harvest

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

• Third-party content exposure detected (medium risk: 0.65). The required workflow can ingest outsider-authored free text when the invocation argument is a “link (issue, PR, file, error)” and the agent “read[s] it for what’s actually being asked,” i.e., it may fetch and include the body/comments of an external GitHub issue/PR/file into the LLM context for analysis.