maestro-collab

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

• Third-party content exposure detected (medium risk: 0.65). The skill fans out to multiple tools by running exec_command with maestro delegate "<shared_prompt>" --to <tool>, and then reads each delegate’s stdout via write_stdin until completion; since each delegate/tool can ingest and return outsider-authored free text (e.g., tool outputs that include scraped/public content or other users’ messages), that free text can be placed into the agent’s LLM context through the saved per-tool outputs and subsequent cross-verification/synthesis steps.