The Agent Skills Directory

[COMMAND_EXECUTION]: The skill invokes ffmpeg and ffprobe to analyze and render video files. These calls are made using subprocess.run with argument lists rather than shell strings, which is a secure method that avoids shell injection vulnerabilities. The operations are consistent with the skill's stated purpose of video editing.\n- [INDIRECT_PROMPT_INJECTION]: The skill processes transcription data derived from user-provided video content. The agent is instructed to review cuts_log.json, which contains tokens from this data, to verify the automated edits. This represents a potential ingestion point for untrusted data that could influence the agent's behavior if malicious instructions were embedded in the transcript and misinterpreted during the review phase.\n
Ingestion points: Transcription JSON files and cuts_log.json containing text from source videos.\n
Boundary markers: None present; the agent is directed to read the log files directly.\n
Capability inventory: Shell command execution (via ffmpeg), file system modifications, and directory management.\n
Sanitization: The build_edl.py script applies regex-based filtering and character normalization to input tokens before they are logged or used in the editing process.

clean-cut