The Agent Skills Directory

[REMOTE_CODE_EXECUTION]: Automated scanners flagged instances of curl | python3 in comfyui_setup.sh and SKILL.md. Manual verification confirms these are false positives; the commands use the safe json.tool module (e.g., python3 -m json.tool) to pretty-print API responses for the user, which does not involve executing arbitrary code.
[COMMAND_EXECUTION]: The skill uses subprocess.run to interact with the local comfy-cli and perform hardware probes. These calls are executed using argument lists without shell=True, which is a secure practice that prevents command injection. The commands are limited to legitimate lifecycle management (installing, launching, and stopping the server).
[DATA_EXFILTRATION]: No unauthorized data exfiltration was detected. The skill includes a proactive security measure in scripts/_common.py that strips sensitive headers (including X-API-Key and Authorization) when an HTTP redirect crosses to a different host. This prevents authentication tokens from being leaked to external storage backends like S3 during file downloads.
[DATA_EXPOSURE]: The executor script scripts/run_workflow.py implements a safe_path_join utility. This ensures that any filenames provided by a remote ComfyUI server (which could contain path traversal sequences like ../) are restricted to the user-specified output directory, preventing arbitrary file writes.
[EXTERNAL_DOWNLOADS]: The skill facilitates the download of AI models and software dependencies from well-known and trusted platforms, including GitHub, HuggingFace, and Comfy Cloud. These downloads are central to the skill's purpose and are performed through the official comfy-cli tool.
[CREDENTIALS_UNSAFE]: No hardcoded secrets were found. The skill manages authentication via environment variables and placeholders, which is the standard and safe approach for handling API keys.

comfyui