The Agent Skills Directory

[COMMAND_EXECUTION]: The script scripts/assemble.py executes ffmpeg and ffprobe binaries to perform video assembly and metadata probing. It uses the subprocess.run function with command arguments passed as a list, which is a secure practice that prevents shell injection vulnerabilities.
[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted external data, specifically transcript.json and edl-lines.json. These files, which may contain content from automated speech-to-text (ASR) engines or user-supplied timing data, are used as inputs for video assembly and quality assurance.
Ingestion points: Data is read from files in scripts/assemble.py (via the load_inputs function) and referenced throughout the instructions in SKILL.md for editing tasks.
Boundary markers: The skill uses structured JSON formats to define data boundaries, reducing the risk of instructions being accidentally misinterpreted as control commands.
Capability inventory: The skill has the capability to execute local ffmpeg and ffprobe commands and write various JSON and MP4 artifacts to a designated project directory (_cut_work/).
Sanitization: The scripts/assemble.py script performs rigorous type checking, range validation, and finite number checks on all numeric inputs (timestamps, durations, indices) before passing them to the rendering engine.
[DYNAMIC_EXECUTION]: The scripts/smoke-test.py script uses importlib.util to dynamically load the assemble.py module for regression testing. This is a standard and safe practice for local Python development and testing within a controlled skill pack environment.

recap-video