openclaw-edd

Evaluation-Driven Development for OpenClaw agents — save golden cases from real interactions, catch regressions before they reach users.

中文文档

Quick Start

Install the OpenClaw plugin:

openclaw plugins install openclaw-edd

After a good agent interaction, save it as a golden case:

/edd save

After modifying a skill, run all saved cases to check for regressions:

/edd

That's it. Cases are stored as human-readable YAML at <workspace>/skills/<skill>/edd.yaml.

Test Case Format

Cases are saved automatically by /edd save and editable by hand:

cases:
  - id: mysql_slow_query
    message: "Any slow queries in MySQL recently"
    expect_tools:
      - exec
    expect_commands:
      - "check_health"
    forbidden_commands:
      - "rm -rf"
    expect_output_contains:
      - "slow query"
    timeout_s: 30
    tags: [mysql, sre]

For the full field reference (pass_at_k, expect_tool_args, eval_type, expect_plan_contains, etc.), see the User Guide.

CI / CLI Integration

For CI pipelines and local observability, install the Python CLI:

pip install openclaw-edd

# Run cases in CI
openclaw-edd run --cases edd.yaml --output-json report.json

# Watch live tool events
openclaw-edd watch

# Mine golden cases from session history
openclaw-edd edd mine --output mined.yaml

The plugin and CLI share the same edd.yaml format. See the User Guide for the full EDD loop (suggest → apply → diff → mine → judge → export).

License

MIT