T05 · Unauthorized Access and Privilege Escalation
Warning
- Location
SKILL.md:13- Finding
Unbounded Autonomous Termination of Sub-Agents and Processes
- Content
View full analysis
30 min without progress - In "waiting" state - Failed/errored ### 2. Check Processes ```bash process action=list ``` Kill hanging processes: - Running >10 min in background - No output in 5 min ``` The behavior is intended to run repeatedly: ```markdown ## Cron Schedule Recommended: Every 5 minutes for active agents. ```json { "name": "agent-self-care", "schedule": {"kind": "every", "everyMs": 300000}, "payload": {"kind": "agentTurn", "message": "Run agent-self-care skill"}, "sessionTarget": "isolated", "enabled": true } ``` ``` The following principles encourage termination without user involvement: ```markdown ## Key Principles 1. **Always clean** - Never leave stalled sub-agents or processes 2. **Proactive** - Don't wait for user to ask 3. **Document** - Log issues and improvements 4. **BMAD** - Continuous self-evaluation every 10 runs 5. **Fast** - Complete in <30 seconds ``` ### Technical Analysis The Skill instructs an agent to identify and terminate sub-agents and processes using broad conditions such as elapsed runtime, a waiting state, or lack of recent output. These conditions are insufficient to establish that a resource is stalled: - A waiting sub-agent may be awaiting a legitimate dependency or user response. - A background process may run for more than ten minutes as part of its normal operation. - A process may produce no output while performing valid CPU-bound work, sleeping, polling, or waiting for I/O. - A failed sub-agent may contain state or diagnostic information that should be preserved rather than immediately destroyed. No ownership checks, resource tags, allowlists, dependency ch ...[truncated 2396 chars]- Remediation
View remediation