file-audit

name	file-audit
description	Use when: per-file AGENTS.md compliance audit with Chinese reports
allowed-tools	Bash, Read, Grep, Glob, Write
triggers	["file-audit","/file-audit","audit files","AGENTS.md compliance"]

File Audit: Per-File AGENTS.md Compliance Audit

Role Detection (READ THIS FIRST -- MANDATORY)

Check your initial prompt. If it contains the literal string "Use the file-audit skill", then:

YOU ARE THE EXECUTOR. Follow these rules:

1. SKIP the "Execution Protocol" section below -- it is for the orchestrator, not you.
2. Read the pattern at ../../PATTERN.md relative to this SKILL.md, and follow it step by step.
3. ABSOLUTE PROHIBITION: Do NOT run csa run, csa review, csa debate, or ANY csa command. You must perform the work DIRECTLY. Running any csa command causes infinite recursion.

Only if you are the main agent (Claude Code / human user):

• You are the orchestrator. Follow the "Execution Protocol" steps below.

---

Purpose

Audit each source file in a given scope against the applicable AGENTS.md rules chain (root-to-leaf). For every file, generate a Chinese audit report with checkbox compliance evidence under ./drafts/audit/, mirroring the source directory structure. Conclude with a Chinese tech blog summarizing findings, patterns, and lessons learned about maintaining coding standards with AI agents.

Execution Protocol (ORCHESTRATOR ONLY)

Prerequisites

• At least one AGENTS.md file must exist in the project
• Target scope (directory or file pattern) must be specified

Quick Start

csa run --sa-mode true --skill file-audit "Audit all files in src/executor/ against AGENTS.md rules"

SA Mode Propagation (MANDATORY)

When operating under SA mode (e.g., dispatched by /sa or any autonomous workflow), ALL csa invocations MUST include --sa-mode true. This includes csa run, csa review, csa debate, and any other execution commands. Omitting --sa-mode at root depth causes a hard error; passing false when the caller is in SA mode breaks prompt-guard propagation.

Step-by-Step

1. Discover AGENTS.md chain: Find all AGENTS.md files from repository root to source directories. Build rule chain (root -> leaf, deepest wins on conflict).
2. Collect target files: List source files in the audit scope (e.g., find src/ -name "*.rs").
3. Prepare output directory: Create mirrored directory structure under ./drafts/audit/.
4. Per-file audit (via CSA, tier-1 for each file): Read file, check each applicable AGENTS.md rule, record pass/fail with evidence. Output a Chinese Markdown report with:
   • File header (path, line count, crate)
   • Applicable rules checklist (checkbox format)
   • Findings (violations with line numbers)
   • Summary verdict (PASS / PASS_WITH_NOTES / FAIL)
5. Write audit reports: Save each report to ./drafts/audit/${RELATIVE_PATH}.audit.md.
6. Violation summary (if violations found): Aggregate all violations into a summary table grouped by rule ID.
7. Tech blog: Write a Chinese tech blog to ./drafts/audit/blog.zh.md covering methodology, key findings, patterns, and lessons learned.

Example Usage

Command	Effect
/file-audit src/executor/	Audit all .rs files in executor module
/file-audit src/	Audit entire src directory
/file-audit crates/csa-acp/src/	Audit the csa-acp crate

Integration

• Standalone: Does not depend on other skills
• Related to: csa-review (diff-based review vs per-file audit), security-audit (security focus vs compliance focus)
• Output: ./drafts/audit/ directory with per-file reports and summary blog

Done Criteria

1. AGENTS.md chain discovered from root to all source directories.
2. All target files in scope identified and audited.
3. Per-file Chinese audit report generated with checkbox compliance evidence.
4. Reports written to ./drafts/audit/ mirroring source directory structure.
5. If violations found: aggregated violation summary table produced.
6. Chinese tech blog written to ./drafts/audit/blog.zh.md.
7. Each report includes verdict: PASS, PASS_WITH_NOTES, or FAIL.