The Agent Skills Directory

[DYNAMIC_EXECUTION]: The script scripts/fx.py contains a call to exec() that dynamically executes the content of a file named fx_private.py if it exists in the skill's root directory. This allows for the execution of arbitrary Python code from a file that is not part of the primary skill logic.
[COMMAND_EXECUTION]: The skill frequently uses subprocess.run() and subprocess.Popen() to interact with system utilities such as ffmpeg, ffprobe, and pip. These tools are used for video metadata probing, frame extraction, and rendering.
[REMOTE_CODE_EXECUTION]: The script scripts/words.py performs a runtime installation of the faster-whisper Python package using pip. While targeting a known library, installing code at runtime is a high-risk pattern that bypasses static dependency verification.
[INDIRECT_PROMPT_INJECTION]: The SKILL.md instructions guide the agent to visually verify rendered frames (check.png) which contain user-provided text. This creates a vulnerability where a user could inject instructions into the video text that the agent might follow upon visual inspection of the frames.
Ingestion points: User-provided video content and effect text options in SKILL.md.
Boundary markers: None used for the visual inspection phase.
Capability inventory: File system access, shell execution, and Python execution.
Sanitization: No sanitization of user-provided text before rendering and visual inspection.
[EXTERNAL_DOWNLOADS]: The scripts/setup.sh and scripts/words.py scripts download and install various system-level dependencies and Python libraries required for video editing and face tracking.

effects