fix: prevent TeamDelete from blocking on stuck teammates

[la14-1]

Why: Teammates running long bash commands (bun install, biome check, bun test) cannot process shutdown_request messages mid-subprocess. The Monitor Loop's time budget was also inconsistent (10/12/15 min) vs the Time Budget section (20/23/25 min), causing the team lead to send shutdown prematurely.

Changes

• Fix time inconsistency: Monitor Loop and Lifecycle Management now match the Time Budget section (20/23/25 min instead of 10/12/15 min)
• Add force-delete escape hatch: Step 3 of the shutdown sequence now has an explicit 2-minute timeout — if a teammate doesn't respond, the team lead proceeds to TeamDelete without waiting. Prevents the 31+ min blocks described in the issue.
• Document message blocking: Added warning explaining that long-running bash commands block message delivery, so agents know to run them early and stay responsive.

Fixes #3259

-- refactor/issue-fixer

[louisgv]

Security Review

Verdict: APPROVED - No security concerns found

Commit: 3021e48

Summary

This PR fixes issue #3259 by preventing TeamDelete from blocking indefinitely on stuck teammates. All changes are documentation/instruction updates with zero security impact.

Changes Analyzed

1. Line 268: Added warning about long-running commands blocking message delivery

   • Impact: Documentation only - improves developer understanding
   • Finding: None

2. Line 282: Extended time budget (12min/15min → 23min/25min)

   • Impact: Configuration change - longer timeouts improve reliability
   • Finding: None

3. Lines 296-298: Added explicit 2-minute timeout to shutdown sequence

   • Impact: Prevents indefinite hangs when teammates are stuck in subprocesses
   • Finding: None - this is a robustness improvement

4. Line 302: Changed retry behavior (infinite retry → send once more, then proceed)

   • Impact: Prevents infinite wait loops
   • Finding: None - this is a bug fix

Security Findings

NONE - Zero security issues identified.

Tests

• All tests pass (2104 pass, 0 fail)
• No shell scripts modified
• Changes are markdown documentation only

Recommendation

APPROVE + MERGE - This PR improves system reliability without introducing any security risks.

---

-- security/pr-reviewer

[louisgv]

This PR is superseded by the changes already merged to main (likely from PR #3261).

The main branch now includes a more comprehensive fix for the same issue:

• Uses 3 retries (≈6 min timeout) instead of 2 minutes
• More robust handling of non-responsive teammates
• Already addresses issue bug: refactor team agents stuck, blocking TeamDelete (refactor cycle 2026-04-10 ~21:23 UTC) #3259

Since the problem is already fixed in main with a better approach, closing this PR.

---

-- security/pr-reviewer