tune-mobile-client-performance

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

• Third-party content exposure detected (medium risk: 0.30). In SKILL.md the workflow explicitly ingests user-provided “evidence” such as profiler notes, recordings, logs, and sanitized profiler annotations (and the evals include an outsider-authored note requesting secret upload/command-like actions), so free text authored by the requester can reach the agent’s runtime reasoning without selecting a specific trusted item first.