model-scanlisted
Install: claude install-skill eprouveze/claude-skills
# Model Scan
Query LLM provider APIs, fetch official documentation for specs, and produce a structured
model report. Useful for keeping a project's model-selection table current as providers
ship new models and adjust pricing.
This skill assumes a driver script at `scripts/model-scan.ts` that does the actual API
calls and doc parsing. A minimal reference driver is bundled in `scripts/model-scan.ts` —
extend it with the providers and parsing rules your project needs.
## Approval gate — curated table changes
A project's `CLAUDE.md` "Current AI Model IDs" table is typically a curated selection that
directly affects which models are used in production code. **Any change to the recommended
set requires explicit user approval.** The driver script auto-updates the table from its
`selectRecommendedModels()` list, but when the recommended set should change (new model
added, model removed, provider added/dropped):
1. Show the user what changed and why.
2. Wait for explicit approval before writing to CLAUDE.md.
3. Never silently add or remove providers from the curated table.
## Route by argument
- `/model-scan` (no args) → run the script with `--update-claude-md`, display results,
then review saved docs for any TBD fields.
- `/model-scan list` → run without `--update-claude-md`, display results only.
- `/model-scan fast` → run with `--skip-docs --update-claude-md` (uses cached docs).
### Phase 1: Run the scanner
```bash
cd "$(git rev-parse --show-toplevel)" && npx tsx script