monitor-episodic-archiver

CRITICAL E004: Prompt injection detected in skill instructions.

• Potential prompt injection detected (medium risk: 0.30). The skill specifies an always-on autonomous monitoring pipeline that proactively profiles users, merges profiles into persistent memory, and mines transcripts for model retraining while insisting the agent "MUST NOT stop", which weakens human oversight and raises privacy/control concerns without clearly malicious intent.

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

• Third-party content exposure detected (medium risk: 0.30). The runtime workflow in monitor_archiver.py reads “session summaries” from the ArangoDB collection session_summaries for the nightly pipeline and then passes session_id/user_id into analyze_archived_session(...), which (by design) processes archived session content with an LLM for taxonomy/user profiling—so outsider-authored text stored in those collections can be ingested indirectly without selecting a specific item.