autopilot-chaos
Pass
Audited by Gen Agent Trust Hub on Jun 28, 2026
Risk Level: SAFEPROMPT_INJECTIONNO_CODE
Full Analysis
- [PROMPT_INJECTION]: The skill instructions (e.g., "execute the whole plan without stopping for confirmation" and "Make reasonable default choices at every fork and keep moving") explicitly direct the agent to override standard interactive safety protocols and human-in-the-loop checkpoints.
- [INDIRECT_PROMPT_INJECTION]: The skill's primary function creates a significant surface for indirect prompt injection by processing external tasks autonomously.
- Ingestion points: Processes task descriptions and potentially associated data files (SKILL.md).
- Boundary markers: No delimiters or specific safety instructions are defined to distinguish between user intent and instructions embedded within external task data.
- Capability inventory: The skill enables the autonomous use of all available agent capabilities and tools.
- Sanitization: No evidence of input validation or sanitization for task-specific instructions is present.
Audit Metadata