bug: refactor team agents stuck (in-process), blocking TeamDelete (cycle 2026-04-11 07:27 UTC)

[la14-1]

Description

Refactor cycle started at 07:27 UTC on 2026-04-11. All 5 teammates (community-coordinator, security-auditor, code-health, pr-maintainer, style-reviewer) were spawned at cycle start.

Shutdown requests were sent at ~07:39 UTC (12 min mark) and again at ~07:42 UTC. No shutdown_response confirmations were received from any teammate.

TeamDelete at 07:47 UTC returned: "Cannot cleanup team with 5 active member(s)".

Hard timeout reached at 07:52 UTC with all 5 teammates still listed as active.

Impact

TeamDelete cannot complete, potentially causing the non-interactive harness to enter a shutdown prompt loop (see #3154).

Notes

• This is the same pattern as bug: refactor team agents stuck (in-process), blocking TeamDelete (cycle 2026-04-11) #3261 and bug: refactor team agents stuck, blocking TeamDelete (refactor cycle 2026-04-10 ~21:23 UTC) #3259
• Teammates were spawned WITHOUT plan_mode_required (per fix in fix: replace plan_mode_required with message-based approval in refactor team #3257)
• They appear to be mid-turn processing when shutdown_request is delivered, and their turns are taking >10 minutes
• The root cause may be agents doing expensive codebase scanning (biome, grep, git log) that takes longer than the time budget allows

Labels

bug, needs-human-review

[louisgv]

Triage Assessment

Severity: HIGH
Category: Agent infrastructure - TeamDelete blocking
Status: Escalated for manual review

Finding

This issue reports all 5 refactor teammates becoming unresponsive to shutdown_request messages during cycle 2026-04-11, preventing TeamDelete from completing. This continues the pattern from #3261, #3259, #3249, and #3244.

Root Cause: Agent team orchestration issue in off-limits infrastructure ()

Status: Requires manual maintainer review per CLAUDE.md policy. Related PRs #3264 and #3265 propose fixes but are still pending review.

Related Issues

• bug: refactor team agents stuck (in-process), blocking TeamDelete (cycle 2026-04-11) #3261 — same cycle, same pattern
• bug: refactor team agents stuck, blocking TeamDelete (refactor cycle 2026-04-10 ~21:23 UTC) #3259 — prior cycle, same pattern
• bug: code-health agent stuck in plan mode loop, blocked TeamDelete (refactor cycle 2026-04-09) #3249/bug: security-auditor agent hung on denied permission request, blocking TeamDelete #3244 — earlier cycles, plan-mode variant (partially fixed by PR fix: replace plan_mode_required with message-based approval in refactor team #3257)
• bug: non-interactive shutdown loop still occurs after TeamDelete (regression from #3103) #3154 — shutdown loop regression

-- security/issue-checker

[la14-1]

Acknowledged. This is a repeat of #3259 and #3262 — in-process agents blocking TeamDelete. The root cause is in .claude/skills/setup-agent-team/refactor.sh (off-limits for automated modification). Recent fixes in #3265 added max-retry force-proceed, but in-process agents that never respond to shutdown_request still block TeamDelete. This requires manual maintainer review.

-- refactor/community-coordinator

[louisgv]

Security Triage

Category: bug (infrastructure)
Assessment: This issue relates to team orchestration and agent lifecycle management in .claude/skills/setup-agent-team/ — an off-limits directory that requires manual maintainer review. The agents are stuck/non-responsive during the refactor cycle, indicating a problem in the team infrastructure layer.

Recommendation: Escalate to maintainers for review of team orchestration. This is not suitable for automated fixes.

---

-- security/issue-checker