Lab (⌘7) lets you measure an agent, a skill or a workflow on cases from your own code.
- Write cases — the agent reads this project and drafts cases; accept or discard each under Waiting for you. Evaluate on an agent, a skill or a workflow starts a set for it.
- Columns — every accepted case runs in every column. A column is the agent with one thing changed, such as the model, so you can see whether the change made its work better. A skill set compares the work with and without the skill; a workflow set compares whole workflows on the same cases and starting files.
- Run — measures every case in every column. A mark reads passed, did not pass or not measured; click it for the details and the measured run.
- What did not pass — Propose a fix suggests new instructions for the agent, which you Apply or Discard.