The Agent Skills Directory

[COMMAND_EXECUTION]: The skill uses subprocess.run across several scripts to perform system environment checks and manage the ComfyUI lifecycle. Specifically, scripts/hardware_check.py executes system commands like nvidia-smi and PowerShell's Get-CimInstance to detect GPU capabilities, while scripts/auto_fix_deps.py runs comfy node install to resolve missing dependencies.
[EXTERNAL_DOWNLOADS]: The skill facilitates the installation of the ComfyUI ecosystem. scripts/comfyui_setup.sh downloads the comfy-cli tool using standard Python package managers (pipx, uv), and the SKILL.md documentation guides users on downloading models from well-known repositories like HuggingFace and CivitAI. The automated scan's detection of piped execution in scripts/comfyui_setup.sh and SKILL.md refers to curl ... | python3 -m json.tool, which is a safe, standard method for formatting JSON output.
[INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest and process workflow files in JSON format, which can contain instructions for generative AI tasks.
Ingestion points: Workflow files are processed in scripts/run_workflow.py, scripts/extract_schema.py, and scripts/check_deps.py.
Boundary markers: No programmatic boundaries exist within the JSON format, but the skill provides a security advisory in SKILL.md regarding the trust profile of workflows from unknown sources.
Capability inventory: The skill possesses capabilities for shell command execution (in scripts/hardware_check.py and scripts/auto_fix_deps.py) and network communication (in scripts/_common.py).
Sanitization: scripts/run_workflow.py utilizes a safe_path_join function to ensure that downloaded output files do not perform path traversal attacks on the host system.

comfyui