| name | ecc-cmd-eval |
| description | Command skill for 'eval'. Use when users request evaluation, benchmarking, quality scoring, and regression checks. |
ecc-cmd-eval
Run evals and produce scorecard + failure analysis.
Workflow
- Identify eval set (fixtures, scenarios, test prompts).
- Run
scripts/run-eval.sh [path] if available.
- Capture metrics: pass rate, failures, latency/cost when available.
- Classify failures by root cause.
- Provide top remediation actions and rerun command.
Output Template
EVAL: PASS|FAIL
Cases: X/Y passed
Pass Rate: Z%
Top Failures:
- ...
Actions:
1) ...
2) ...
Rules
- Keep eval dataset version/path in report.
- Distinguish flaky from deterministic failures.
- Include reproducible command lines.
ECC Parity Notes
- This skill is adapted from ECC intent and translated to OpenClaw-native tool workflows.
- Prefer reproducible, tool-backed execution over harness-specific slash semantics.
- Keep outputs evidence-based and verification-first.
OpenClaw Constraints
- If upstream flow assumes unavailable hooks/tools, use documented fallback scripts in
openclaw-ecc/scripts/.
- Mark non-applicable checks as
SKIP rather than hard-failing when environment support is absent.