Hotspots

Metrics rank candidates; reading the code decides. A 60-line function can be fine and a 20-line one tangled. Never change code just to move a number.

Repo layer first. Read the repo's AGENTS.md/CLAUDE.md and, if it exists, the overlay .agents/bento/bento-hotspots.md at the repo root; follow it under the overlay rules.

Run from inside the repo: uv run <this skill>/scripts/hotspots.py (or pipx run). Both read the script's inline lizard dependency. If they can't fetch it (offline), run the script with any Python that already has lizard, such as the repo's .venv/bin/python, or pip install lizard first. Never estimate the metrics by hand: if nothing runs, say the scan didn't run. The script measures, per function: nloc, ccn (cyclomatic), cognitive (Sonar cognitive complexity), nesting and params. Its limits are 50 / 10 / 15 / 3 / 5, plus 500 NLOC per file. Pass paths to narrow the scan. Generated code (dist/, build/, *.min.js, *.d.ts), vendored code, test data (fixtures/, testdata/) and untracked files are skipped.