agentsclimarketplace

Wiki lint

Skill AIwork4me/open-llm-wiki/skills/wiki-lint

Claude Code skill bundle for building auditable research wikis from papers with independent QA, safe writeback, and validation.

Install
npx -y skills add AIwork4me/open-llm-wiki --skill wiki-lint

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

One thing to look at

  • 2 stars2 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

Run a health check for an open-llm-wiki vault. Use when the user asks to lint, audit, validate, or periodically check wiki quality. The default mode is report-only: inspect schema compliance, QA coverage, links, stale claims, contradictions, and log health. Modify files only when the user explicitly requests fix mode or a scheduled automation has been configured to allow safe maintenance writes.

The file declares its own license as MIT. That is the author’s claim about this one file, and it is not the same thing as the license GitHub reports for the repository, which is listed with the other numbers below.

SKILL.md

4.4 KB, as published. Nobody here has run it

Wiki Lint

Audit an open-llm-wiki vault for structural, traceability, and maintenance problems. Default to report-only.

Runtime Tool

Use the deterministic linter whenever available. For semantic-growth audits, also use claim/QA scripts when available:

uv run python scripts/wiki_lint.py "<vault>" --fail-on p1
uv run python scripts/wiki_discover_sources.py "<vault>"
uv run python scripts/wiki_claims.py "<vault>"
uv run python scripts/wiki_normalize_metrics.py "<vault>" --in-place
uv run python scripts/wiki_semantic_qa.py "<vault>" --write-report --fail-on p1
uv run python scripts/wiki_contradictions.py "<vault>" --write-report
uv run python scripts/wiki_science_review.py "<vault>" --queue --write-report

The script checks structure, frontmatter, QA gates, contradiction reports, links, index coverage, stale claims, claim graph availability, and log format. Read reports before doing any manual inspection.

Safety Boundary

  • Read-only by default.
  • Fix mode requires explicit user approval or an automation prompt that clearly authorizes maintenance writes.
  • Never edit files in raw/.
  • Never rewrite QA reports; they are append-only audit records.
  • Do not rewrite claims/claims.jsonl unless the user asked for semantic refresh or maintenance fix mode.
  • Queue files under _state/ are state, not evidence. They may be updated by explicit maintenance or scheduled growth runs.
  • When fixing, show a write plan first and keep edits targeted.

Checks

1. Structure

  • required directories exist: raw/, sources/, concepts/, drafts/, qa-reports/, claims/, _state/, templates/
  • required root files exist: SCHEMA.md, index.md, log.md
  • source page filenames match LLM-NNNN.md

2. Frontmatter

  • source pages include id, title, status, created, updated, source, and tags
  • stable source pages live in sources/
  • draft source pages live in drafts/
  • concept pages include id, title, created, and updated
  • IDs are unique and sequential enough to audit

3. QA Coverage

  • every stable source has qa-reports/LLM-NNNN.md
  • every QA report has overall and verdict
  • no stable source has a failing or missing QA gate
  • contradiction reports exist after publish when required by the workflow

4. Links and Index

  • [[LLM-NNNN]] links resolve to source pages
  • [[concept-name]] links resolve to concept pages
  • index.md lists all stable source pages and concept pages
  • orphan pages are reported, not automatically deleted

5. Claim Hygiene

  • flag words such as "latest", "current", and "state of the art" when the page is older than 90 days
  • verify claims/claims.jsonl can be parsed and references stable sources
  • verify _state/source-registry.jsonl and _state/growth-queue.jsonl are valid JSONL
  • inspect science-review queue items before automatic writeback and keep review-required claims out of concept revisions until approved
  • report [CONTRADICTION ...] markers that need follow-up
  • suggest concept pages for topics appearing in three or more sources

6. Log Health

  • log entries should follow: [YYYY-MM-DD HH:MM] action | target | agent | note
  • report entries older than 30 days
  • in fix mode, archive old entries to log-archive/YYYY-MM.md without changing archived history

Output

Return a concise report:

# Wiki Lint Report
- date: YYYY-MM-DD
- mode: report-only|fix
- structure: PASS|FAIL
- frontmatter: PASS|FAIL
- qa: PASS|FAIL
- links: PASS|FAIL
- claim hygiene: PASS|WARN|FAIL
- log: PASS|WARN|FAIL

## Findings
- [P1] path: issue and suggested fix

## Files changed
- none

Use priorities:

  • P0: data corruption, missing stable-source QA, broken schema
  • P1: broken links, unsafe status, duplicate IDs
  • P2: stale claims, missing index rows, orphan pages
  • P3: style or cleanup

Fix Mode

If the user asks to fix issues, address only the reported findings. Re-run wiki_lint.py after edits and report remaining findings.

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.