issue-writeup
Pass
Audited by Gen Agent Trust Hub on Jun 16, 2026
Risk Level: SAFEPROMPT_INJECTION
Full Analysis
- [SAFE]: No malicious patterns, obfuscation, or unauthorized access attempts were found. The skill is designed for professional documentation and adheres to regulatory frameworks.
- [EXTERNAL_DOWNLOADS]: The skill includes links to official regulatory and standards bodies (e.g., federalreserve.gov, occ.gov, finra.org) for documentation purposes. These are well-known services and do not involve executable code.
- [PROMPT_INJECTION]: No direct prompt injection or safety bypass attempts were detected. The skill identifies a potential surface for indirect prompt injection as it ingests untrusted fieldwork notes and examiner letters. Boundary markers are present in the form of structured CCCE (Condition, Criteria, Cause, Effect) headers and templates. The capability inventory is restricted to document generation (Word/Markdown) via platform plugins, and no high-privilege operations or network-write capabilities are triggered by the processed data. No explicit sanitization is performed within the instructions.
- [DATA_EXFILTRATION]: No code or instructions for exfiltrating sensitive data were detected. The skill uses standard utilities for document generation.
- [COMMAND_EXECUTION]: No shell commands, privilege escalation attempts, or dangerous system calls were detected.
Audit Metadata