error-handling
Pass
Audited by Gen Agent Trust Hub on Feb 17, 2026
Risk Level: SAFEPROMPT_INJECTION
Full Analysis
- PROMPT_INJECTION (LOW): The file references/anti-rationalization.md contains meta-instructions explicitly designed to steer the LLM's reasoning engine. The section 'Why This Works' describes using 'Authority language' and 'commitment statements' to 'trigger compliance mechanisms in LLM decision-making'. While intended to enforce professional software engineering standards, these techniques represent a deliberate attempt to override the agent's default instructional weights.
- DATA_EXFILTRATION (SAFE): No hardcoded credentials or unauthorized data access patterns were found. The provided examples in references/overview.md correctly demonstrate security best practices, such as scrubbing sensitive fields (e.g., passwords) before transmitting data to external error tracking services like Sentry.
- REMOTE_CODE_EXECUTION (SAFE): The skill consists entirely of documentation and illustrative code snippets. There are no automated scripts that download or execute external code, and no use of dangerous functions like eval() or exec().
Audit Metadata