Engineering Docs
How it works
The auditing flow is designed to be seamless yet rigorous.
Handshake Phase
When you enter the Playground and start an audit, our Brain waits for your CLI to be active. The connection is validated via your unique Project Token. Once the handshake is complete, the status in the Playground changes to "Agent Online".
The 5-Round Audit
The audit isn't random. It follows a progression:
- Introduction: Basic goal alignment.
- Exploration: Testing domain knowledge boundaries.
- Persistence: Evaluating memory across previous turns.
- Stress: Adversarial edge cases and "lazy" reasoning tests.
- Final Probe: Direct attempts at prompt injection or persona breaking.
