three-radius-model

Pass

Audited by Gen Agent Trust Hub on Aug 22, 2026

Risk Level: SAFE
Full Analysis
  • [SAFE]: The skill does not contain any executable scripts, shell commands, or automated subprocess calls. It consists entirely of markdown instructions for agent behavior.
  • [EXTERNAL_DOWNLOADS]: The skill references external URLs for documentation and historical context. Links target deciqai.com, github.com/deciqai, gutenberg.org, berkshirehathaway.com, and doi.org. These are either vendor-owned resources or well-known trusted domains for public literature, corporate reports, and academic research.
  • [PROMPT_INJECTION]: No malicious patterns such as 'ignore previous instructions', 'DAN', or 'developer mode' overrides were detected. The instructional language follows a standard coaching flow with step-by-step gates.
  • [DATA_EXFILTRATION]: No access to sensitive system paths (e.g., .ssh, .env, .aws) or unauthorized network exfiltration patterns were found. The skill only processes user-provided responses to coaching prompts.
  • [OBFUSCATION]: Analysis for Base64 encoding, zero-width characters, homoglyphs, and hidden Unicode tags yielded no findings. The content is transparent and human-readable.
  • [REMOTE_CODE_EXECUTION]: There are no package managers (npm, pip) or remote script execution patterns (curl | bash) present in the analyzed files.
Audit Metadata
Risk Level
SAFE
Analyzed
Aug 22, 2026, 08:15 AM
Security Audit — agent-trust-hub — three-radius-model