Archived agent repeatedly restarts after mngr stop/archive

[joshalbrecht]

Description

An archived worker agent (setup-repo, ID agent-0153f5791adc4332a0c172de90e595d7) has restarted at least 3 times after being archived via mngr archive -f and stopped via mngr stop. Each restart receives stray hi messages and transitions to WAITING before being stopped again.

Steps to reproduce

1. Create a working agent via mngr create
2. Agent completes its task and transitions to WAITING
3. Archive the agent with mngr archive -f <agent-id>
4. Agent transitions to STOPPED as expected
5. ~40 minutes later, agent restarts (new start_time) and transitions back to WAITING
6. Stop the agent again with mngr stop <agent-id>
7. Repeat from step 5 — agent restarts again

Observed behavior

• Agent has archived_at label set (e.g. 2026-03-28T18:41:00.832418+00:00)
• Despite this, the agent's process restarts multiple times (~40-60 min apart)
• Each time it resumes, its transcript shows stray hi messages being injected
• The agent responds "No response requested." suggesting the session is being resumed by something

Timeline:

• 18:41 — archived
• 18:45 — first STOPPED confirmed
• 20:45 — restarted (start_time), transitioned WAITING at 20:50
• 21:00 — stopped again via mngr stop
• 21:37 — restarted again (new start_time), transitioned WAITING at 21:42

Expected behavior

An archived+stopped agent should not restart. The archived_at label should prevent the agent from being restarted by the system.

Environment

• Agent type: worker
• Mind: product_manager
• Host: host-9c184f4ad4e147bba6c5877f5a3efbe7 (local, running)
• Agent ID: agent-0153f5791adc4332a0c172de90e595d7
• Agent name: setup-repo