メニュー
Use when: reviewing agents/*.md, skills/*/SKILL.md, or .mdc rules for contradictions, ambiguity, persona drift, cognitive load, coverage gaps, and composition conflicts before merge.
Codex または Claude でインストール この Prompt をコピーして Codex、Claude、または他のアシスタントに貼り付けると、Skill ページを確認してインストールできます。
SOC 職業分類に基づく
Use when running workspace tests, choosing a test command, or summarizing results before handoff to debugger.
Use when running workspace tests, choosing a test command, or summarizing results before handoff to debugger.
Use when: running pre-commit or pre-push CI-equivalent checks in any workspace — discovers workflow commands with Cursor tools, filters for local executability, scopes to staged changes, and runs checks cheapest-first.
Use when the user wants to install or customize cursorAssistant in the current project (GitHub install or configure).
Use when: running pre-commit or pre-push CI-equivalent checks in any workspace — discovers workflow commands with Cursor tools, filters for local executability, scopes to staged changes, and runs checks cheapest-first.
Use when: discovering package manifests, assessing dependency health, finding replacements, or confirming import usage before removal — discovery only; mutating installs are handled by the deps agent.
| name | surfaceReview |
| description | Use when: reviewing agents/*.md, skills/*/SKILL.md, or .mdc rules for contradictions, ambiguity, persona drift, cognitive load, coverage gaps, and composition conflicts before merge. |
| version | 1.1 |
| license | MIT |
| disable-model-invocation | true |
| paths | ["agents/**","skills/**","template/rules/**",".cursor/agents/**",".cursor/skills/**",".cursor/rules/**","AGENTS.md"] |
Six-module audit of Cursor instruction surfaces. Invoke with /surfaceReview (not auto-applied). For module checklists and severity rules, read references/modules.md when executing.
Use Read, Grep, Glob; run cursorEval check / tokens on SKILL.md when tools/cursorEval/ exists.
.mdc rulesreview subagent/surfaceReviewlifecycleAudit or cursorLifecycleLabel each module in output as Module 1 … Module 6 (names below).
| # | Module | Output prefix |
|---|---|---|
| 1 | Contradiction detection | contradiction: |
| 2 | Semantic ambiguity | ambiguity: (include rewrite) |
| 3 | Persona consistency | persona: |
| 4 | Cognitive load | cognitive-load: (metric + threshold) |
| 5 | Semantic coverage | coverage-gap: + standalone scope-coverage: line |
| 6 | Composition conflicts | composition: |
After all modules, emit a table: Severity | Module | Section | Finding | Suggested fix. Empty module → one row — | <module> | — | No findings. | —.
Close with:
Result: N critical, N high, N medium, N low — [ready to merge | needs revision before merge | block: restructure required]
| Rule | Outcome |
|---|---|
| Any Critical | block: restructure required |
| Any High | needs revision before merge |
| Else | ready to merge |
Optional: cursorEval quality / dev (needs GITHUB_MODELS_TOKEN).