skill-eval

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

• Third-party content exposure detected (medium risk: 0.30). In scripts/eval_runner.py, run loads an outsider-controlled suite JSON (evals/<skill-name>/evals.json), then renders each evals[*].prompt and evals[*].expectations/common_assertions into the executor and grader prompts (render_executor_prompt/render_grader_prompt), so arbitrary free text from that file is ingested at runtime.