AgentSkillsCN

documentation-maintenance

通过可重复的审计闭环(衡量、修复、验证、报告),持续维护并优化任意代码仓库的技术文档。

SKILL.md
--- frontmatter
name: documentation-maintenance
description: Maintain and improve technical documentation with a repeatable audit loop (measure, fix, verify, report) across any repository.

Documentation Maintenance Skill

When to use

Use this skill when the user asks to review, update, or continuously improve docs such as design docs, ADRs, standards, runbooks, or guides.

Goal

Produce documentation that is correct, consistent, traceable, and easy to review.

Workflow

  1. Set scope
  • Identify target docs and related source files.
  • Define what "done" means for this pass.
  1. Measure current quality
  • Check metadata consistency (Last Updated, owner, status, links).
  • Scan for stale terms, broken references, and duplicated rules.
  1. Fix with minimal diffs
  • Update source-of-truth documents first.
  • Sync derivative docs (index pages, summaries, onboarding docs).
  • Keep wording concise and testable.
  • Continue the fix loop until all discovered in-scope issues are resolved, unless blocked.
  1. Verify
  • Re-check stale terms and links.
  • Confirm examples and commands still match current implementation.
  1. Report
  • Summarize file-level changes.
  • List remaining risks and every unresolved task (with blocker/assumption).

Quality rubric (0-5 each)

  • Accuracy
  • Consistency
  • Completeness
  • Traceability
  • Maintainability
  • Verifiability
  • Readability
  • Discoverability
  • Non-duplication
  • Iterability

Output format

  • Target level (Correctness / Operability / Self-improving)
  • Change policy (max 3 lines)
  • Edited files
  • Before/After quality notes
  • Evidence sources (official links or repo source)
  • Resolved tasks (all)
  • Unresolved tasks (all, with blockers)

Guardrails

  • Prefer source-of-truth first, then downstream docs.
  • Never hide uncertainty; state assumptions explicitly.
  • Avoid broad rewrites unless the user asks for restructuring.
  • Default behavior is full in-scope repair, not "top N" sampling.