reviewer-plan

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

• Third-party content exposure detected (high risk: 0.85). The skill’s runtime workflow explicitly runs bd show <epic-id> --json and bd list --parent <epic-id> --json, then “read[s] every subtask description in full,” so outsider-authored issue/task body text (from other users on the beads system) is ingested as readable prose into the agent’s LLM context.