dum-dum

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

• Third-party content exposure detected (medium risk: 0.30). Runtime path collect.py → find_transcripts() scans ~/.claude/projects/**.jsonl session transcripts and scan_transcript() reads record["content"]/record["reason"]/tool inputs from those JSONL lines to produce user_correction, hook_denial, etc.; this ingest is not restricted to first-party/authenticated text and could include outsider-authored transcript text if the attacker can get content into the user’s session logs.