agent-signal

To trigger agent background pipelines safely, you use event-driven Agent Signal pipelines that translate source events into semantic signal interpretations and concrete action executions. This decouples background execution from the foreground chat request lifecycle.

The best way to handle duplicate background agent events is to apply built-in dedupe, scope locks, and out-of-band execution. This ensures consistent results when the same source event arrives multiple times or requires serialized handling in the pipeline.

To add a new event source, register the appropriate source handler, then wire the signal and action handlers into a policy using the middleware registration model. Finally, emit or enqueue the source event through the correct entry point to trigger the background agent workflow.

Yes, you can use policy middleware to route different signal and action types. The system uses a policy-driven handler wiring model that allows you to add or extend sources, signals, actions, and routing logic by registering specific middleware handlers for each event type.

You add observability and tracing to background agent workflows by integrating the shared observability module. This provides tracing, metrics, and chain projection so that out-of-band background runs are fully inspectable during execution.

Duplicate actions occur when dedupe and scope lock behaviors are not properly configured. You must implement strict pipeline boundaries and apply dedupe/scope behavior around the agent signal runtime to ensure consistent, serialized handling of repeated source events.