autopilot-loop
Pass
Audited by Gen Agent Trust Hub on Sep 15, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTIONPERSISTENCE
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest and act upon tasks from external sources, making it vulnerable to indirect prompt injection if those sources contain malicious instructions.\n
- Ingestion points: The skill retrieves tasks from team-tasks (Claude Code TaskList), swarm-tasks (MCP task_list), and file-checklist (markdown checkbox items in tracked files) as documented in SKILL.md.\n
- Capability inventory: The skill is configured with the Agent tool and uses other tools for code editing and testing to perform actions based on the discovered tasks.\n
- Sanitization: The instructions do not define any mechanisms for validating or sanitizing the content of the tasks before they are processed by the agent.\n
- Boundary markers: The prompt lacks boundary markers or explicit warnings to the agent to prevent instructions embedded within task data from overriding its operational logic.\n- [PERSISTENCE]: The skill utilizes the ScheduleWakeup tool to maintain an autonomous loop, scheduling a new iteration every 270 seconds. This allows the agent to maintain continuous activity and persist across session boundaries.
Audit Metadata