pr-comment-fact-check

Étoiles4

Forks0

Mis à jour6 juillet 2026 à 10:43

STOP and fact-check PR review comments before applying or dismissing reviewer or bot feedback. Use on CodeRabbit, Copilot, Cursor bot, dependabot, or human reviewer comments.

Installation

Installer avec Codex ou Claude Copiez ce prompt, collez-le dans Codex, Claude ou un autre assistant, puis laissez-le vérifier la page du skill et l'installer pour vous.

Source

stainless-code

stainless-code/persist

Ouvrir le dépôt GitHub Voir les dépôts du créateur

Téléchargement

Métiers associésSOC

Basé sur la classification professionnelle SOC

Analystes en assurance qualité des logiciels et testeursProfessions informatiques et mathématiques·SOC 15-1253

Explorateur de fichiers

2 fichiers

SKILL.md

readonly

name	pr-comment-fact-check
description	STOP and fact-check PR review comments before applying or dismissing reviewer or bot feedback. Use on CodeRabbit, Copilot, Cursor bot, dependabot, or human reviewer comments.

PR comment fact-check

STOP before applying or dismissing any PR comment. Each comment is a claim — verify against the actual code and this repo's authoritative sources (.agents/, docs/architecture.md, JSDoc, tests) before acting. Workflow: WORKFLOW.md.

Verdict	Default action
✅ Correct	Apply + resolve
⚠️ Partial	Apply salvageable part + explain nuance
❌ Hallucinated	Push back with evidence; resolve-on-merge-gate exception in WORKFLOW
🕒 Outdated	Point at fix commit + resolve
💭 Style	Apply if cheap, else defer

Hallucination catalog (scrutinize harder)

Common LLM-reviewer patterns on this repo:

"This isn't tested" without checking siblings — persist-core contracts are pinned across src/core/persist-core.test.ts, src/adapters/codecs/seroval.test.ts, src/adapters/backends/idb.test.ts, src/adapters/sources/tanstack-store.test.ts; useHydrated spans src/adapters/frameworks/react.test.ts (bun, SSR/snapshot) and tests-dom/**/*.test.tsx (vitest, rerender/detach). Verify coverage before accepting.
Type-safety alarms — if bun run typecheck (tsgo) passes, the claim is almost always wrong, or about runtime behavior the type system can't see (then the reviewer must justify with the runtime case).
Generic "best practice" claims unsupported by our rules — "always destructure", "prefer interfaces over types", "add useMemo/useCallback" — stylistic; we either have a rule or we don't. Grep .agents/ for the convention.
Convention citations that don't exist — "this breaks the library's API conventions" — grep .agents/ + docs/architecture.md. If not codified, it's preference, not rule.
Memory-leak / race-condition claims with no concrete trigger — "this could leak" without a scenario is speculation; ask for the path. Real candidates here: hydration-gate ordering, throttle trailing edge, cross-tab listener teardown (destroy()), useSyncExternalStore unmount-detach — demand the specific path.
Wrong API for our seams — bot suggests coupling to a specific store when the seam is PersistableSource; suggests a factory-per-combination when factory policy is "backend earns a factory only when it needs real adaptation (IndexedDB)".
Public-API suggestions that leak internals — PersistStorage.raw is public and typed unknown on purpose (identity compare only); internal listener aliases are deliberately kept out of public signatures. Push back on "tighten this type" / "export this helper".

Anti-patterns

❌ Applying every suggestion to clear the queue.
❌ Replying "fixed!" without verifying.
❌ Dismissing without evidence.
❌ Resolving a thread you rejected as hallucinated (the reviewer needs to see the receipts).

Reference

verify-after-each-step — run after applying a fix.
harden-pr — optional full pass on the branch once comments are triaged.
docs/architecture.md — seam model, public-surface policy, test matrix (authoritative for "is this the right seam / is this tested" claims).

Plus depuis ce dépôt

même dépôt

alpine-persist

stainless-code/persist

Gate Alpine.js UI on Persist hydration (Alpine.plugin + useHydrated / $hydrated). Use when wiring HydrationSignal into Alpine data components.

2026-07-234

angular-persist

stainless-code/persist

Gate Angular UI on Persist hydration with useHydrated (readonly Signal). Call inside an injection context; use when avoiding flash of default state on async backends.

2026-07-234

lit-persist

stainless-code/persist

Gate Lit element UI on Persist hydration with HydrationController. Use when wiring HydrationSignal into a ReactiveControllerHost; not a hook.

2026-07-234

persist-async-storage

stainless-code/persist

Persist with React Native AsyncStorage via @stainless-code/persist/backends/async-storage. Use for RN string-wire JSON state; always gate with useHydrated.

2026-07-234

persist-compressed

stainless-code/persist

Compress a string-wire StateStorage with @stainless-code/persist/backends/compressed (gzip/deflate). Backend wrapper — compose with createStorage; stack compress-then-encrypt.

2026-07-234

persist-crosstab

stainless-code/persist

Cross-tab sync for backends without storage events via @stainless-code/persist/transport/crosstab (createBroadcastCrossTab). Wrap storage + pass crossTabEventTarget; call close() on teardown.

2026-07-234

PR comment fact-check

Verdict

Default action

✅ Correct

Apply + resolve

⚠️ Partial

Apply salvageable part + explain nuance

❌ Hallucinated

Push back with evidence; resolve-on-merge-gate exception in WORKFLOW

🕒 Outdated

Point at fix commit + resolve

💭 Style

Apply if cheap, else defer

Hallucination catalog (scrutinize harder)

Common LLM-reviewer patterns on this repo:

"This isn't tested" without checking siblings — persist-core contracts are pinned across src/core/persist-core.test.ts, src/adapters/codecs/seroval.test.ts, src/adapters/backends/idb.test.ts, src/adapters/sources/tanstack-store.test.ts; useHydrated spans src/adapters/frameworks/react.test.ts (bun, SSR/snapshot) and tests-dom/**/*.test.tsx (vitest, rerender/detach). Verify coverage before accepting.

Type-safety alarms — if bun run typecheck (tsgo) passes, the claim is almost always wrong, or about runtime behavior the type system can't see (then the reviewer must justify with the runtime case).

Generic "best practice" claims unsupported by our rules — "always destructure", "prefer interfaces over types", "add useMemo/useCallback" — stylistic; we either have a rule or we don't. Grep .agents/ for the convention.

Convention citations that don't exist — "this breaks the library's API conventions" — grep .agents/ + docs/architecture.md. If not codified, it's preference, not rule.

Memory-leak / race-condition claims with no concrete trigger — "this could leak" without a scenario is speculation; ask for the path. Real candidates here: hydration-gate ordering, throttle trailing edge, cross-tab listener teardown (destroy()), useSyncExternalStore unmount-detach — demand the specific path.

Wrong API for our seams — bot suggests coupling to a specific store when the seam is PersistableSource; suggests a factory-per-combination when factory policy is "backend earns a factory only when it needs real adaptation (IndexedDB)".

Public-API suggestions that leak internals — PersistStorage.raw is public and typed unknown on purpose (identity compare only); internal listener aliases are deliberately kept out of public signatures. Push back on "tighten this type" / "export this helper".

Anti-patterns

❌ Applying every suggestion to clear the queue.

❌ Replying "fixed!" without verifying.

❌ Dismissing without evidence.

❌ Resolving a thread you rejected as hallucinated (the reviewer needs to see the receipts).

Reference

verify-after-each-step — run after applying a fix.

harden-pr — optional full pass on the branch once comments are triaged.

docs/architecture.md — seam model, public-surface policy, test matrix (authoritative for "is this the right seam / is this tested" claims).