The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes user-provided Markdown and JSON content, which is an inherent surface for indirect prompt injection or XSS.
Ingestion points: Reads artifact files (MD/JSON) and sidecar state files as input to the rendering process.
Boundary markers: The skill uses a structured prompt for the independent Codex review step, although it does not implement formal delimiters for the processed content itself.
Capability inventory: The skill has read/write file access and can execute shell commands via the Bash tool.
Sanitization: The rendering script (render_html.py) implements both standard HTML escaping and a dedicated sanitization function to strip dangerous tags (e.g., <script>, <iframe>) and event handlers.
[EXTERNAL_DOWNLOADS]: The generated reports fetch styling and mathematical rendering assets from cdn.jsdelivr.net, which is a well-known and trusted content delivery network.
[COMMAND_EXECUTION]: The skill uses git commands to resolve repository-relative paths for metadata inclusion. This is a standard project operation and uses static command arguments.

render-html