The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes freeform user briefs in references/01-brief-and-context.md which are used to construct prompts for image generation models in references/02-production.md.
Ingestion points: The Brief Intake section in references/01-brief-and-context.md accepts untrusted parameters such as platform, format, and content/copy.
Boundary markers: Absent. There are no explicit delimiters or instructions provided to the agent to ignore potentially malicious instructions embedded within the user-provided brief.
Capability inventory: The skill utilize shell commands for network requests (via curl), file system write access for saving assets and manifests (references/03-validation-and-export.md), and browser-based rendering via Puppeteer or Playwright (references/02-production.md).
Sanitization: Absent. The content from the user brief is interpolated into design prompts and template parameters without evidence of escaping or validation.
[COMMAND_EXECUTION]: The workflow utilizes shell commands including curl and jq to interact with external APIs and process JSON responses during the production stage (references/02-production.md).
[DYNAMIC_EXECUTION]: The skill employs browser automation tools (Puppeteer or Playwright) to render HTML and CSS templates into visual assets, which involves executing code in a browser environment to generate images from dynamic content.
[EXTERNAL_DOWNLOADS]: Fetches generated image assets from the fal.run API, which is a recognized external service for AI image generation.
[DATA_EXFILTRATION]: Transmits prompt data and authorization headers to the fal.ai API. This is the intended behavior for generating social media assets but involves sending brand and campaign context to an external provider.

paw-cra-design-social