The Agent Skills Directory

[SAFE]: The skill performs localized video editing tasks using standard system tools like FFmpeg and FFprobe. All file operations are directed to local paths specified in the configuration.\n- [SAFE]: Robust integrity verification is implemented in scripts/replace_hook.py. It uses PSNR (Peak Signal-to-Noise Ratio) to ensure retained video frames are identical within encoding tolerances and measures audio correlation to prevent unintended changes to the sound track.\n- [SAFE]: Input validation is thorough throughout the scripts. Numerical inputs (timestamps) are checked for validity and positivity, and color codes are validated using regular expressions. The preflight_source function verifies the integrity and format of the source video before any processing begins.\n- [COMMAND_EXECUTION]: The skill utilizes subprocess.run to invoke ffmpeg and ffprobe for media processing. These calls are constructed as lists (avoiding shell execution) and use paths validated through the resolve utility function.\n- [INDIRECT_PROMPT_INJECTION]: The skill processes untrusted media and metadata. Evidence chain:\n
Ingestion points: Reads config.json, source video files, words JSON, and SRT/JSON captions in scripts/replace_hook.py.\n
Boundary markers: SKILL.md mandates manual review ('needs_review' status) and examination of preservation booleans in the manifest.\n
Capability inventory: File read/write and subprocess.run for media tools in scripts/replace_hook.py.\n
Sanitization: Includes strict schema validation, path resolution via resolve(), and high-tolerance decoded frame comparisons to verify the body was not altered.

render-hook-replacement