‹ 首页

doc-evidence-compare

@quanxiaoxiao · 收录于 今天 · 上游提交 3 个月前

Compare repository documentation and code evidence for a topic, check whether a business rule or relationship is documented, locate implementation evidence, and identify mismatches or missing documentation with file-backed conclusions.

适合你,如果常需确认文档是否准确反映代码实现

/ 通过 npx 安装 校验哈希
npx oh-my-skill add quanxiaoxiao/quan-skills/doc-evidence-compare
/ 通过 bash 安装
curl -fsSL https://oh-my-skill.com/install.sh | bash -s -- quanxiaoxiao/quan-skills/doc-evidence-compare
/ 已经装过?验证本机副本,不用重装
npx oh-my-skill verify quanxiaoxiao/quan-skills/doc-evidence-compare
安装目标可用 --agent / --scope 或 --to 明确指定;省略时只会在唯一已存在的 agent 目录上自动选择,零命中或多命中会停止并提示。content_hash 缺失或不一致均拒装。
0GitHub stars
~758最小装载
~1.8K含声明引用
~1.8K文本包总量
索引托管

怎么用

技能原文 SKILL.md作者撰写 · MIT · f6f857f

Doc Evidence Compare

Search documentation and code together, then answer with evidence instead of a freeform guess.

Use This Skill

Use this skill when the task involves:

  • checking whether a business rule, relationship, or behavior is documented
  • comparing docs and code for consistency
  • locating where a topic is implemented in code
  • checking whether documented behavior is actually implemented
  • finding missing documentation or documentation drift
  • questions such as 有没有写, 代码里在哪, 是否一致, 文档和实现是否对得上
When NOT to use

Do not use this skill when:

  • the task is only to summarize a single document the user already provided
  • the task is only to edit documentation without needing repo-wide evidence gathering
  • the task is pure code debugging with no documentation comparison angle
Pre-Read Order

Read the target repository first:

  1. AGENTS.md if present
  2. README.md or repository index docs
  3. docs/opencode-context.md if present
  4. obvious architecture, API, or glossary docs related to the topic

Then load bundled guidance from this skill pack:

  1. references/search-workflow.md
  2. references/evidence-matrix.md
Workflow
  1. Read repo context before searching the topic. Start with AGENTS.md, README.md, and any obvious architecture or context docs so term matching happens in the right domain.
  1. Expand the query into search aliases. Include Chinese and English terms, synonyms, entity names, field names, API names, and likely implementation identifiers before running searches.
  1. Search docs first and code second, but do not stop after docs. Use rg against docs/, README*, AGENTS.md, openapi*, schema*, then inspect src/, app/, packages/, services/, and tests/ when relevant.
  1. Collect evidence, not just matches. For each claim, capture the file references and the specific meaning of each hit. Distinguish direct evidence from inference.
  1. Classify each claim. Use exactly one status per claim: documented_and_implemented, documented_only, implemented_only, contradictory, or not_found.
  1. Output the fixed evidence matrix before any recommendation. Keep the final answer in the matrix format defined in references/evidence-matrix.md with columns Claim / 问题点, Doc Evidence, Code Evidence, Status, Gap / Notes, and Confidence, then add a short conclusion and recommendation block.
Reference Guide

Load these only when needed:

  • references/search-workflow.md for the exact search order, alias expansion, and status rules
  • references/evidence-matrix.md for the required response template and evidence rules
Verification

Before finalizing an answer, verify:

  1. both docs and code were searched unless the user explicitly limited scope
  2. every positive claim includes file references
  3. not_found means searched but not found, not guessed absent
  4. every inference is labeled as inference
  5. documented-only and implemented-only cases are kept separate
Output Contract

Return results in this order:

  1. Top Summary
  2. Evidence Matrix Use columns Claim / 问题点, Doc Evidence, Code Evidence, Status, Gap / Notes, and Confidence.
  3. Recommendation
按 MIT 许可原样转载,未经改动 · 在 GitHub 查看 →

评论

登录即可评论;带「已验证安装」的,是发布者名下有本店的安装或持有记录。