executing-plans
Pass
Audited by Gen Agent Trust Hub on Mar 29, 2026
Risk Level: SAFEPROMPT_INJECTION
Full Analysis
- [PROMPT_INJECTION]: The skill is susceptible to indirect prompt injection because its core function is to load and execute instructions from a 'plan file' provided in the agent's environment. \n
- Ingestion points: The process starts by reading an external plan file (Step 1 in SKILL.md). \n
- Boundary markers: There are no technical delimiters (e.g., XML tags or unique markers) specified in the instructions to separate the plan data from the agent's system instructions. \n
- Capability inventory: The 'Execute Tasks' phase implies the usage of general-purpose tools such as shell execution and file system modifications to implement the plan. \n
- Sanitization: The skill mitigates risks by requiring the agent to 'review critically', raise concerns with a human partner, and 'stop executing immediately' if instructions are unclear or blockers are encountered.
Audit Metadata