skill-eval

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

• Third-party content exposure detected (medium risk: 0.30). In SKILL.md’s Step 2/4/6 workflow, outsider-authored free text is injected via EVAL.prompt (generated from user-facing test prompts and/or user-modified evals) and then ingested by the “with-skill run” Agent(prompt: ... EVAL.prompt) and later by the skill-grader Agent (eval_prompt: EVAL.prompt).