AgentSkillsCN

context-audit

针对 CLAUDE.md 上下文文件进行全面的质量审核。适用于执行深度质量检查、为发布做准备、确保上下文效率，或验证令牌优化效果时使用。

SKILL.md
--- frontmatter
name: context-audit
description: Comprehensive quality audit for CLAUDE.md context files. Use when performing thorough quality checks, preparing for releases, ensuring context efficiency, or verifying token optimization.
allowed-tools: Read, Glob, Grep, Bash
context: fork
agent: context-doc-agent

LLM Context Comprehensive Audit

Performs deep quality audits of CLAUDE.md context files, checking structure, content quality, efficiency, design doc references, and compliance with standards.

Overview

This skill provides comprehensive quality auditing for LLM context files by running all validation checks, analyzing content efficiency, verifying design doc pointers, checking line limits, and generating detailed audit reports with prioritized recommendations.

Quick Start

Audit all context files:

bash
/context-audit

Audit specific file:

bash
/context-audit CLAUDE.md

Audit package context:

bash
/context-audit pkgs/effect-type-registry/CLAUDE.md

Quick audit (non-strict):

bash
/context-audit --strict=false

Parameters

Optional

  • •target: Path to CLAUDE.md file or "all" (default: all)
  • •strict: Enable strict mode with additional checks (default: true)
  • •check-refs: Validate design doc references exist (default: true)
  • •output: Output file path for audit report

Workflow

High-level audit process:

  1. •Parse parameters to determine audit scope and strictness
  2. •Load design.config.json to get quality standards (line limits, etc.)
  3. •Discover CLAUDE.md files using Glob (root + package-level)
  4. •Run validation checks (structure, formatting, markdown quality)
  5. •Analyze content quality (efficiency, organization, token usage)
  6. •Check design doc pointers (existence, validity, coverage)
  7. •Verify line limits (root: 500, child: 300 from config)
  8. •Calculate health scores (file, package, overall)
  9. •Identify issues by severity (critical, high, medium, low)
  10. •Generate recommendations prioritized by impact
  11. •Output audit report with actionable improvements

Instructions

IMPORTANT: Follow the detailed step-by-step instructions in instructions.md to perform the audit correctly.

For usage examples and common scenarios, see examples.md.

Output Format

The audit generates a structured report with:

Summary Section

  • •Total files audited
  • •Overall health score (0-100)
  • •Critical/high/medium/low issue counts
  • •Pass/fail status

File-Level Details

For each CLAUDE.md file:

  • •File path and role (root vs child)
  • •Line count vs limit
  • •Structure validation results
  • •Content quality score
  • •Design doc pointer status
  • •Specific issues found

Recommendations

Prioritized list of improvements:

  1. •Critical issues (must fix)
  2. •High priority (should fix)
  3. •Medium priority (nice to have)
  4. •Low priority (optional)

Quality Metrics

  • •Average line count
  • •Design doc pointer coverage
  • •Content efficiency score
  • •Token optimization score

Success Criteria

The audit passes when:

  • •All files under line limits (root: 500, child: 300)
  • •No critical or high severity issues
  • •All design doc pointers valid and exist
  • •Content is lean imperative instructions (not implementation details)
  • •Proper separation between root and child contexts

Related Skills

  • •/context-validate - Basic structure and formatting validation
  • •/context-review - Quality and efficiency review
  • •/context-update - Update context files based on audit findings
  • •/context-split - Split large files that exceed limits
  • •/design-audit - Similar comprehensive audit for design docs