agentHost: Fix sessions stuck in progress after subagent completion

[Benjamin Christopher Simmonds (benibenj)]

Summary

Fix completed Copilot Agent Host sessions remaining In Progress in the Agents window after a subagent follow-up finishes.

The observed SDK sequence supplies the subagent ID in hook.start.data.input.agentId for a subagentStop hook, but omits it from the hook.end envelope. Our handler only completed subagents from agentStop events with an envelope agent ID, leaving the follow-up turn active. #333999 made that stale child status visible on the parent session row.

Fix

• Remember the subagent ID by hookInvocationId until the matching stop hook ends.
• Reuse the existing subagent completion path and remove the correlation immediately.
• Keep the sessions-list behavior unchanged; no polling, timeouts, or forced UI status.
• Cover follow-up completion without an envelope agent ID, overlapping hooks, and duplicate completions that must not close a later turn.

Validation

• Both new regression tests failed before the fix.
• All 430 tests in copilotAgentSession.test.ts passed afterward.
• npm run transpile-client
• Targeted ESLint and staged hygiene (npm run precommit)
• git diff --check

Release cherry-pick requested via ~release-cherry-pick.

[vs-code-engineering]

This PR will be automatically cherry-picked to release/1.136 when merged.

[Copilot]

Copilot review overview

🟢 Approval recommended

The focused implementation handles the reported event sequence and includes appropriate regression coverage.

Review tier: Balanced
Findings: None

What changed in this PR

Fixes stale In Progress state by correlating subagent stop hook events.

Changes:

• Tracks subagent IDs across matching stop hooks.
• Completes resumed subagent turns without envelope IDs.
• Adds regression coverage for overlapping and duplicate events.

File	Description
copilotAgentSession.ts	Correlates stop hooks and completes subagent turns.
copilotAgentSession.test.ts	Tests missing IDs, overlapping hooks, and duplicates.

---

💡 Add a code-review agent skill or configure MCP servers for context-aware, tailored reviews. Learn more in the docs.