engineer-agent-loop
Pass
Audited by Gen Agent Trust Hub on Jul 21, 2026
Risk Level: SAFE
Full Analysis
- [SAFE]: The skill is instructional in nature, providing best practices for engineering agentic loops with a focus on reliability, deterministic caps, and context budgeting.
- [DATA_EXFILTRATION]: No sensitive data access or unauthorized network operations. The documentation explicitly advises practitioners to keep secrets out of state files and traces.
- [REMOTE_CODE_EXECUTION]: No execution of untrusted remote code. References to external scripts and patterns (e.g., the Ralph pattern) are provided as illustrative documentation rather than executable code.
- [EXTERNAL_DOWNLOADS]: All external URL references target well-known research repositories (arXiv) or established technology organizations (Anthropic, Microsoft, Cloudflare, Google, LangChain) for informational purposes.
- [PROMPT_INJECTION]: No instructions attempt to override agent safety guidelines. The skill actually includes security advice on how to protect loops from indirect prompt injection via untrusted inputs.
Audit Metadata