The Agent Skills Directory

[COMMAND_EXECUTION]: The skill uses npm, npx, and playwright to scaffold projects, install dependencies, and capture screenshots. It provides custom scripts in the template/scripts/ directory for brand extraction and media generation.
[EXTERNAL_DOWNLOADS]: The skill triggers the download of browser binaries via playwright install chromium (a well-known service from Microsoft) and fetches project templates and font assets from standard registries and the Google Fonts API.
[PROMPT_INJECTION]: The skill instructions contain a directive for the agent to autonomously start an onboarding workflow ('do NOT wait to be asked') upon first load. While this increases agent autonomy, the actions performed (scaffolding a local theme and requesting local brand media) are benign and serve the skill's primary setup purpose.
[INDIRECT_PROMPT_INJECTION]: The skill has an attack surface where the agent is instructed to read user-supplied images from the brand-input/ folder to extract branding data.
Ingestion points: Media files located in the brand-input/ directory (SKILL.md, Step 1).
Boundary markers: The instructions lack specific boundary delimiters for the vision-to-text task but rely on the agent to populate a specific JSON schema.
Capability inventory: File system writes (src/theme.ts), network requests (fal.ai API), and browser automation (Playwright).
Sanitization: The skill implements a deterministic validation step using Zod schema validation and a WCAG contrast gate in scripts/extract-brand.ts to ensure the generated theme data is safe and valid before being applied.
[SAFE]: The skill demonstrates security best practices by implementing an escapeString function to prevent command or code injection when writing to theme.ts, and it explicitly instructs the agent and user to manage sensitive keys (FAL_KEY) and sessions (auth.json) locally without committing them to version control.

demoz-skill