prompt-engineering

Pass

Audited by Gen Agent Trust Hub on Jun 19, 2026

Risk Level: SAFECOMMAND_EXECUTIONEXTERNAL_DOWNLOADS
Full Analysis
  • [COMMAND_EXECUTION]: The skill makes extensive use of the belt CLI tool via belt app run to demonstrate prompting techniques. These commands are hardcoded examples used to interact with the inference.sh platform.
  • [EXTERNAL_DOWNLOADS]: The skill references images and installation documentation hosted on inference.sh domains and the official inference-sh GitHub repository. These are recognized as legitimate service resources.
  • [PROMPT_INJECTION]: Indirect prompt injection attack surface identified in SKILL.md.
  • Ingestion points: User-controlled data placeholders such as [article text] and [code] in prompt templates.
  • Boundary markers: Uses markdown code blocks and clear descriptive labels (e.g., 'Code:') to separate instructions from data.
  • Capability inventory: Uses the Bash(belt *) tool to send prompts to external AI models.
  • Sanitization: No specific escaping or validation logic is provided for the content inserted into the template placeholders.
Audit Metadata
Risk Level
SAFE
Analyzed
Jun 19, 2026, 02:15 AM
Security Audit — agent-trust-hub — prompt-engineering