AgentSkillsCN

review-doc-consistency

文档一致性审核员,负责核查代码实现与文档之间的匹配度。当用户请求审核文档与代码的一致性、检查README/文档是否已过时,或验证API文档的准确性时,可使用此工具。具体应用场景包括:(1) 审核README与实现的一致性;(2) 检查docs/目录内容是否已过时;(3) 验证API/配置文档的准确性;(4) 生成文档一致性报告。触发关键词包括:文档审核、文档一致性、检查过时文档、验证文档。

SKILL.md
--- frontmatter
name: review-doc-consistency
description: 'Documentation consistency reviewer that checks alignment between code implementation and documentation. Use when user requests reviewing documentation vs code consistency, checking if README/docs are outdated, verifying API documentation accuracy. Applicable for (1) reviewing README vs implementation consistency (2) checking if docs/ directory content is outdated (3) verifying API/config documentation accuracy (4) generating documentation consistency reports. Trigger words include doc review, documentation consistency, check outdated docs, verify docs.'

Documentation Consistency Reviewer

Goal

Systematically identify all "outdated" or "inconsistent with implementation" descriptions in README + docs/, outputting ≥30 issue items.

Core Principles

  1. Code is truth - When documentation conflicts with code, source code/config/contract files are authoritative
  2. Evidence before conclusions - Each issue must cite code/config location as evidence
  3. Contracts first - OpenAPI/proto/schema/TS types are treated as SSOT (Single Source of Truth)
  4. Security default tightening - Security-related inconsistencies are prioritized as high severity

Review Process

1. Document Enumeration

bash
# Scan scope
- README.md (root directory)
- docs/**/*.md (all documentation)
- Contract files: OpenAPI/proto/GraphQL schema/TS types

2. Document-by-Document Review

For each document:

  1. List key claims/commitments/configs/interface items
  2. Search for corresponding implementation in code
  3. Compare differences: missing/renamed/behavior mismatch/default value mismatch
  4. Record issues using template

3. Cross-Check

  • Reverse check documentation from contract files
  • Reverse check documentation from config files

See checklist.md for detailed review checklist.

Severity Levels

LevelDefinitionExample
P0Security issue/serious misleadingDocs say sandbox enabled but code doesn't enable it
P1Core functionality inconsistencyFollowing docs leads to failure
P2Incomplete examples/naming inconsistencyDoesn't directly block usage
P3Wording/formatting/link minor issuesDoesn't affect functionality
Pending EvidenceSuspicious but insufficient evidenceNeeds further investigation

Output Format

See output-format.md for detailed templates.

Single Issue Item

markdown
### [Title]
- **Severity**: P0/P1/P2/P3/Pending Evidence
- **Location**: `<file_path>:<line_number>`
- **Evidence**:
  - Documentation: [quote]
  - Code: [quote]
- **Impact**: [Misleading consequences]
- **Suggestion**: [Minimal fix]
- **Related Principle**: Code is truth/Contracts first/Security default tightening/...

Review Conclusion

markdown
## Review Conclusion
- **Verdict**: Pass/Conditional Pass/Fail
- **Summary**: P0:x P1:x P2:x P3:x Pending:x
- **Fix Priority**: P0 → P1 → P2 → P3

Multi-Agent Parallel

For acceleration, split by following dimensions for parallel multi-agent execution:

  1. By document type - One agent each for README, API docs, development guide
  2. By module - One agent per functional module's documentation
  3. By check direction - One checks docs against code, another checks code against docs

Deduplication and unified severity rating needed when aggregating.

Execution

After review completion, output doc-consistency.md report file, and also output doc-consistency.json (structured issue list for aggregation/deduplication/statistics).