brainstorm
Pass
Audited by Gen Agent Trust Hub on Aug 20, 2026
Risk Level: SAFEPROMPT_INJECTION
Full Analysis
- [PROMPT_INJECTION]: The skill exhibits an indirect prompt injection surface where user-controlled input (the topic) is directly interpolated into prompts for sub-agents.
- Ingestion points: The {topic} variable, derived from user arguments or the current conversation and rephrased by the agent, is embedded in instructions for sub-agents in SKILL.md.
- Boundary markers: The prompt template for sub-agents (You are {persona}, brainstorming about: {topic}) lacks boundary delimiters or instructions to ignore embedded commands within the topic string.
- Capability inventory: The sub-agents are explicitly granted permission to perform WebSearch queries, which could be manipulated or triggered by a malicious topic.
- Sanitization: No input sanitization, filtering, or validation is performed on the topic before it is passed to the sub-agents.
Audit Metadata