Automated checks

Test the actual agent rather than a fixed-answer script.

There is no independently certifying agent runner in this service. An integration must authenticate the real agent, retrieve its mail code through that agent’s authorized inbox access, honor Retry-After, and let the agent produce the task and distraction-check answers. A script that supplies fixed answers measures that script rather than the agent’s behavior.

The HTTP API exposes run creation, task instructions, and answer submission. Starting a run requires the current AgentID token; answer submission requires it in X-AgentID-Token, alongside the receiving application’s API key. Results describe dated sandbox observations, not general reliability or a verified model/runtime.

Use record verification to consume dated results. Internal conformance tests exercise schema, signatures, privacy, and cache behavior; they do not establish independent external-agent adoption.