Lab

Lab tests a platform agent by playing simulated people against it, judging the answers and reporting the result, billed to the workspace.

Building blocks

Part	What it is
Environment	Test cases (simulated people who write to the agent), the files they hold, and the checks the answers must pass
Gym	Plays an environment against an agent to measure it, or to improve its configuration round by round
Diagnosis	Reads conversations the agent already had and says what fails, why, and where to fix it
Certification	Measures an agent on test cases against a pass bar; a person approves or rejects the result in the web app

The agent must be a platform agent; publish one from files with Create and configure an agent.

Every request sends X-Space-Slug and one credential; see Authentication and API keys. An API key needs the Lab scope.

Task	Auth
Environments, gym sessions and diagnoses	Key or bearer
Launch, read and share a certification; change its reviewer	Key or bearer
An agent's chats, through `/api/v1/diagnostics/agents/{agent_id}/chats/`	Key or bearer
Approve or reject a certification	Bearer only

A key without Lab gets 403 {"detail": {"code": "API_KEY_SCOPE_MISSING", "scope": "lab"}}.

Field	On	Meaning
`space_seq`	Diagnosis runs, gym sessions, certifications	The number the web app shows (#3), counted per workspace. `id` stays the key in every path.
`kit_name`	Diagnosis runs, gym sessions	The environment's name, also when the caller cannot read that environment.

All paths start with /api/v1/diagnostics.

Report	Conversation IDs	Read them with
Gym	`GET /gym/{session_id}/conversations/` → `[].chat_id`	`GET /agents/{fork_agent_id}/chats/{chat_id}/`, with `fork_agent_id` from `GET /gym/{session_id}/`
Diagnosis	`trace_ids`	`GET /agents/{agent_id}/chats/{chat_id}/`
Certification	`report.cases[].executions[].chat_id`	`GET /certifications/{certification_id}/chats/{chat_id}/`

GET /agents/{agent_id}/chats/ lists the agent's chats the caller can open. With a key, these are the only routes that read an agent's chats.