agent-reliability-and-guardrails
Pass
Audited by Gen Agent Trust Hub on Aug 26, 2026
Risk Level: SAFE
Full Analysis
- [SAFE]: The skill consists entirely of educational content, architectural guidelines, and pseudocode examples related to AI safety and reliability. It does not include any executable scripts or active agent instructions.
- [PROMPT_INJECTION]: The static analysis flag for prompt injection is a false positive. While the text contains phrases such as "ignore previous instructions," these are used descriptively within a section explaining how engineers should detect and defend against jailbreak attempts. The content does not attempt to override the instructions of the LLM analyzing the file.
- [COMMAND_EXECUTION]: The document mentions various system commands (e.g., rm, chmod, DROP) and tool patterns, but these are cited strictly as examples of high-risk actions that developers should gate with human approval. No actual shell commands are executed or recommended for execution by the agent.
- [DATA_EXPOSURE_AND_EXFILTRATION]: No network operations, sensitive file access, or credential harvesting patterns were found. The mention of sensitive paths or PII is part of the defensive guidance for scrubbing data before it reaches a model.
Audit Metadata