model-council
Pass
Audited by Gen Agent Trust Hub on Sep 26, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill processes user-supplied queries and transmits them to external AI model APIs. This creates a surface for indirect prompt injection if the user input contains malicious instructions designed to influence the behavior of the sub-models. However, the skill mitigates this by instructing sub-models to provide analysis only and using the primary agent to synthesize the final output.
- Ingestion points: User input is passed as a prompt to the
api_council.pyscript via command-line arguments. - Boundary markers: The skill uses basic structural labels like 'Problem:' in the prompt template to separate instructions from user content.
- Capability inventory: The skill possesses the capability to perform parallel network requests to official AI service endpoints.
- Sanitization: User input is interpolated into the external prompt template without specific sanitization or escaping mechanisms.
Audit Metadata