fix(worker-bus): P0 timeout enforcement + P1 memory-tier inference pool sizing

.cursorrules

@@ -21,9 +21,9 @@ Analysiere **jede Datei, jedes Feature und jedes Repo individuell und ganzheitli
 
 ---
 
-## StoryCraft-Studio (dieses Repo) — Kurzkontext
+## WorldScript-Studio (dieses Repo, ehem. StoryCraft-Studio) — Kurzkontext
 
-Offline-first, **Privacy**, Redux Toolkit + IndexedDB + Zustand (transient UI), Hybrid-KI (Gemini, Vercel AI SDK, Ollama/WebLLM), pnpm-Workspace (`packages/*`), PWA, 5-Locale-i18n. Fokus: **Erzählung**, Schreibflow, KI als Co-Pilot — nicht als Ghostwriter. Vollständiger Stack und Regelkatalog: **`.cursor/index.mdc`**.
+Offline-first, **Privacy**, Redux Toolkit + IndexedDB + Zustand (transient UI), Multi-Provider-KI (Gemini, Vercel AI SDK, OpenAI/Claude/Grok/OpenRouter, Ollama/WebLLM/ONNX/Transformers.js), pnpm-Workspace (`packages/*`), PWA + Tauri 2, 19-Locale-i18n. Fokus: **Erzählung**, Schreibflow, KI als Co-Pilot — nicht als Ghostwriter. Vollständiger Stack und Regelkatalog: **`.cursor/index.mdc`**.
 
 ---
 
@@ -46,7 +46,7 @@ Inhalt des Kommentars: **warum** die Änderung, **was** sie verbessert, optional
 
 Bei jeder offenen PR **alle** Inline-Review-Kommentare (CodeRabbit + jeder andere Bot/Mensch) **proaktiv, ungefragt** abarbeiten. Pro Finding: gegen den **aktuellen** Code validieren (Anker oft veraltet), dann **Root-Cause-Fix** vollständig umsetzen (Code **+ Tests + i18n + Doku**) oder mit Begründung als False-Positive ablehnen. Jeden Thread mit Commit-Referenz beantworten, **resolven**, **0 unresolved** hinterlassen, committen + pushen.
 
-**Iron Rule — Loop bis Ruhe:** Nach jedem Push **Review erneut triggern** (`gh pr comment <N> --body "@coderabbitai review"`) und die komplette Review-Historie prüfen, nicht nur den letzten Status (ein „rate limited"-Status kann eine frühere echte Review verdecken); eine frische Review bringt regelmäßig **neue** Findings (eine „Welle"). Zyklus wiederholen, bis **BEIDES** gilt: frische Review = **0 neue Kommentare** UND **0 offene Threads**. Niemals aufhören, solange noch Kommentare kommen. **Niemals ein neues `biome-ignore`/Suppression** einbauen, um ein Finding stummzuschalten — der Suppression-Ratchet-Gate lässt die CI scheitern; stattdessen sauber refaktorieren. CodeAnt AI erscheint nur als 5 CI-Status-Checks (Quality Gates/SAST/SCA/SCR/Test Coverage) zur Grün-Prüfung, kein Kommentar-Thread — dort nicht erneut triggern in Erwartung eines Kommentars. Vollständige Prozedur: **`docs/CODEANT-REVIEW-LOOP.md`**.
+**Iron Rule — Loop bis Ruhe:** Nach jedem Push **Review erneut triggern** (`gh pr comment <N> --body "@coderabbitai review"`) und die komplette Review-Historie prüfen, nicht nur den letzten Status (ein „rate limited"-Status kann eine frühere echte Review verdecken); eine frische Review bringt regelmäßig **neue** Findings (eine „Welle"). Zyklus wiederholen, bis **BEIDES** gilt: frische Review = **0 neue Kommentare** UND **0 offene Threads**. Niemals aufhören, solange noch Kommentare kommen. **Niemals ein neues `biome-ignore`/Suppression** einbauen, um ein Finding stummzuschalten — der Suppression-Ratchet-Gate lässt die CI scheitern; stattdessen sauber refaktorieren. CodeAnt AI erscheint nur als 5 CI-Status-Checks (Quality Gates/SAST/SCA/SCR/Test Coverage) zur Grün-Prüfung, kein Kommentar-Thread — dort nicht erneut triggern in Erwartung eines Kommentars. Vollständige Prozedur (inkl. des komplementären, token-freien DeepSource-Loops): **`docs/CODEANT-REVIEW-LOOP.md`** + **`docs/DEEPSOURCE-REVIEW-LOOP.md`**.
 
