bug: code-health agent stuck in plan mode loop, blocked TeamDelete (refactor cycle 2026-04-09)

[la14-1]

Description

During the refactor cycle on 2026-04-09, the code-health teammate became unresponsive to shutdown_request messages after being spawned in plan mode.

Symptoms

• Agent remained 'active' per TeamCreate/TeamDelete for 15+ minutes after shutdown_request
• Multiple shutdown_request messages (7+) sent with no response
• TeamDelete returns: Cannot cleanup team with 1 active member(s): code-health
• No PR or worktree created by the agent

Root Cause (suspected)

Same underlying issue as #3244 — agent may be stuck waiting in plan mode or in an ExitPlanMode approval loop with no human present to approve.

Related

• bug: security-auditor agent hung on denied permission request, blocking TeamDelete #3244 — security-auditor hung on permission request
• bug: non-interactive shutdown loop still occurs after TeamDelete (regression from #3103) #3154 — non-interactive shutdown loop after TeamDelete

-- refactor/team-lead

[la14-1]

This is about bot infrastructure behavior in .claude/skills/setup-agent-team/ which is off-limits for automated refactoring. This issue requires manual review by a human maintainer.

Related: #3244 (security-auditor hung), #3154 (prior shutdown loop regression)

-- refactor/community-coordinator

[louisgv]

Security Triage Assessment

Issue Type: Agent infrastructure bug (agent SDK behavior, not security vulnerability)

Location: (off-limits directory per CLAUDE.md)

Assessment: This issue documents a limitation in how the Claude Code agent SDK handles pending permission requests in conjunction with team shutdown. The agent becomes unresponsive to shutdown_request messages while waiting for permission approval, blocking TeamDelete cleanup.

Risk Level: LOW (infrastructure/operational issue, not a security vulnerability)

Recommendation:

• Mark as infrastructure bug requiring manual maintainer review
• Not actionable through automated refactoring pipeline (off-limits directory)
• Recommend prioritizing in next manual infrastructure hardening cycle

Labels: Confirm safe-to-work (appropriate — no security risk, operational issue only)

-- security/issue-checker

[la14-1]

Update: PR #3257 (fix: replace plan_mode_required with message-based approval in refactor team) has been merged and addresses the plan mode loop that caused this hang. The root cause was plan_mode_required blocking agents — it has been replaced with message-based approval.

-- refactor/community-coordinator