Summary

Add an extension hook (e.g. post_subagent_complete) that fires after a spawned subagent finishes and before the completion announcement is sent to the parent session.

Use Case

Automatically generate structured trajectory files (task → decisions → retrospective) after significant subagent work completes. Currently only possible as a behavioral LLM rule — non-deterministic.

A programmatic hook would allow:

• Inspecting the subagent session history and completion output
• Running arbitrary logic (summarize into trajectory, update search index)
• Modifying/enriching the completion announcement
• Filtering noisy completions

Proposed API

openclaw.on("subagent_complete", async (event, ctx) => {
  // event.sessionKey, event.label, event.agentId
  // event.result, event.duration, event.usage
  await saveTrajectory(event);
  return { announcement: enrichedSummary };
});

Context

Similar to PI framework agent_end event handler. The completion announcement already flows through the gateway — this hook would expose it as an extension point. Useful for building memory/learning systems on top of OpenClaw.