The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill takes user-provided input (headlines, colors, logo URLs) and interpolates them directly into HTML templates without sanitization.
Ingestion points: User inputs for headlines, brand colors, logo URLs, and product image paths gathered in Step 1 of SKILL.md.
Boundary markers: No delimiters or instructions are provided to the agent to treat user-provided text as data only, increasing the risk that the agent or the rendering engine interprets embedded scripts.
Capability inventory: The skill performs file system writes (ad-{slug}.html) and utilizes the Playwright MCP tool to render and screenshot the generated file.
Sanitization: No escaping, validation, or filtering of external content is specified before interpolation into the HTML templates.
[DYNAMIC_EXECUTION]: The skill dynamically constructs HTML files (which may include JavaScript) from templates and user inputs, then triggers a browser-based rendering process to capture screenshots.
[COMMAND_EXECUTION]: The skill instructs the agent to perform file writes and interact with the Playwright MCP tool, specifically recommending the --allow-unrestricted-file-access flag which lowers the browser's security sandbox.

ad-creative