harvest

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

• Third-party content exposure detected (high risk: 0.70). The skill reads docs/harvest/<workflow_id>.md (workflow trace) and docs/issues-filed.json and then injects extracted free-text sections (e.g., “Signals observed” and gate rejection reasons) into the LLM context for analysis before calling substrate-write; these files are outsider-authored content if the operating user did not author the workflow trace/issues themselves.