The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill instructs the setup of a virtual environment and installation of well-known software packages playwright and weasyprint. It also downloads brand assets like fonts and logos from Vercel's public blob storage and the vendor's repository as listed in brand/brand.json and brand/finals.json.
[COMMAND_EXECUTION]: The skill executes local scripts brand/materialize.mjs (via Node.js) and assets/render.py (via Python). These scripts are used to stage brand resources and launch browser processes for PDF generation respectively.
[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted external data, such as consulting deal terms and pricing, by interpolating them into assets/proposal-template.html. The resulting file is then rendered by a full browser (Playwright/Chromium) in assets/render.py, creating a surface for potential XSS-to-PDF or data exposure attacks if the input data is malicious.
Ingestion points: External consulting deal copy, pricing, and agreement text as described in SKILL.md.
Boundary markers: The instructions in SKILL.md and assets/copy-and-design-frameworks.md advise escaping text values, but the templates and rendering scripts do not enforce technical boundary markers.
Capability inventory: assets/render.py utilizes Playwright, which permits JavaScript execution, and performs file system writes to the workspace.
Sanitization: The skill relies on the agent following natural language instructions to escape text values during the template-filling process; no programmatic sanitization is evident in the rendering script.

recoup-internal-consulting-proposal-designer