grep-with-context

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

• Third-party content exposure detected (medium risk: 0.30). In SKILL.md Step 1/3, the skill uploads user-provided files (and their contents) as Grep attachments and Step 4/5 refines and runs a research job with --attachment-ids, so any outsider-authored text included in those uploaded files becomes LLM-readable input at runtime.