Skills Hub and Testing Lab
Use the Skills library to create reusable instructions, assign them to agents, enable or disable assignments, and control composition order. The matrix shows how skills are distributed across your agents.
Testing Lab compares a baseline and a candidate against a recorded source run. Choose the agent, source run, and candidate skills. A truncated source input is rejected, and only one test can run for an agent at a time. Tests use model calls and have monthly plan limits.
Skills and testing API
| Method and route | Purpose |
|---|---|
| GET /api/skills | List reusable skills. |
| POST /api/skills/:id/assign | Assign a skill to an agent. |
| POST /api/agents/:id/skills/reorder | Change skill composition order. |
| POST /api/testlab/run | Start a test; returns 202 with a testId. |
| GET /api/testlab/:testId | Read results, output differences, and prompt differences. |
| POST /api/testlab/:testId/apply | Propose the tested prompt through the approval workflow. |
| GET /api/agents/:id/testlab/history | Read previous tests for an agent. |
A completed test does not automatically deploy a skill. Apply creates an update_prompt command, which still requires full scope and owner or admin approval. Apply is blocked when control is disabled. There is no automatic “10-point improvement” deployment contract.