The Agent Skills Directory

[COMMAND_EXECUTION]: The skill uses subprocess.run to invoke external utilities for its primary functionality.
It calls ffmpeg and ffprobe in lane_speech_coach.py for audio analysis of rehearsal recordings.
it calls whisper-cli for speech-to-text transcription.
It calls node and chrome (in headless mode) in diagram_from_json.py to render JSON-based diagrams into PNG files.
These calls use constructed argument lists and target local files within the project structure.
[DYNAMIC_EXECUTION]: The tool employs dynamic loading of Python modules to handle project-specific logic and modular check "lanes".
In preprep.py, it uses importlib.util to load synchronization scripts (e.g., ifkakao_pull_notes.py) and several standalone lane modules.
This execution is constrained to the project's own directory or paths specified in the local surfaces.yaml configuration.
[INDIRECT_PROMPT_INJECTION]: The skill ingests and processes untrusted data from user-supplied manuscripts and slide files, creating a potential surface for indirect injection.
Ingestion points: Markdown manuscripts, PowerPoint (.pptx) notes, and JSON diagram specifications are read from the local file system (e.g., in preprep.py and lane_speech_coach.py).
Boundary markers: The skill uses specific Korean markers like 🖥 (screen text) and 🗣 (speech text) to delimit different sections within manuscript files.
Capability inventory: The skill possesses capabilities for file system operations (os.replace in safe_install.py) and local command execution as noted above.
Sanitization: Content is structurally parsed using regex and XML trees to isolate text before it is presented to the agent or processed by specialized lanes.

preprep