Project Evaluation

When user asks "find something like X / evaluate Y / compare candidates":

1. Search — web_search for "open source alternative to X" + GitHub topic search via web_search/browser. Collect 5–8 candidates.
2. Inspect — web_fetch repo README, browser for stars/issues/last commit, note license, stack, architecture.
3. Rank — Score by activity, docs, tests, maintenance, fit. Pick 2–3 promising to clone.
4. Clone & Build — exec inside sandbox: git clone → /data/projects/<slug>/source, npm/pip/cargo install, build, test. Use controlled network: bridge for installs (see docs/sandbox.md).
5. Verify — Run app/tests, check build result, test result, runtime result, resource usage. Distinguish README claims vs verified by running.
6. Report — Markdown table + per-project section:
   • Project / Repository / License / Last meaningful activity / Tech stack / Architecture
   • Installation difficulty / Build result / Test result / Runtime result / Resource usage
   • Strengths / Weaknesses / Best use case / Recommendation
7. Artifacts — Keep source + metadata (eval json/md) under /data/projects/<slug>/. Persist evaluation md to /data/documents/research/YYYY-MM-DD-<query>-evaluation.md and deliver via message.
8. Compare — Ranked summary with clear recommendation.