 ---
 

KIMI-INSTRUCT.md

@@ -40,6 +40,11 @@ Dieses Projekt läuft auf Low-End-Hardware. Beachte strikt:
 
 Dieser Workflow ist die Standardvorgehensweise, wenn der Nutzer um Behebung von CodeAnt-Kommentaren bittet.
 
+> **Kanonische Quelle:** Die vollständige, agenten-unabhängige Prozedur (inkl. GraphQL-Snippets, PR-Split
+> bei >100 Dateien, Merge-Regeln) lebt in **`docs/CODEANT-REVIEW-LOOP.md`**; der komplementäre,
+> token-freie statische-Analyse-Loop in **`docs/DEEPSOURCE-REVIEW-LOOP.md`**. Der Abschnitt unten ist
+> eine Kimi-spezifische Kurzfassung — bei Abweichungen gelten die beiden kanonischen Dateien.
+
 > **In der Praxis (Beobachtung 2026-07-30):** Der Bot, der tatsächlich Inline-Kommentare postet, ist
 > **CodeRabbit** (`@coderabbitai review` zum erneuten Triggern) — nicht CodeAnt AI. CodeAnt AI
 > erscheint als 5 CI-Status-Checks (`CodeAnt - Quality Gates/SAST/SCA/SCR/Test Coverage`), die auf
@@ -102,7 +107,8 @@ pnpm exec vitest run <betroffene-testdateien>
 
 - Lint/Format-Fehler vor dem Commit beheben (`biome check --write ...`).
 - TypeScript-Fehler sofort beheben.
-- i18n-Keys bei neuem UI-Text zu **allen 11 Locales** hinzufügen und Bundles neu bauen.
+- i18n-Keys bei neuem UI-Text zu **allen 19 Locales** hinzufügen (`node scripts/check-i18n-keys.mjs --fix`) und Bundles neu bauen (`pnpm run i18n:bundle`).
+- Commit-Nachrichten mit der Zuordnung des tatsächlich arbeitenden Agenten/Modells abschließen, z. B. `Co-Authored-By: Kimi K2 <noreply@moonshot.ai>` — niemals eine generische oder falsche Modellangabe übernehmen.
 
 ### 3.4 Commit & Push
 

README.md

@@ -505,7 +505,7 @@ The Settings → AI panel shows a live GPU status badge with adapter details and
 | **PDF Export**       | jsPDF                                                     | Client-side, configurable PDF document generation                    |
 | **Document Export**  | docx + jszip                                              | Word-compatible `.docx` generation (lazy-loaded)                     |
 | **PWA**              | Service Worker + Web App Manifest v3                     | Offline support, installability, Workbox chunking                    |
-| **i18n**             | Custom React Context (`I18nContext.tsx`)                  | 2861 keys × 19 locales (de/en/es/fr/it + ar/he/fa RTL Beta + ja/zh/pt/el/fi/sv/hu/is/eu Beta); EN fallback; `localStorage` persistence |
+| **i18n**             | Custom React Context (`I18nContext.tsx`)                  | 2861 keys × 19 locales (de/en/es/fr/it + ar/he/fa RTL Beta + ja/zh/pt/el/fi/sv/hu/is/eu/ru/ko Beta); EN fallback; `localStorage` persistence |
 | **Testing**          | Vitest 4.x (6477+ tests / 532 files) + Playwright E2E     | Unit/integration + cross-browser E2E; Stryker mutation (manual workflow)          |
 | **Code Quality**     | Biome (lint + format) + TypeScript 7 (tsgo) strict       | `--error-on-warnings` in CI; zero `any` policy                      |
 | **Visualization**    | Force-directed graph                                      | Interactive character relationship network                           |
@@ -748,7 +748,7 @@ Shared Playwright helpers (`waitForSpaReady`, `ensureBlankProject`, `clickNavIte
 
