awwwards-doctor

Pass

Audited by Gen Agent Trust Hub on Nov 11, 2026

Risk Level: SAFECOMMAND_EXECUTIONEXTERNAL_DOWNLOADSINDIRECT_PROMPT_INJECTIONDYNAMIC_EXECUTION
Full Analysis
  • [COMMAND_EXECUTION]: The skill instructs the agent to execute various shell commands and npm scripts for maintenance purposes.
  • Runs npm run doctor, npm run build, and npm test within the local repository checkout to diagnose and fix toolchain issues.
  • Performs file operations (cp, mv, mkdir) on the application's local cache directory (~/.awwwards-mcp) to backup and reset the SQLite database.
  • [EXTERNAL_DOWNLOADS]: The skill manages the installation and update of necessary software components from vendor and well-known sources.
  • Guides the user to install or update the awwwards-mcp package via npm.
  • Triggers the installation of browser dependencies using npx playwright install chromium.
  • Mentions optional installation of the well-known spectorjs package for WebGL frame capture.
  • [INDIRECT_PROMPT_INJECTION]: The skill processes data scraped from the external website awwwards.com, which introduces a potential surface for indirect prompt injection.
  • Ingestion points: Data enters the system via tools like capture_page, inspect_page, and search_references (in SKILL.md) which parse HTML and JSON from external web pages.
  • Boundary markers: The skill documentation indicates the use of specific response modes (e.g., mode: "anchors") and redacted inventories to structure external data.
  • Capability inventory: The system can execute npm scripts, perform file system writes (scaffolding), and make network requests for scraping.
  • Sanitization: The skill explicitly documents security measures in v2-routing.md, such as storing shader source from untrusted pages as raw data while ensuring it is never executed, and implementing DNS/IP guards to prevent SSRF during build verification.
  • [DYNAMIC_EXECUTION]: The skill includes tools for scaffolding and reconstructing projects.
  • Tools like build_project and build_motion generate code structures (Vite/Next.js scaffolds) and configuration files based on captured data.
  • Documentation in v2-routing.md clarifies that these stubs are for reconstruction and do not execute captured callbacks, shader source, or user data from the source pages.
Audit Metadata
Risk Level
SAFE
Analyzed
Nov 11, 2026, 10:21 PM