Aj Openai Whisper

Local speech-to-text with the Whisper CLI (no API key).

Security Scan

The skill is consistent with a local Whisper CLI wrapper: it only needs the 'whisper' binary, provides straightforward usage, and requests no credentials, though you should verify the Homebrew formula origin and be aware of model downloads and a minor naming inconsistency in the docs.

Purpose & Capability

Name/description, required binary ('whisper'), and the brew install for 'openai-whisper' align with a local Whisper CLI transcription skill. No extraneous credentials, binaries, or config paths are requested.

Instruction Scope

SKILL.md only tells the agent to run the local 'whisper' CLI on audio files (transcription/translation). It explicitly notes that models download to ~/.cache/whisper on first run — expected but means network downloads and disk writes. Also the doc claims '--model' defaults to 'turbo', which is not a standard Whisper model name (tiny/base/small/medium/large) and looks like an inconsistency to verify.

Install Mechanism

Install uses a Homebrew formula ('openai-whisper') which is a low-to-moderate risk install method when the formula is from a trustworthy tap. The skill does not include direct download URLs or extracted archives. You should verify which tap/formula will be used (official homebrew/core vs third-party tap).

Credentials

No environment variables, credentials, or config paths are requested. This is proportional for a local CLI transcription skill.

Persistence & Privilege

Skill does not request always:true and is user-invocable only. It does not attempt to modify other skills or system-wide settings in the provided instructions.

Assessment

This skill appears to be what it says: a thin instruction layer for using a local 'whisper' CLI. Before installing or running it: (1) verify the Homebrew formula origin (run 'brew info openai-whisper' and inspect the formula or the tap) to ensure you're installing from a trusted source; (2) expect the model files to be downloaded to ~/.cache/whisper on first run — downloads can be large and require network access and disk space; (3) confirm the 'whisper' implementation's behavior (which hosts it downloads models from) if you need offline-only operation; (4) note the SKILL.md mentions a '--model' default of 'turbo', which is likely incorrect — verify the model names supported by the installed 'whisper' binary; (5) because transcription processes audio data, avoid running it on highly sensitive audio on an untrusted machine or without verifying the binary. If you want greater assurance, install Whisper manually from the official upstream repository or inspect the Homebrew formula before proceeding.

Whisper (CLI)

Use whisper to transcribe audio locally.

Quick start

• whisper /path/audio.mp3 --model medium --output_format txt --output_dir .
• whisper /path/audio.m4a --task translate --output_format srt

Notes

• Models download to ~/.cache/whisper on first run.
• --model defaults to turbo on this install.
• Use smaller models for speed, larger for accuracy.