prompt
Pass
Audited by Gen Agent Trust Hub on Oct 3, 2026
Risk Level: SAFE
Full Analysis
- [SAFE]: The skill serves as a pedagogical resource for advanced prompt engineering. It provides structured guidance on context architecture, few-shot design, and model-specific optimizations.
- [PROMPT_INJECTION]: Static detectors flagged instructions in
references/review.mdregarding instruction overrides. However, analysis confirms these are benign methodological guidelines intended to reduce confirmation bias during audits (e.g., instructing the agent to frame reviews as a search for defects) and educational 'smoke tests' for verifying prompt security. They are not attempts to bypass model safety filters. - [INDIRECT_PROMPT_INJECTION]: The skill explicitly addresses this threat category by providing technical strategies for mitigation, such as XML delimitation of untrusted data and the implementation of information barriers between planning and data-parsing components.
- [DATA_EXFILTRATION]: There are no network operations, hardcoded credentials, or sensitive file access patterns. The skill includes a checklist for detecting credentials in other prompts as a defensive measure.
Audit Metadata