The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill instructs the agent to construct and execute shell commands using external data, such as file paths and names provided by users or found in the environment.
Ingestion points: User-provided file paths, filenames, and media metadata processed using the commands described in SKILL.md and reference.md.
Boundary markers: No boundary markers or explicit instructions for handling untrusted input are provided.
Capability inventory: Extensive shell command execution capabilities via ffmpeg and ffprobe are detailed in both SKILL.md and reference.md.
Sanitization: There is no evidence of input sanitization or escaping instructions, which creates a potential surface for command injection if malicious filenames (e.g., using shell metacharacters) are processed.
[COMMAND_EXECUTION]: The skill is primarily centered around the execution of complex CLI tools for media transformation.
Evidence: The skill contains numerous bash examples for format conversion, resizing, and filtering in SKILL.md and reference.md.
[EXTERNAL_DOWNLOADS]: The skill contains a reference to an external GitHub repository for project contributions.
Evidence: Mentions github.com/digitalsamba/claude-code-video-toolkit as a target for feedback and contributions in SKILL.md.

ffmpeg