css-first

Pass

Audited by Gen Agent Trust Hub on Jun 3, 2026

Risk Level: SAFENO_CODEEXTERNAL_DOWNLOADSPROMPT_INJECTION
Full Analysis
  • [NO_CODE]: The skill consists exclusively of static Markdown and CSS files. It does not ship with any executable scripts (Shell, Python, JavaScript) that run on the host environment.
  • [EXTERNAL_DOWNLOADS]: The skill provides a structured workflow for the agent to fetch live browser support metadata and documentation from well-known and trusted platforms: unpkg.com (for web-features and MDN compatibility data) and developer.mozilla.org. These are legitimate developer tools used to ensure information accuracy.
  • [PROMPT_INJECTION]: The fetch workflow for MDN data creates an indirect prompt injection surface as the agent is instructed to ingest and process data from external URLs.
  • Ingestion points: references/live-mdn-fetch.md describes fetching JSON and Markdown from unpkg.com and developer.mozilla.org.
  • Boundary markers: Absent. The instructions do not specify explicit delimiters for the external content.
  • Capability inventory: references/rules/browser-verification.md encourages the agent to use browser-based tools (MCP servers) for rendering verification, which includes navigating to URLs and executing simple JavaScript (CSS.supports).
  • Sanitization: Absent.
  • [COMMAND_EXECUTION]: Instructions in references/rules/browser-verification.md guide the agent to use standard browser automation tools (MCP servers) for Render Checks and Computed Style Inspection. These are development-oriented tools and do not involve arbitrary host command execution.
Audit Metadata
Risk Level
SAFE
Analyzed
Jun 3, 2026, 09:04 AM
Security Audit — agent-trust-hub — css-first