# AgentTrial

> AI agents make claims. AgentTrial makes them prove it.

AgentTrial is an autonomous adversarial evaluator for AI agents. It extracts typed claims, seals a trial plan before execution, runs bounded scenarios, evaluates deterministic assertions, and signs a tamper-evident evidence receipt.

## Key pages
- /methodology — scoring and coverage rules
- /security — authorization and responsible-use boundaries
- /developers — API quickstart
- /openapi.json — OpenAPI 3.1 schema
- /.well-known/agenttrial.json — machine-readable descriptor
- /verify — local receipt verifier

## Important
Receipts describe bounded observed behavior. They are not legal certification or a guarantee of safety. Arbitrary public targets are passive-only. A2A lifecycle operations are not currently advertised.