 - **🐛 Report Bugs** — Open a GitHub Issue with details and reproduction steps
 - **💡 Suggest Features** — Open a Discussion or Issue
-- **🌍 Improve Translations** — Five locale trees (`en` is the reference); native polish for FR/ES/IT especially welcome in PRs
+- **🌍 Improve Translations** — Five core locale trees (`en` is the reference); native polish for FR/ES/IT especially welcome in PRs
 - **🧪 Write Tests** — Coverage thresholds: branches ≥ 60 %, functions ≥ 67 %, lines ≥ 74 %; contributions to large components (collaboration, AI streaming paths, OpenRouter provider) are particularly valuable
 
 See **[`CONTRIBUTING.md`](CONTRIBUTING.md)** for the full dev setup, Biome / Vitest / Playwright guide, and architecture notes.
@@ -767,6 +767,7 @@ See **[`CONTRIBUTING.md`](CONTRIBUTING.md)** for the full dev setup, Biome / Vit
 | [`AUDIT.md`](AUDIT.md) | Security & quality audit trail + scorecard |
 | [`docs/AUDIT-PERFECTION-PLAN-v1.23.md`](docs/AUDIT-PERFECTION-PLAN-v1.23.md) | Living master plan for the v1.23 audit/perfection engagement (phase status + batch log) |
 | [`docs/CI.md`](docs/CI.md) | GitHub Actions jobs, Node/pnpm parity, Act examples |
+| [`docs/CODEANT-REVIEW-LOOP.md`](docs/CODEANT-REVIEW-LOOP.md) | Canonical, agent-agnostic PR review correction loop (CodeRabbit/CodeAnt/any bot) — GraphQL fetch/reply/resolve, PR-size split, merge rules |
 | [`docs/DEEPSOURCE-REVIEW-LOOP.md`](docs/DEEPSOURCE-REVIEW-LOOP.md) | DeepSource static-analysis correction loop (living runbook; complements the CodeAnt loop) |
 | [`docs/DEEPSOURCE-REMEDIATION-PLAN.md`](docs/DEEPSOURCE-REMEDIATION-PLAN.md) | Prioritised DeepSource backlog tracker (P0-security→P5-docs) with triage decisions |
 | [`docs/adr/`](docs/adr/README.md) | Architecture Decision Records — state-management boundaries, local-AI stack layering, WorkerBus v2 hybrid routing |

docs/CODEANT-REVIEW-LOOP.md

@@ -38,7 +38,7 @@ Each wave is handled exactly like the first.
 > Until both are true, **keep iterating**. Never declare the PR done while new comments are still
 > arriving or any thread is open.
 
-```
+```text
         ┌─────────────────────────────────────────────┐
         │ 1. Fetch unresolved threads (GraphQL)        │
         │ 2. Validate each against CURRENT code        │
@@ -167,10 +167,10 @@ git commit -m "refactor(scope): address CodeAnt wave N review feedback"
 git push origin <feature-branch>
 ```
 
-End commit messages with:
+End commit messages with the attribution matching whichever agent/model actually made the change, e.g.:
 
-```
-Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
+```text
+Co-Authored-By: GitHub Copilot (Claude Sonnet 5) <noreply@github.com>
 ```
 
