菜单
Audit an agent or skill prompt file for internal inconsistencies (phase numbering, behavioral contracts, terminology, stale references). Produces a Q&A with info-gain scores, a learnings section, and a remediation list. Use before any prompt migration.
用 Codex 或 Claude 帮你安装 复制这段 Prompt，粘贴到 Codex、Claude 或其他助手里，让它检查 Skill 页面并帮你完成安装。
基于 SOC 职业分类
| name | prompt-audit |
| description | Audit an agent or skill prompt file for internal inconsistencies (phase numbering, behavioral contracts, terminology, stale references). Produces a Q&A with info-gain scores, a learnings section, and a remediation list. Use before any prompt migration. |
Audit a prompt file for internal inconsistencies. Every finding cites file:line.
/prompt-audit <path-to-prompt-file>
<path-to-prompt-file> is a repo-relative path to any agents/*.md or skills/*/SKILL.md.
Read the full content of <path-to-prompt-file> using the Read tool.
Search for files that reference the target prompt:
grep -rn "<prompt-file-basename>" ${CLAUDE_PLUGIN_ROOT}/../../ --include="*.md" --include="*.js" -l
Read each related file (envelope template, test harness skill, test fixtures, test JS file).
For each dimension below, enumerate findings. Every finding must cite file:line.
| Dimension | What to check |
|---|---|
| Phase model | Phase table ↔ specialist catalog ↔ test harness all use the same phase IDs and numbering |
| Behavioral contracts | Every contract stated in one file (e.g. envelope) is consistent with the agent body |
| Terminology | Terms like "Orchestrator", "cascade", "DISPATCHED" used consistently across the file |
| Stale references | Date literals, commit SHAs, removed section names, deprecated field names |
| Skip conditions | Skip rules in the phase table match skip rules in the test harness |
| Standard wiring | Dependency graph stated in the agent matches the graph in the harness |
| Specialist catalog coverage | Every non-trivial non-inline phase has a catalog entry |
For each inconsistency that requires a design decision (not just a mechanical fix), emit:
### Q<N> — <short question>
**Info-gain: <0.0–1.0>** — <one sentence explaining what uncertainty this resolves>
**Evidence:**
- <file:line>: <what it says>
- <file:line>: <what it says (contradicting or context)>
**Answer:** <your reasoned answer, citing evidence>
Sort questions highest-info-gain first. Pure mechanical fixes (typos, stale dates) go in Step 5 directly without a Q entry.
List non-obvious discoveries that a future prompt author should know:
## Learnings
1. <finding> — <why it matters>
Emit a prioritized list of changes needed:
## Remediation
| Priority | File | Change |
|----------|------|--------|
| CRITICAL | <file> | <what to fix> |
| HIGH | <file> | <what to fix> |
| MEDIUM | <file> | <what to fix> |
| LOW | <file> | <what to fix> |
Resolve ambiguous user prompts by choosing whether to answer now, answer with assumptions, ask 1–2 high-value clarification questions, replan, or stop. Use when the best response depends on hidden intent, audience, scope, constraints, risk, output format, or desired depth.
Analyzes an approved plan and decomposes it into a dependency-ordered task graph via a two-phase TaskCreate and TaskUpdate pass. Detects linear DEPENDS ON chains (shared worktree), identifies independent work streams (parallelized), and executes the full worktree-isolated run. Also handles Branch B (learnings from session context). **AUTOMATICALLY INVOKE** only when: - The PostToolUse(ExitPlanMode) hook has injected its schedule-tasks nudge in this turn (i.e., the user just approved a plan), AND the user has not signaled they want to defer execution. - /schedule-plan-tasks is invoked explicitly. - The user explicitly says "schedule tasks", "execute the plan", "decompose plan into tasks" — only when a plan exists at ~/.claude/plans/*.md AND ExitPlanMode has already been approved in this session. **DO NOT auto-invoke** while plan mode is active, while review-plan is iterating, or before the user has approved ExitPlanMode. **References:** JIT-loaded from `${CLAUDE_SKILL_DIR}/references/`
Headless `claude -p` test harness V2 — runs a fixture's literal prompt through real Claude sessions (k replicas in parallel), grades each replica on five string views plus structured tool-call / tool-result / run-health / LLM-judge condition families, then quorum-aggregates row verdicts. Use when an LLM-driven prompt needs behavioral verification beyond contract-string parser tests.
Universal plan review: senior-engineer directive + deterministic count-based severity rule. AUTOMATICALLY INVOKE when: - MANDATORY_PRE_EXIT_PLAN directive applies (before ExitPlanMode) - User says "review plan", "check plan", "plan ready?" - Any plan file needs review NOT for: Code review of existing files (use /gas-suite:gas-review or /review-fix)
Compare an agent or skill prompt against its test harness skill for phase-model, skip-condition, and wiring consistency. Reports mismatches and identifies which file is authoritative.
TDD-based prompt migration — given a target agent/skill prompt and a remediation list from prompt-audit, writes failing tests first, then updates the prompt to make them pass, and commits both in a single atomic commit.