The Agent Skills Directory

[COMMAND_EXECUTION]: The skill executes multiple local shell scripts (narrate.sh, finalize.sh, setup_tts.sh) and Python scripts (asr.py, build_vo.py, timing_md.py) from a directory defined by the $PAPERCUT_LIB environment variable. This is standard behavior for a local developer automation tool.
[DATA_EXPOSURE]: The skill reads from ~/.config/papercut-explainer/config.env to resolve library paths, which is expected for local configuration management.
[INDIRECT_PROMPT_INJECTION]: The skill ingests data from narration/script.json and papercut/episode.json which contains text and instructions for AI models (Qwen3-TTS). This represents a vulnerability surface for influencing model output, but it is the intended purpose of the tool.
Ingestion points: narration/script.json, papercut/episode.json in the local project directory.
Boundary markers: None mentioned in the instructions.
Capability inventory: Local subprocess execution of shell/Python scripts and audio processing tools (ffmpeg, say).
Sanitization: Not mentioned.

papercut-voice