resume

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

• Third-party content exposure detected (medium risk: 0.65). At runtime the skill reads outsider-authored free text from the experiment’s persisted history files—specifically .autoresearch/{domain}/{name}/results.tsv and .autoresearch/{domain}/{name}/program.md (via cat)—and injects that content into the agent’s LLM context when summarizing/reporting state.