agentsclimarketplace

Wiki lint

Skill Baikodis/claude-knowledge-base-skills/skills/wiki-lint

Claude Code skills that turn research and raw data into an LLM-ready knowledge base: source-discovery → corpus-ingest → wiki-build → wiki-lint (+ batch-analyze). Installable plugin marketplace.

Install
npx -y skills add Baikodis/claude-knowledge-base-skills --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

  • 0 stars0 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

Health-check for a Karpathy wiki. Checks broken links, INDEX intent-routes, source map, source notes & semantic layer (digested vs routing-only), duplicates, orphans, stale/caveats, contradictions, corpus drift, and structure hygiene. Use after /wiki-build, after a large corpus ingest, or when the user asks you to "check the wiki/knowledge base".

SKILL.md

6.0 KB, ~1.4k tokens by cl100k_base, as published. Nobody here has run it

/wiki-lint — Karpathy wiki health-check

Purpose

Validate that a markdown wiki is navigable, source-backed, and safe to use as LLM context.

Input:

knowledge/<topic>/INDEX.md
knowledge/<topic>/_sources.md
knowledge/<topic>/_log.md
inventory/              # optional but expected for fresh builds
text_corpus/            # optional corpus backing

When to use

Triggers:

  • "Check the wiki / knowledge base"
  • "Run lint on <topic>"
  • "Are there any duplicates / broken links / orphans?"
  • After /wiki-build
  • After a big /corpus-ingest
  • Before using the wiki as a sales/client deliverable

Do NOT use:

  • on daily memory notes
  • on random non-Karpathy folders without INDEX.md
  • as a replacement for corpus conversion/OCR

Checks

1. INDEX integrity

  • INDEX.md exists
  • all listed node links exist
  • all actual nodes are listed or intentionally excluded
  • "When to read what" routes to real files/anchors
  • subfolders with 3+ nodes have a local INDEX.md

2. Markdown links

  • all local links resolve
  • no stale renamed paths after cleanup
  • no broken relative paths to inventory/ or text_corpus/

3. Source map integrity

  • _sources.md exists
  • source refs used in nodes appear in _sources.md
  • source paths/reports exist where local
  • coverage caveats mention skipped/failed buckets from inventory reports

3a. Source notes & semantic layer

  • source_notes/README.md exists and explains the layer
  • topic nodes cite source notes via [[source_notes/<bucket>/<id>#anchor]] evidence-links, NOT raw text_corpus/ paths
  • every [[source_notes/...]] link resolves to an existing file with the referenced anchor
  • significant useful_text sources from inventory reports either have a source note OR are explicitly listed as deferred in caveats
  • routing-only files (_pdf_sources.md, _image_sources.md) are not cited from topic nodes as the route to comprehension — they may exist for audit but should be flagged as audit-only
  • contradictions.md exists (even empty); cross-source disagreements are not silently resolved in topic nodes
  • topic nodes do not contain phrases like "see these PDFs", "list of routed sources", "full extraction map → PDF" — that is shallow integration

4. Corpus drift

Compare the wiki/source map against the project corpus:

  • text_corpus/ files not referenced anywhere
  • inventory/*report* newer than the wiki _log.md
  • conversion reports mention failures not reflected in caveats
  • raw buckets not represented in the sources map

5. Duplicates

  • similar node names
  • repeated sections across nodes
  • duplicate source listings
  • same topic split across multiple nodes without a cross-reference

6. Orphans

  • node exists but is not in INDEX.md
  • source/output exists but is not in _sources.md or _pdf_sources.md
  • wiki file has no inbound/outbound references

7. Stale and caveats

  • old updated: dates
  • recency markers older than 12 months for unstable claims
  • caveat says "not parsed" but the corpus now has extracted outputs
  • known failed buckets not listed in caveats

8. Contradictions

LLM-assisted pass when needed:

  • conflicting numbers/dates/rules
  • opposite recommendations
  • incompatible source interpretations

Contradictions require a user decision; do not auto-fix them.

