managed-model-endpoints

Pass

Audited by Gen Agent Trust Hub on Sep 8, 2026

Risk Level: SAFEINDIRECT_PROMPT_INJECTIONDYNAMIC_EXECUTION
Full Analysis
  • [INDIRECT_PROMPT_INJECTION]: The skill processes data from external model endpoints during health checks and inference smoke tests. 1. Ingestion points: Inference response data captured in Step 5 of SKILL.md. 2. Boundary markers (absent): The instructions do not define delimiters or warnings to ignore embedded instructions in model responses. 3. Capability inventory: The skill performs endpoint lifecycle management (start/stop) and inference routing defined in SKILL.md. 4. Sanitization (absent): No sanitization or validation logic is defined for the external data ingested.
  • [DYNAMIC_EXECUTION]: The skill instructs the agent to implement and execute start, stop, and health check scripts for model lifecycle management as described in SKILL.md.
Audit Metadata
Risk Level
SAFE
Analyzed
Sep 8, 2026, 01:46 AM
Security Audit — agent-trust-hub — managed-model-endpoints