Vocal Isolation, Background Music Removal then De-Noise

Vocal isolation / background music removal on remote (FREE) L4 GPU. Trigger when user says: isolate vocals, remove background music, extract voice, 提取人声, 去除背...

This skill appears to do what it says (remote Demucs + ClearerVoice inference on Modal), but you should be cautious before installing or running it: - Do NOT run isolate.py directly on your local machine unless you audit the code first. The code will remove and replace ~/.cache (shutil.rmtree on cache_dir) and manipulate ~/checkpoints (rmdir + symlink), which could delete or change files in your home directory if executed locally. The SKILL.md does not mention these destructive filesystem changes. - Prefer running via 'modal run' as intended (the code is designed for Modal remote containers). If you will run anywhere else, inspect and modify the symlink / rmtree logic so it cannot touch your real home directory. - The image build will pip-install large ML packages (torch, demucs, clearvoice). That is expected for the task but be aware it pulls code and models from PyPI/HF at build/runtime; if you require provenance, ask the author for exact package sources/versions or a reproducible image. - The docs mention HF_TOKEN for higher-rate model downloads; you may need to provide that to avoid download failures — the skill does not declare it as required in metadata. - Ask the skill author to (a) explicitly document the symlink/deletion behavior and why it is safe within Modal, (b) avoid destructive operations on Path.home() or at least guard them so they only run inside the Modal environment, and (c) fix minor docs inconsistencies (npx mention, Python version mismatch). If you can't get these clarifications, treat the skill as untrusted for local execution and only run it in a controlled environment (isolated account or throwaway container) after code review.

Type: OpenClaw Skill Name: speech-isolate Version: 1.3.1 The skill bundle provides a legitimate pipeline for vocal isolation and speech enhancement using the Modal cloud platform. The code in `isolate.py` and `src/images.py` correctly implements the two-stage processing (Demucs and ClearerVoice) and uses standard Modal CLI commands for data management as described in `SKILL.md`. No evidence of data exfiltration, malicious execution, or prompt injection was found.