tau

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

• Third-party content exposure detected (medium risk: 0.30). In SKILL.md’s “Mandatory Live DAG Monitoring” read loop, the agent continuously reads local Tau JSON artifacts under the active <run_dir> (e.g., dag-progress.json, node receipts, handler-response receipts), which are produced from externally authored targets like GitHub issues/PRs and therefore can contain outsider-provided free text.