Agent harness

An agent harness connects an application's request to declared tools and handles the result. Siliqun Rotor can make a bounded decision, run configured Z-Script at supported hooks, and record a trace. The consuming application still owns credentials, authorization, tool execution, and user-facing responses.

flowchart LR
    Request[Application request] --> Pre[Configured pre script]
    Pre --> Decide[Engine decision]
    Decide --> Post[Configured post script]
    Post --> Trace[Returned result and trace]
    Trace --> Caller[Caller validates permissions and executes tool]

Scripts and fuel

Supported script endpoints accept an optional fuel value and return output, step count, and script hash when successful. Scripts can also be attached to supported request hooks. Review the Z-Script reference for the exact contract and supported locations. A script is a programmable check or transformation; it is not an automatic prompt-injection filter, PII scrubber, or general security boundary.

Keep the action with the application

For a decided tool route, inspect the tool and arguments, apply the application's own authorization checks, then execute it. Handle ask, unresolved, and refused without dispatching an action. A trace records the request path and available result fields; it does not prove the action was safe, authorized, or correct.