The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill maintains project state through a progress.md file which is read and updated across development iterations. 1. Ingestion points: The agent reads progress.md in game/game-testing.md and processes user-supplied prompts for site generation. 2. Boundary markers: The skill mandates prefixing progress.md with Original prompt: to maintain context. 3. Capability inventory: The skill allows shell execution via pty_spawn, file system modifications, and browser automation via Playwright. 4. Sanitization: No explicit sanitization step for the progress.md content before it is re-processed.
[COMMAND_EXECUTION]: The skill uses pty_spawn to run local development servers (e.g., Vite, Python http.server) and the Playwright test client during the build and QA process as described in SKILL.md and shared/12-playwright-interactive.md.
[EXTERNAL_DOWNLOADS]: The skill fetches UI libraries, icons, and 3D assets from reputable CDNs including Google Fonts, Fontshare, unpkg, jsdelivr, and esm.sh, which are well-known and trusted providers for web development assets.
[DYNAMIC_EXECUTION]: The Playwright testing script in game/scripts/web_game_playwright_client.js injects a virtual time shim into the browser environment to facilitate deterministic game testing.
[SAFE]: The skill implements a robust, mandatory security review process via a dedicated subagent prompt in website-publishing/security_subagent_prompt.md that scans for exposed credentials, dangerous code patterns, and insecure CORS settings before deployment.

website-building