The Agent Skills Directory

[REMOTE_CODE_EXECUTION]: The skill provides instructions to install the higgsfield CLI by piping a shell script from the vendor's official GitHub repository (higgsfield-ai/cli) to the shell. This is a standard installation method for the vendor's own tooling.
[EXTERNAL_DOWNLOADS]: Fetches a canonical Brandbook template from Google Docs and downloads font files from Google Fonts when required for rendering. These are well-known and trusted services.
[COMMAND_EXECUTION]: Utilizes Python's subprocess.run to execute local binaries for image conversion and document processing, including ImageMagick (magick/convert), rsvg-convert, LibreOffice (soffice), and Poppler utilities (pdffonts, pdftoppm). These calls use list-based arguments to prevent shell injection.
[INDIRECT_PROMPT_INJECTION]: The skill has an attack surface for indirect injection as it ingests and processes untrusted user assets (SVGs, PDFs, PPTXs). However, it implements significant mitigations:
Ingestion points: User-supplied files and local paths analyzed in references/asset-analysis.md.
Boundary markers: Uses explicit delimiters like [BRAND LOCK — DO NOT DEVIATE] in prompts to separate brand constraints from generation instructions.
Capability inventory: Extensive use of subprocess calls for media manipulation across all scripts.
Sanitization: The validate_svg function in scripts/brandkit.py actively strips potentially malicious content from SVGs, such as tags, foreignObject, and event handlers.

higgsfield-brandkit