The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes user-provided code snippets and titles by interpolating them directly into HTML templates for image rendering.
Ingestion points: Content from user requests is placed into  and  placeholders within assets/template.html and assets/grid-2x2.html.
Boundary markers: The instructions do not specify any boundary markers or instructions to the agent to ignore or escape instructions embedded within the user-provided code.
Capability inventory: The skill invokes a shell script that runs google-chrome in headless mode to render the HTML, which is loaded via the file:// protocol.
Sanitization: There is no evidence of sanitization or filtering for HTML/JavaScript tags in the ingested code, which could allow a malicious snippet to execute scripts or attempt to access local files during the rendering process.
[DYNAMIC_EXECUTION]: The skill generates and processes dynamic content by creating temporary HTML files from templates and executing a browser-based rendering pipeline.
Evidence: The workflow described in SKILL.md involves programmatically modifying HTML templates and then invoking scripts/render.sh to process them.
[COMMAND_EXECUTION]: The skill executes local system commands through a shell script to perform image rendering.
Evidence: scripts/render.sh executes google-chrome --headless with the --no-sandbox flag, and uses identify or file to verify the output PNG metadata.
[EXTERNAL_DOWNLOADS]: The skill's HTML templates reference official font resources from a well-known service.
Evidence: Both assets/template.html and assets/grid-2x2.html use CSS @import rules to fetch the 'JetBrains Mono' font family from Google Fonts (fonts.googleapis.com).

code-snippet-images