| name | cli-evals |
| description | Run, inspect, and debug cli-evals suites for live coding-agent CLI evaluations. |
CLI Evals
Use this skill when a user wants to run a live eval suite, compare agents, inspect
reports, rebuild metrics, or diagnose local prerequisites.
Process
- Run
cli-evals doctor first when the target agent or auth state is unclear.
- List scenarios with
cli-evals list --suite <config>.
- Run canaries before full corpora:
cli-evals run --suite <config> --canary --agent <agent> --model <model>.
- Use
--filter <EV-ID> for a failing scenario and add --keep to preserve the
workdir.
- Rebuild summaries from existing reports with
cli-evals rebuild --report <evals/reports/latest.json>.
Do not trigger --all live agent runs without the user explicitly asking; they
cost time, tokens, and may use authenticated agent sessions.