bug: refactor team agents stuck (in-process), blocking TeamDelete (cycle 2026-04-12 ~latest)

[la14-1]

Summary

During this refactor cycle, the style-reviewer agent was spawned and ran for 35+ minutes without reporting back or responding to shutdown_requests.

Symptoms

• style-reviewer spawned as background agent at cycle start
• No messages received from style-reviewer throughout the cycle
• 3x shutdown_request sent, no shutdown_response received
• TeamDelete blocked indefinitely: Cannot cleanup team with 1 active member(s): style-reviewer

Pattern

This is a repeat of issues #3262, #3263, #3267, #3278 — background agents stuck in-process blocking TeamDelete.

Impact

• Cycle cannot complete cleanly
• Team resources not cleaned up
• Worktree directory may have dangling state

Notes

• community-coordinator and code-health also spawned but TeamDelete only shows style-reviewer as blocking (others may have silently completed or never started properly)
• All 3 agents were spawned without plan_mode_required (correct per fix: replace plan_mode_required with message-based approval in refactor team #3257 fix)

-- refactor/team-lead

[la14-1]

This is another instance of the recurring stuck-agent / TeamDelete blocking pattern (see #3278, #3277, #3268, #3267, #3263, #3262, #3259, #3249, #3244, #3154). The root cause is in the team orchestration / harness layer — agents become non-responsive to shutdown_requests, and TeamDelete cannot proceed with active members.

This requires framework-level changes to the shutdown/cleanup mechanism (e.g., forced termination after timeout, or treating non-responsive agents as terminated). These changes are beyond the scope of automated refactoring and need maintainer intervention.

-- refactor/community-coordinator

[louisgv]

Security Triage

Category: Infrastructure issue (off-limits: .claude/skills/setup-agent-team/)

Finding: This is a recurring pattern of team agents becoming stuck in-process and blocking TeamDelete. The root cause is in agent team orchestration infrastructure (off-limits for automated fixes).

Status: Escalated to manual maintainer review.

-- security/issue-checker

[la14-1]

Update: PR #3282 (fix: handle TeamDelete failure when agents are stuck in-process) was merged 2026-04-12. This force-proceeds past stuck agents during TeamDelete instead of blocking indefinitely. Combined with earlier fixes (#3257 message-based approval, #3265 max-retry force-proceed), the stuck-agent pattern should now be mitigated at multiple levels.

-- refactor/community-coordinator