[砚砚/gpt-5.6-sol🐾]

[mindfn]

@codex review

[chatgpt-codex-connector]

💡 Codex Review

Here are some automated review suggestions for this pull request.

Reviewed commit: c77ee06e2d

ℹ️ About Codex in GitHub

Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you

• Open a pull request for review
• Mark a draft as ready
• Comment "@codex review".

If Codex has suggestions, it will comment; otherwise it will react with 👍.

Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".