The Agent Skills Directory

[COMMAND_EXECUTION]: The script scripts/render.js uses child_process.execSync to call the convert utility (ImageMagick) for image compression. The command incorporates variable paths (rawPng, outputFile) which are manipulated within the script and could potentially be subject to command injection if filenames are derived from untrusted user input.
[DATA_EXFILTRATION]: The workflow involves reading local image files from the user's workspace using fs.readFileSync (e.g., in scripts/example_render.js) to convert them into Base64 strings for embedding in HTML. While intended for asset management, this demonstrates access to the local file system.
[DYNAMIC_EXECUTION]: The skill relies on generating and running Node.js scripts that utilize the Playwright library to render dynamically generated HTML into images. This involves runtime execution of code constructed from templates.
[INDIRECT_PROMPT_INJECTION]: The skill processes text to generate card content which is directly interpolated into HTML templates (as seen in scripts/render.js and scripts/example_render.js). This creates an attack surface for indirect prompt injection.
Ingestion points: Text content extracted from user documents or workspace files in Stage 1 of the workflow.
Boundary markers: None identified in the HTML generation scripts to delimit untrusted data from the template logic.
Capability inventory: The skill has access to the workspace via Bash(*), Write, and Edit tools. The Node.js scripts utilize fs for file reads, child_process for shell execution, and playwright for browser automation.
Sanitization: There is no evidence of HTML escaping or sanitization of the title, body, or subtitle variables before they are injected into the HTML templates rendered by Playwright.

Buda Social Cards