The Agent Skills Directory

[COMMAND_EXECUTION]: The skill utilizes several Python scripts (audio-inspect, extract-review-frames, ffmpeg-preflight, media-intake) to execute local ffmpeg and ffprobe commands via the subprocess module. This is the primary intended purpose of the skill. The implementation follows security best practices by using list-based arguments instead of shell strings to prevent command injection, and includes hardcoded resource bounds such as MAX_FILES_HARD, MAX_TIMEOUT_HARD, and MAX_OUTPUT_BYTES_HARD to prevent resource exhaustion or denial-of-service attacks.
[INDIRECT_PROMPT_INJECTION]: The skill handles untrusted external data in the form of media files and transcripts. It mitigates potential risks by defining a strict 'Evidence-driven media work' workflow (described in references/media-intake-and-manifest.md) that separates technical measurements from editorial interpretations. It explicitly instructs the agent not to infer meaning or identity from technical metadata and requires human review for editorial decisions.
[REMOTE_CODE_EXECUTION]: No remote code execution patterns were detected. All tool execution is limited to local binaries (ffmpeg, ffprobe) and local Python scripts that do not fetch or execute remote payloads.
[DATA_EXFILTRATION]: The skill documentation (references/advanced-operations-and-safety.md) explicitly warns against sending sensitive data over network protocols and recommends restricting network access. It emphasizes the use of a private task workspace to contain media and artifacts.
[PRIVILEGE_ESCALATION]: The skill does not perform any operations requiring elevated privileges. It uses -n (no-overwrite) by default in FFmpeg commands to prevent accidental file modification.

ffmpeg