specs-design

Pass

Audited by Gen Agent Trust Hub on Aug 27, 2026

Risk Level: SAFEINDIRECT_PROMPT_INJECTIONEXTERNAL_DOWNLOADS
Full Analysis
  • [INDIRECT_PROMPT_INJECTION]: The skill extracts design elements from potentially untrusted HTML and CSS files located in docs/product/demo/ to generate documentation.
  • Ingestion points: The skill explicitly instructs subagents to read and process docs/product/demo/**/*.html and docs/product/demo/**/*.css files in SKILL.md (Step 1, Step 3) and templates/design-md-prompt.md.
  • Boundary markers: The skill does not define clear delimiters or provide explicit instructions to the subagent to ignore any embedded directives or natural language instructions within the demo files.
  • Capability inventory: The subagents are granted the capability to write the resulting documentation to the local file system (docs/specs/design/). The main agent also utilizes network tools to fetch remote specifications.
  • Sanitization: There is no evidence of sanitization or structural validation of the incoming demo content before it is processed by the language model.
  • [EXTERNAL_DOWNLOADS]: The skill fetches format specifications and reference examples from the official GitHub repository of google-labs-code (a well-known organization). These remote requests are performed to ensure compliance with the design.md standard if local cached copies in the refs/ directory are unavailable or outdated.
Audit Metadata
Risk Level
SAFE
Analyzed
Aug 27, 2026, 02:49 PM
Security Audit — agent-trust-hub — specs-design