Auto Whisper Safe

• Risk level: Low
• Analyzed: Feb 14, 2026, 6:18 AM
• Findings: 0
• Version: 1.1.0

Overview

The skill's files, instructions, and requirements are consistent with a RAM-safe Whisper-based transcription helper; no signs of hidden exfiltration or unrelated privileges, though there are minor implementation inconsistencies and a privacy/automation consideration to review before enabling automatic use.

This skill appears to do exactly what it says: chunk long audio, run the local whisper CLI, and merge transcripts. Before installing or enabling automatic transcription, check the following: (1) Confirm you have the whisper CLI and ffmpeg/ffprobe installed from trusted sources; the script calls ffprobe but only checks for ffmpeg presence. (2) Fix the minor help-text inconsistency (script prints 'default: de' but actually defaults to 'en') if you rely on the default language. (3) Consider the privacy implications of adding the 'ALWAYS transcribe' BOOTSTRAP entry — automatic transcription of user audio may expose private content to logs or downstream processing; only enable automatic behavior if you trust the environment and want all audio transcribed. (4) If you need larger models, verify RAM availability before switching MODEL. Overall the skill is coherent and contains no obvious hidden network calls or credential requests.