The Agent Skills Directory

[COMMAND_EXECUTION]: The skill uses subprocess.run to execute ffprobe for media metadata extraction and to call the filebrowser CLI through uv run. These calls are implemented using argument lists rather than shell strings, which mitigates many command injection risks.
Evidence in scripts/upscale_cli/service.py (_run_ffprobe function).
Evidence in scripts/upscale_cli/filebrowser.py (FileBrowserGateway._run method).
[INDIRECT_PROMPT_INJECTION]: The skill processes external image and video files using libraries like Pillow and ffprobe. While these operations are necessary for the skill's functionality, they represent a potential attack surface if the libraries have unpatched vulnerabilities to maliciously crafted media files. The skill mitigates this by validating file types and sanitizing filenames.
Evidence of filename sanitization in scripts/upscale_cli/naming.py using regex to strip dangerous characters.
[DATA_EXPOSURE_AND_EXFILTRATION]: The skill implements strong guidelines for handling sensitive information. It explicitly instructs the agent not to output authentication headers, API keys, or tokens in its responses.
Evidence in SKILL.md under the "Must Comply" (必须遵守) section.
Evidence in scripts/upscale_cli/config.py where secrets are resolved from environment variables rather than being hardcoded.
[PATH_TRAVERSAL]: The integration with FileBrowser includes explicit checks to prevent directory traversal attacks.
Evidence in scripts/upscale_cli/filebrowser.py (normalize_remote_path function) which checks for .. in remote paths.

upscale