The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes external data such as upstream installers and workflow JSON files. These files are untrusted inputs that could potentially contain malicious links or commands.
Ingestion points: The skill reads data from upstream installer scripts (.bat/.sh) and workflow JSON files (workflow.json) to extract model URLs and node sources.
Boundary markers: No specific boundary markers or 'ignore' instructions are provided for the AI when parsing these external files.
Capability inventory: The skill uses the apply_manifest tool to install nodes and download models, and generates shell scripts for local installation. It also includes npm scripts for URL validation.
Sanitization: The instructions explicitly direct the AI to resolve model names to canonical Hugging Face URLs and verify they are live, flagging any that cannot be resolved. The system also mentions that contributions are reviewed for safety upstream.
[COMMAND_EXECUTION]: The skill describes the generation and execution of shell scripts (install-windows.bat, install-runpod.sh) for installing components. These scripts are intended for setup automation within the ComfyUI environment.
[EXTERNAL_DOWNLOADS]: The skill manages downloads of model weights and custom nodes, primarily targeting Hugging Face and the author's GitHub repository. These operations are core to the skill's purpose of providing installer packs and follow vendor-specific patterns.

installer-packs