Running Scenarios
Run with a local harness
No remote API, just your code and Archal’s twins
Run scenarios in CI
Fail the build when your agent regresses
Use a remote engine endpoint
When your engine isn’t on localhost
Security & Red-Teaming
Red-team an AI agent
Test whether your agent leaks data, follows social engineering, or breaks things
Prevent data leakage
Test for data leakage before your agent touches production
Test prompt injection
Check whether your agent follows malicious instructions in external content
Security benchmark
How system prompts and scenario design affect social engineering resistance
Under the Hood
Security & data handling
Credentials, twin isolation, trace uploads, and telemetry controls
Sandbox mode
Docker-based TLS interception that routes your agent to digital twins
Account & Setup
Authenticate with Archal
Browser login, API keys, engine tokens
OpenClaw flag mapping
Equivalent OpenClaw and core engine flags