Paste the prompts you actually typed this week — one per line — and see which skill each one should have used. Everything runs in your browser. Turn on the typed router to replace keyword matching with one calibrated decision per prompt.
Set the worker URL of a PM Skills MCP deployment with a JEV_API_KEY secret (see mcp-remote). Prompts are sent to that worker and to the decision model only when this is set; nothing is stored.
| Prompt | Skill | Confidence | Tier |
|---|
The same router runs as a Claude Code hook, an npm package, and POST /route on the worker. Prefer to browse? describe your task · the catalog.