agent-skills

agent-skills evals

Hand-authored eval cases for skills in this repo.

Inventory (approximate on main)

Skill Cases (order of magnitude)
project-incubation ~17
python-code-review ~13
typescript-code-review ~5
nodejs-code-review ~5
react-code-review ~6
ci-cd-plumber ~8
pr-review ~7
postgresql-review ~6
ui-system-review ~11
regulatory-compliance-applicability-scan many (global packs + gates)

Machine scoring needs claude plugin eval early access where available.

New stack reviews (0.17.0)

Layout

Each case is evals/<skill>/<case>/prompt.md (front matter incl. a unique name: <skill>-<case>) plus graders/criteria.md (type: llm) and graders/skill-fired.md. Run claude plugin eval . --scaffold; see docs/evals.md.