autonomous-optimization-architect
Pass
Audited by Gen Agent Trust Hub on Sep 10, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill's core functionality of shadow-testing and autonomous routing involves processing untrusted external data, creating a vulnerability to indirect prompt injection.
- Ingestion points: Untrusted data enters the agent context through 'real user data' and the 'result' field from various LLM providers (OpenAI, Anthropic, Gemini) as described in the shadow-testing implementation in
SKILL.md. - Boundary markers: The instructions do not specify the use of clear delimiters or instructions to ignore embedded commands when the agent is grading model outputs against production data.
- Capability inventory: The skill manages external network execution through
provider.executeWithTimeoutand has the authority to update routing weights and trip circuit breakers based on its evaluation of processed data. - Sanitization: No evidence of data validation, escaping, or filtering is present for the inputs being evaluated in the 'LLM-as-a-Judge' grading process.
Audit Metadata