SenseVoice Transcribe
Security checks across malware telemetry and agentic risk
Overview
The skill fits a local audio transcription workflow, with disclosed setup downloads, local transcript writes/deletes, and optional Discord progress notifications that users should review before use.
This appears safe to use for its stated purpose. Install it in the documented virtual environment, review the dependency/model sources, run --dry-run before batch work, be careful with --force-dates because it deletes transcript folders, and only use the Discord webhook if you are comfortable sharing progress metadata with that Discord destination.
VirusTotal
VirusTotal findings are pending for this skill version.
Risk analysis
Artifact-based informational review of SKILL.md, metadata, install specs, static scan signals, and capability signals. ClawScan does not execute the skill or run runtime probes.
Installing the skill’s dependencies may download and run third-party ML packages and cache large model files locally.
The setup relies on external Python packages and model downloads. This is expected for a transcription skill, but it means the user is trusting those package/model sources.
pip install funasr modelscope onnxruntime # Models auto-download on first run (~234MB SenseVoice + ~4MB VAD)
Use the documented virtual environment, consider pinning package versions, and verify the package/model sources before first use.
Using the force option with the wrong date could remove existing transcript outputs for that date.
The --force-dates option recursively removes the transcript directory for a selected date before re-transcribing. This is disclosed and scoped, but it can delete existing local transcript files.
if FORCE_DATES and not DRY_RUN: ... force_dir = TRANSCRIPT_DIR / fd ... shutil.rmtree(force_dir)
Run --dry-run first, keep backups of important transcripts, and use --force-dates only when you intend to replace that date’s transcript folder.
If enabled, transcription progress details such as dates, counts, and status may be posted to a Discord channel, and the webhook URL itself should be treated as sensitive.
The skill optionally sends progress notifications to a user-supplied Discord webhook. This is disclosed and optional, but it moves progress metadata outside the local machine.
--discord-webhook URL | Post start/milestone/finish to Discord
Only provide a trusted webhook URL, avoid using it for sensitive daylogs if progress metadata is private, and rotate the webhook if it is exposed.