| name | find-catalog-gaps |
| description | Find tools that probably belong in CATALOG.md but aren't there, via referenced-but-not-catalogued detection + a foundational-tools checklist. Use to audit the catalog for missing heavyweight tools. Triggers - "find catalog gaps", "what's missing from the catalog", "audit catalog completeness". |
Find Catalog Gaps
The highest-yield discovery method once star-ranked sweeps saturate: find tools the
catalog already refers to but never entered. This surfaced codex (92K), cline
(63K), MetaGPT (69K), aider (46K) — all missing despite being the largest tools in
their categories. This skill is read-only: it reports candidates; you vet and
add them (with /add-catalog-entry).
Run it
python3 .claude/skills/find-catalog-gaps/find-gaps.py
This combines two signals:
- Detector F (
audit-evals.py --overlaps) — "Overlaps with" tokens naming a
tool that has no entry of its own. Multi-reference tokens are the likeliest gaps;
single-reference ones are usually deliberate external/conceptual peers.
- Foundational checklist — a curated list (in
find-gaps.py) of heavyweight
in-scope tools whose peers are catalogued. Anything MISSING is a candidate.
Vet before adding
A candidate is a lead, not a decision. Add only dev-loop tooling (produces /
reviews / tests / ships code, or agent/MCP infra for that). Skip — and note why —
proprietary-only tools, model-serving infra (ollama, vllm), chat UIs (open-webui),
and general business automation. Confirm the peer-consistency argument: if a
candidate's peers are catalogued, its absence is a real inconsistency.
Extend the checklist
find-gaps.py's CHECKLIST dict is meant to grow as the ecosystem moves — add new
heavyweight tools under the right group. Keep names matching how they'd appear in a
[name]( link. A clean run prints "every checklist tool has a catalog entry".
Related
/add-catalog-entry — add a vetted candidate (handles count propagation + audit).
- Detector F lives in
audit-evals.py; this skill wraps it with the checklist.