reflect
Pass
Audited by Gen Agent Trust Hub on Nov 2, 2026
Risk Level: SAFECOMMAND_EXECUTIONINDIRECT_PROMPT_INJECTION
Full Analysis
- [COMMAND_EXECUTION]: The skill references and utilizes local Python scripts (scripts/bias_pattern_detector.py, scripts/conversation_depth_analyzer.py, and scripts/directional_recommendation_validator.py) to process conversation text. These scripts are self-contained, use only standard libraries, and are used for heuristic analysis of the conversation's depth and bias signals.
- [INDIRECT_PROMPT_INJECTION]: The skill processes the full conversation history to reassess goals and assumptions (Ingestion point: conversation history). While it lacks explicit boundary markers or sanitization for this data, its capability inventory is restricted to executing local Python scripts for textual scanning. The risk of responding to instructions embedded in the history is mitigated by a strict 5-Dimension Analysis Framework and Honest Output Discipline that requires the agent to cite specific evidence for every claim, preventing the blind following of potentially malicious instructions within the history.
Audit Metadata