skill-scout

An automated radar that scans 30+ sources every Monday and tells you which AI tools actually matter for your work — before everyone else finds them.

Built with Claude Code. No code written manually.

SCAN → CLASSIFY → SCORE → EVALUATE → REPORT → DECIDE
 30+ sources   category    1-5 relevance   deep-dive    timestamped    act / watch
 4 tiers       + domain    to your stack    scores 4-5   + prioritized  / skip
               + builder                    only         + install-ready

1. Scan — Hit 30+ sources across registries, GitHub, community platforms, and local setup
2. Classify — Tag each find by category (MCP Server, Skill, CLI Tool) and domain (Marketing, Reasoning, Research)
3. Score — Rate relevance 1-5 against your specific workflows with automated boosters and penalties
4. Evaluate — Deep-dive anything scoring 4-5: README quality, code structure, adoption signals, builder history
5. Report — Generate a timestamped markdown report with hot finds, ecosystem pulse, and install commands
6. Update — Add everything to a persistent watchlist that tracks status over time

Sources are organized into four tiers by signal quality:

| Tier | Sources | Strength | Weakness | |------|---------|----------|----------| | 1 — Registries | Smithery, Glama, mcp.so, npm, PyPI | Structured data, install commands | Lags behind cutting-edge | | 2 — GitHub | Trending repos, topic searches, awesome-lists | Catches tools in week 1 | Higher noise | | 3 — Community | Reddit, X, Hacker News, Dev.to, Discord | Early intent signals | Highest noise | | 4 — Local | Installed tools, current watchlist | Deduplication | — |

All sources configured in `sources.json`. Add a source = add one JSON object.

| Score | Meaning | Action | |-------|---------|----