The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill ingests untrusted data from user briefs and project files which is then interpolated directly into prompts for AI generation and content for HTML rendering. A malicious brief could attempt to influence the generation logic or inject content into the rendered templates.
Ingestion points: SKILL.md (Step 1: Brief Intake) processes data from the user or campaign brief files.
Boundary markers: None identified; the skill does not appear to use delimiters or instructions to ignore embedded commands within the ingested data.
Capability inventory: The skill can execute shell commands (curl), write files to the project directory, and render HTML via Puppeteer/Playwright.
Sanitization: No sanitization or validation of user-supplied text before interpolation into prompts or HTML was observed.
[DYNAMIC_EXECUTION]: The skill generates HTML and CSS code at runtime based on brand guidelines and user briefs, which is subsequently executed in a browser environment (Puppeteer or Playwright) to render flyers.
Evidence: references/flyer-rendering.md details the architecture for building and rendering these dynamic templates.
[EXTERNAL_DOWNLOADS]: The skill fetches AI-generated assets from fal.ai's official API endpoints.
Evidence: references/generation-pipeline.md uses curl to interact with queue.fal.run to submit jobs and download the resulting image files.
[COMMAND_EXECUTION]: The skill utilizes several CLI utilities to process assets and communicate with remote services, including curl, jq, potrace, vtracer, and ImageMagick (convert).
These operations are used for standard asset conversion (bitmap-to-vector) and API interaction.

paw-cra-design-brand