 ## 6. Reply to every thread, then resolve it → 0 unresolved

docs/DEEPSOURCE-REMEDIATION-PLAN.md

@@ -88,6 +88,9 @@ Fix in this order — highest user/security impact first; cosmetic last. Within
 |---|---|---|---|---|---|---|
 | **JS-0323** | Detected usage of `any` (Critical) | 34 / 20 files | **100% test files** (browser-API mocks: MediaRecorder/AudioContext …), all already `biome-ignore noExplicitAny`'d; **0 production** | **rule-ignore, test scope** — redundant with Biome `noExplicitAny` + the suppression ratchet (baseline 52). Do NOT add 34 inline `skipcq`. Long-term: opportunistic typed-mock cleanup (§4c). | maintainer-click (rule-ignore) | — |
 | **JS-0415** | JSX tree too deeply nested (5–6 levels) | 2 | `components/HelpView.tsx` | pre-existing structural; genuine refactor candidate (extract sub-components to cut nesting). Minor severity → schedule as a low-priority hygiene PR; **not** a false positive. | todo (future refactor) | — |
+| **JS-0067** | Unexpected function declaration in the global scope | 2+ (repo-wide pattern) | `services/ai/localAiDeviceProfiler.ts`, `services/workerBusManager.ts` (representative — every top-level `services/*.ts` function hits this) | **rule-ignore, repo-wide** — idiomatic top-level ES-module functions, not `<script>`-tag globals; fixing per-occurrence would touch dozens of pre-existing files for no behavioral gain. | todo (maintainer dashboard click) | #305 |
+| **JS-R1005** | Cyclomatic complexity (medium risk) | 5 | `packages/worker-bus/src/workerBus.ts` (`enqueue`=11), `services/ai/localAiDeviceProfiler.ts` (`detectWebnn`=10, `detectMemoryTier`=9, `recommendBackend`=9, `detectDirectML`=6) | **pre-existing, out of scope for #305** — confirmed via diff that none of these 5 functions were touched by the PR (only `runOnPool`/`waitForIdle` in `workerBus.ts` and an `export` keyword add on `detectMemoryTier`); DeepSource's PR view widens "introduced" to every issue in a touched file, not just changed lines. Genuine refactor candidates (extract sub-checks to cut branching) — schedule as a dedicated low-priority hygiene PR, not bundled into feature work. | todo (future refactor) | #305 |
+| **JS-C1002** | Variable name is too small | 2 | `packages/worker-bus/src/workerBus.ts` (`toError`'s `e`), `packages/worker-bus/tests/workerBus.test.ts` (`t` in a telemetry assertion) | **pre-existing, out of scope for #305** — same "introduced" widening as JS-R1005 above; neither line is in this PR's diff. Trivial rename (`e`→`err`, `t`→`telemetry`) whenever either file is next touched for an unrelated reason. | todo (opportunistic) | #305 |
 
 ### P4 — Hygiene / Style
 | Issue code | Title | Occ. | Files / modules | Decision | Status | PR |
@@ -148,6 +151,7 @@ config level (`.deepsource.toml`) instead of scattering inline `skipcq`. Log eac
 |---|---|---|---|
 | **JS-0323** (`any`) | dashboard rule-ignore, **test-file scope** | 34 occ all in test mocks, already governed by Biome `noExplicitAny` + the suppression ratchet (baseline 52); double-suppressing is noise. 0 production `any`. Long-term cleanup → §4c. | 2026-06-24 |
 | **JS-0440** (`dangerouslySetInnerHTML`) | dashboard "Ignore" (single occurrence) | reviewed-safe — `__html` is DOMPurify-sanitized (repo policy). In-code skipcq can't attach to a JSX attribute (§4b). | 2026-06-24 |
+| **JS-0067** ("unexpected function declaration in the global scope") | recommend dashboard rule-ignore (not yet actioned — maintainer click) | flagged on PR #305 for ordinary top-level `function`/`async function` declarations in `services/ai/localAiDeviceProfiler.ts` + `services/workerBusManager.ts` — idiomatic Biome-approved ES-module code, not `<script>`-tag global pollution; repo-wide pattern, not a per-file fix. Informational-only check (not in required-status list). See `DEEPSOURCE-REVIEW-LOOP.md` §11, 2026-08-01 entry. | 2026-08-01 |
 
 ## 6. Definition of done
 

docs/DEEPSOURCE-REVIEW-LOOP.md

@@ -60,7 +60,7 @@ List + resolve via the GraphQL snippet in §3a.
 A push that fixes findings triggers a **fresh** DeepSource run, which can surface **new** findings
 caused by the fix (a "wave"). Handle each wave like the first.
 