9. Structure hygiene

Project-level sanity when running after ingest/build:

  • raw files under raw/
  • extracted text under text_corpus/
  • runtime caches under _runtime/
  • accidental/legacy dirs under _archive/
  • logs grouped by domain
  • README explains the canonical structure

Phase 0 — Scope

Determine the lint target:

  • /wiki-lint <topic> → one wiki
  • /wiki-lint → active project wiki
  • /wiki-lint --all → all project wikis

If multiple candidates exist, ask once.

Phase 1 — Run deterministic checks first

Use file-system checks before LLM reasoning:

  • list files
  • parse markdown links
  • grep source refs
  • compare timestamps and report paths
  • count orphans/broken links

For large contradiction checks, /batch-analyze can be used internally as an extract/verify utility.

Phase 2 — Report

Report categories:

INDEX integrity: ok / issues (intent-routes? topic-listing-only?)
Broken links: N
Source map: ok / issues
Source notes & semantic layer: ok / shallow / missing
  - source notes coverage: M of K useful_text sources
  - evidence-link integrity: N broken / N raw-path citations
  - audit-only files cited as memory: N
Contradictions log: present / missing / silently-resolved
Corpus drift: N
Duplicates: N
Orphans: N
Stale/caveats: N
Contradictions found by LLM pass: N
Structure hygiene: ok / issues

Separate:

  • auto-fixable
  • needs a user decision
  • informational caveats

Phase 3 — Fix policy

Auto-fixable after confirmation:

  • broken local links caused by renames
  • missing INDEX entries for obvious nodes
  • stale caveat wording that contradicts current reports
  • source map path updates after cleanup
  • README/log updates

Never auto-fix without a decision:

  • merge/delete duplicate content
  • resolve contradictions
  • delete raw/source files
  • remove wiki nodes

Prefer moving questionable files to _archive/ or _orphans/, not deleting.

Phase 4 — Log

Append to the wiki _log.md:

- YYYY-MM-DD: wiki-lint — broken links N, source issues N, orphans N, fixed M, deferred K.

Also append to the daily note for meaningful cleanup.

Anti-patterns

  • Do not delete files as a lint fix.
  • Do not report "ok" if source reports show skipped/failed buckets.
  • Do not use the LLM contradiction pass before deterministic link/source checks.
  • Do not silently rewrite node content while "linting".
  • Do not let lint become ingest or wiki-build.

What ships with it

Read from the repository

Just SKILL.md. No reference files, no scripts.

Gives 0 of the 12 instructions most quality gates skills give in ~1.4k tokens

Counted across 1,195 of the 2,094 authors here whose files we hold, read 2026-08-07

  • read the output and check the exit codein 54 of 1195, across 14 files
  • verify requirements using a line-by-line checklistin 53 of 1195, across 12 files
  • identify the verification command proving the claimin 51 of 1195, across 12 files
  • run the full verification commandin 50 of 1195, across 11 files
  • verify output confirms the claimin 49 of 1195, across 12 files
  • check version control diff after agent delegationin 46 of 1195, across 6 files
  • state claim with evidencein 44 of 1195, across 4 files
  • run the test suitein 33 of 1195, across 26 files
  • keep state in memory by defaultin 27 of 1195, across 6 files
  • make prototype runnable with one commandin 26 of 1195, across 5 files
  • produce a verification reportin 25 of 1195, across 14 files
  • detect the package manager from lockfilesin 24 of 1195, across 5 files

Said here and by no other author read

  • list broken links and missing index entries
  • flag shallow integration of source notes
  • check for corpus drift against inventory reports
  • find duplicate or orphaned files
  • check for stale dates and failed bucket caveats
  • separate findings into auto-fixable and needs-decision

Grouped from the skills themselves: near-identical wordings counted once, and counted by distinct author, so one author publishing three of these counts once. Length counted with cl100k_base; the agent that loads this file may tokenize it differently.

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.