chaos-design
Pass
Audited by Gen Agent Trust Hub on Jul 9, 2026
Risk Level: SAFEPROMPT_INJECTION
Full Analysis
- [PROMPT_INJECTION]: The skill is susceptible to indirect prompt injection as it ingests and processes untrusted data from the user to generate its outputs.\n
- Ingestion points: Step 1 specifically directs the agent to gather a resilience hypothesis and target system architecture details via user interaction (AskUserQuestion).\n
- Boundary markers: The instructions do not define clear delimiters or data framing constraints to prevent embedded instructions in the user context from influencing agent behavior.\n
- Capability inventory: The execution environment allows the use of Bash, Write, and WebFetch tools, which increases the potential impact if the agent follows malicious instructions hidden in user input.\n
- Sanitization: There is no evidence of input validation, filtering, or instruction-trapping logic to neutralize potentially malicious prompts within the gathered context.
Audit Metadata