ai-forge-judge

Pass

Audited by Gen Agent Trust Hub on Aug 9, 2026

Risk Level: SAFE
Full Analysis
  • [SAFE]: The skill is primarily instructional, focused on evaluating the quality of other LLM prompts using a multi-dimensional scoring rubric.
  • [COMMAND_EXECUTION]: While the skill mentions evaluating shell/bash prompts (Group B), it does not contain executable scripts or perform shell operations itself. It serves as a static analyzer/judge for text provided in the conversation context.
  • [EXTERNAL_DOWNLOADS]: The skill mentions an optional WebFetch to https://agentskills.io/specification to get the latest version of the specification. This is a well-known, domain-relevant source and is treated as safe.
  • [DATA_EXFILTRATION]: No network operations or sensitive file access patterns were detected. The skill operates entirely within the provided context to analyze user-supplied text.
  • [PROMPT_INJECTION]: The skill contains 'MANDATORY' instructions and 'NEVER' lists (e.g., in SKILL.md and references/bash-dimensions.md), but these are legitimate behavioral constraints for its role as an evaluator and do not attempt to bypass core safety filters or extract system prompts.
Audit Metadata
Risk Level
SAFE
Analyzed
Aug 9, 2026, 09:00 AM
Security Audit — agent-trust-hub — ai-forge-judge