-```
+```text
         ┌─────────────────────────────────────────────┐
         │ 1. Fetch DeepSource findings (check-runs +   │
         │    annotations) for the PR head SHA          │
@@ -215,7 +215,8 @@ git commit -m "refactor(scope): address DeepSource wave N (<issue codes>)"
 git push origin <feature-branch>
 ```
 
-End commit messages with `Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>`.
+End commit messages with the attribution matching whichever agent/model actually made the change, e.g.
+`Co-Authored-By: GitHub Copilot (Claude Sonnet 5) <noreply@github.com>`.
 **No manual re-trigger** — DeepSource analyses the new head SHA on its own. Wait for the fresh run,
 then go back to §3.
 
@@ -270,6 +271,21 @@ GitHub App resumes auto-reviewing, run **both** loops: CodeAnt for narrative/AI
   new service/hook needs its branches tested or patch coverage dips below the ~72% target and the
   `codecov/patch` check fails. #236's hook+panel needed a dedicated hook test (offline/error/disabled/
   stale-apply/dictionary/clear branches) to clear it — same posture as #232's binderDepth test.
+- **2026-08-01** — On PR #305, `DeepSource: JavaScript` surfaced **JS-0067** ("unexpected function
+  declaration in the global scope") on top-level `function`/`async function` declarations in
+  `services/ai/localAiDeviceProfiler.ts` and `services/workerBusManager.ts` — a pattern used
+  throughout this repo's ES modules (every service file declares top-level functions; this is
+  idiomatic, Biome-approved module code, not `<script>`-tag global-scope pollution). Confirmed via
+  `gh api .../branches/main/protection`: only the `CI Success` aggregator is a **required** status
+  check — `DeepSource: JavaScript` is informational-only and does not block merge. Per §4a.3, do not
+  scatter `# skipcq: JS-0067` across dozens of pre-existing functions repo-wide to chase this; the
+  correct fix is a **dashboard rule-level ignore** for JS-0067 (maintainer action, not toml-expressible
+  — DeepSource rule toggles live under Settings → Issue Types on app.deepsource.com, not
+  `.deepsource.toml`). Real, valuable findings from the same wave (real bugs, not style noise) were
+  fixed instead: a stale-`MessagePort`-reuse bug (late PROGRESS/RESULT from a cancelled task could
+  reset/resolve a new task sharing the same released port — fixed by filtering on `msg.taskId` in
+  `workerBus.ts`), and a missing `port.start()` call in `workerPool.ts::spawnWorker()` (an
+  `addEventListener`-based `MessagePort` never dispatches without it per the WHATWG spec).
 
 ---
 

docs/SESSION-HANDOFF-MAIN-CI-COVERAGE.md

@@ -25,7 +25,7 @@ The diagnostic narrative below is retained for the record.
 **Symptom:** On `main`, the **`CI / CD` → `🔍 Quality Gate`** job (both Node 22 and Node 24) fails on the
 **coverage threshold**:
 
-```
+```text
 ERROR: Coverage for lines (5.16%) does not meet global threshold (74%)
 ERROR: Coverage for functions (4.91%) ... (67%)
 ERROR: Coverage for statements (5.02%) ... (72%)
@@ -113,7 +113,7 @@ the coverage-threshold path (i.e. not as a hard test failure). Candidates, in pr
 **⚠️ BLOCKER on #180 — real typecheck failure (must fix before commit):**
 `pnpm run typecheck` fails with **3× TS2352** at `features/proForge/machine/proForgeMachine.ts` lines
 **68, 71, 75** (the pre-existing `setPreSnapshot`/`setResult`/`setDecision` actions):
-```
+```text
 Conversion of type 'ProForgeMachineEvent' to type '{ output: ... }' may be a mistake ...
 Property 'output' is missing in type '{ type: "ABORT"; }' ...
 ```
@@ -158,5 +158,6 @@ Flow-Mode fullscreen + docs). Each = one PR off main, CodeAnt deferred.
 ## Environment reminders
 - **ONE Bash call per turn** (low-end hardware ~3.7 GB RAM); no concurrent heavy shells. A full local
   coverage run ≈ 1 hour — prefer single-file runs for diagnosis.
-- Co-Author commits with `Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>`.
+- Co-Author commits with the attribution matching whichever agent/model actually made the change, e.g.
+  `Co-Authored-By: GitHub Copilot (Claude Sonnet 5) <noreply@github.com>`.
 - Never lower coverage thresholds or `it.skip` to go green — fix the root cause.