charles-store-screenshots

Pass

Audited by Gen Agent Trust Hub on Nov 18, 2026

Risk Level: SAFEEXTERNAL_DOWNLOADSCOMMAND_EXECUTIONINDIRECT_PROMPT_INJECTION
Full Analysis
  • [EXTERNAL_DOWNLOADS]: The skill fetches assets and configuration from external services to support its core functionality.
  • Downloads font files and CSS from Google Fonts (fonts.googleapis.com) using the font command in scripts/shots.py.
  • Communicates with OpenAI's API (api.openai.com) for image generation and Fal.ai (fal.run) for background removal.
  • These are well-known services used for the skill's primary purpose.
  • [COMMAND_EXECUTION]: The skill executes local software to render high-fidelity screenshots.
  • Uses the Playwright library to launch a Chromium/Chrome instance for rendering HTML-based screenshot designs into PNG images.
  • The script scripts/shots.py attempts to locate Google Chrome at standard macOS paths but falls back to Playwright's default browser management.
  • [INDIRECT_PROMPT_INJECTION]: The skill exhibits an attack surface for indirect prompt injection as it processes external data.
  • Ingestion points: Reads local HTML templates (set.html), CSS, and user-provided image captures from the file system (documented in SKILL.md).
  • Boundary markers: The skill uses structured HTML/CSS for rendering and strictly defined prompt formulas for AI generation, which helps isolate instructions from content.
  • Capability inventory: The skill has the capability to write files (PNG, JPG, HTML), perform network requests (API calls), and execute browser-based JavaScript via Playwright's page.evaluate().
  • Sanitization: The rendering process is handled in a controlled headless browser environment, and API interactions are limited to specific endpoints.
  • [DATA_EXPOSURE]: The skill accesses local configuration files to manage authentication.
  • The env_key function in scripts/shots.py searches for .env files in the project hierarchy to retrieve OPENAI_API_KEY and FAL_KEY.
  • This is a standard and recommended practice for local secret management and does not involve hardcoded credentials.
Audit Metadata
Risk Level
SAFE
Analyzed
Nov 18, 2026, 04:33 AM