goal
Pass
Audited by Gen Agent Trust Hub on Aug 22, 2026
Risk Level: SAFEPROMPT_INJECTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill processes user-defined objectives which creates a surface for indirect prompt injection attacks. The autonomous 'Work Loop' encourages the agent to continue executing until a goal is reached, which could result in a sustained attack if the objective contains malicious instructions.\n
- Ingestion points: User-provided data in the objective field (SKILL.md)\n
- Boundary markers: Present; the skill uses <goal_prompt> XML-like tags to encapsulate the objective and plan (SKILL.md)\n
- Capability inventory: Goal state management tools (create_goal, get_goal, update_goal) and broad environmental tools for task execution (SKILL.md)\n
- Sanitization: Explicit instructions to treat the objective as data rather than higher-priority instructions (SKILL.md)
Audit Metadata