acp-router

To route ACP harness tasks, the router analyzes plain-language requests and directs them to either the OpenClaw ACP runtime using sessions_spawn or the direct acpx path via exec, ensuring reliable task execution.

The sessions_spawn path manages OpenClaw ACP runtime operations and threading, while the direct acpx path via exec handles immediate execution. The router chooses between them based on the specific intent of the request.

AgentId mapping translates user-harness references into internal agent IDs, ensuring that ACP session management and threading operations are routed to the correct runtime identity for reliable execution.

When ACP backend or adapters are unavailable, the router provides guarded fallbacks and guidance, utilizing thread-spawn recovery and pinned acpx handling to manage the failure and maintain workflow stability.

Yes, you can use plain-language requests to manage multiple ACP harness identities because the router maps these natural language inputs to internal agent IDs and directs them to the appropriate runtime path.

ACP thread-spawn operations require recovery when the standard runtime path encounters issues, prompting the router to initiate thread-spawn recovery and apply pinned acpx handling to restore the workflow.