twoslash

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

• Third-party content exposure detected (medium risk: 0.30). The runtime workflow in scripts/run-evals.py ingests outsider-authored free text from evals/evals.json (prompt/expected_output) and evals/trigger-queries.json (query) and sends it to the Claude CLI via run_claude_json(...prompt=...) for generation and trigger classification.