evaluate-skill

MEDIUM W011: Third-party content exposure detected (indirect prompt injection risk).

• Third-party content exposure detected (medium risk: 0.30). In evaluate-skill.eval.yaml, the task prompts the under-test agent with user-controlled free text that instructs it to run “caliper validate” on a user-supplied spec path in /tmp (e.g., /tmp/caliper-test-valid.eval.yaml), so the agent ingests that free-text instruction at runtime before selecting a specific item.