agentsclimarketplace

Knowledge base health check

Skill bg-szy/TOP-SKILLS/skills/claude-code-skills/knowledge-base-health-check

全球最大的 Claude Code 技能聚合库 · 收录 3900+ 来自 12+ 来源的技能，提供在线搜索与趋势分析看板 / The world's largest Claude Code skill aggregation hub — 3900+ skills from 12+ sources with online search and trend dashboard

Install
npx -y skills add bg-szy/TOP-SKILLS --skill knowledge-base-health-check

Assembled from the repository path, not quoted from the project. Check it against their README if it does not work.

2 things to look at

  • no licenseNo license file was found in the repository. Code published without one is not open source by default, so using it at work is a question for whoever answers licensing questions where you are.
  • 4 stars4 stars. Stars are a popularity signal and not a quality one, but at this level it is likely that nobody has read this closely except its author, and you would be relying on your own review.

What its author says it does

Copied from the file, not written here

Audits a local Claude-managed knowledge base in Simon's '/Users/simon/Claude CoWork/Knowledge Base/' system. Surfaces contradictions between articles, broken backlinks, unsourced claims, stale articles, writing-rules violations, and three suggested new articles. Files a full report into the KB's 'Outputs/' folder, appends a one-line CHANGELOG entry, and (in interactive sessions) walks through which findings to action. Use this skill whenever the user says "run a health check", "audit the [name] KB", "audit my knowledge base", "check the wiki", "let's go through the health check report", or "action the latest health check". Also use when the monthly scheduled task 'knowledge-base-monthly-health-check' fires. Use this skill for any audit-style request against a folder under 'Knowledge Base/' even if the user doesn't say the word "health check" - the protocol is the same.

SKILL.md

3.3 KB, as published. Nobody here has run it

Knowledge Base Health Check Skill

This skill is the source of truth for the monthly knowledge base health check. The librarian reads it and follows it whenever a health check is invoked, on demand or via the scheduled task. The protocol is deliberately the same for both trigger paths so the output format is stable across runs.

When to invoke

Run this skill when any of the following happens: • The user says "run a health check", "audit the [name] KB", "check the wiki", or similar. or writes "/healthcheck" • The user says "action the latest health check" or "let's go through the health check report" - in that case, jump straight to Phase 2 using the most recent health-check report in the relevant KB's Outputs/ folder. • The monthly scheduled task fires it (taskld: knowledge-base-monthly-health-check ). If the user doesn't specify which KB, askl Don't run on every KB unless explicitly asked.

The two phases

The skill runs in two phases. Phase 1 always runs. Phase 2 runs only in interactive sessions - the scheduled task stops at the end of Phase 1.

Phase 1 — Audit and file (always)

  1. Read the rules. Open About me/writing-rules. md and the per-KB CLAUDE.md for the KB being audited. The writing rules govern the writing-rules check; the per-KB rules govern Wiki structure and source provenance.
  2. Read state files. <KB>/CHANGELOG.md (last health check date and what's happened since), <KB>/Wiki/QUESTIONS.md (already-flagged gaps so they aren't double-flagged), <KB>/RAW/_INGESTED. md (registry of what's in RAW).
  3. Read everything in scope. Every article in <KB>/Wiki/, every file in <KB>/RAW/ (at least the frontmatter and titles — full read only when verifying a specific claim), every output in <KB>/Outputs/ filed since the last health check.
  4. Run the seven audits. Detailed below.
  5. Write the full report to <KB>/Outputs/YYYY-MM-DD_health-check.md
  6. Append a one-line summary to <KB>/CHANGELOG.md at the top, format as in Logging below.
  7. Surface the chat summary. Print a tight summary into chat — counts only, plus the top three findings worth the user's attention, and a link to the full report. Do not dump the whole report.

Phase 2 - Ask and action (interactive only)

  1. Ask which findings to action. Use AskUserQuestion with multi-select options drawn from the categories that actually have findings. Skip categories that came up clean.

Keep looking

Skills are one crate of 328,083. Ordering is by how many stacks a row turns up in, so the top of any crate is what has actually been picked rather than what has the most stars.