菜单
Transcribe audio files to text using mlx_whisper on Apple Silicon. Use when the user has an audio file (MP3, WAV, AAC, OGG, etc.) they want transcribed.
用 Codex 或 Claude 帮你安装 复制这段 Prompt，粘贴到 Codex、Claude 或其他助手里，让它检查 Skill 页面并帮你完成安装。
基于 SOC 职业分类
Run consistent, findings-first code reviews with Codex CLI for any target: uncommitted/dirty working tree, current branch vs base branch, specific commit SHA, or pull request. Use when asked to review bugs, risks, regressions, and test gaps with actionable file/line findings.
Deep-dive into complex code using Codex CLI. Traces call chains, maps dependencies, and explains how code works. Use when asked to understand, unwind, or explain tangled or unfamiliar code.
Use Codex CLI to diagnose and fix a bug, error, or failing test. Use when you want a second opinion on a bug or when Claude's fix isn't working.
Use Codex CLI to refactor a module, file, or code pattern. Use when you want an independent take on how to restructure complex or tangled code.
| name | transcribe |
| description | Transcribe audio files to text using mlx_whisper on Apple Silicon. Use when the user has an audio file (MP3, WAV, AAC, OGG, etc.) they want transcribed. |
Transcribe audio files locally on Apple Silicon using mlx_whisper. Audio never leaves the device.
Identify the audio file path from the user's request. If not provided, ask.
mlx_whisper "<audio_file_path>" --model mlx-community/whisper-large-v3-turbo --output-format txt --output-dir /tmp/
whisper-large-v3-turbo by default for best qualitymlx-community/whisper-tiny instead--output-format json instead of txtThe output file will be written to /tmp/ with the same base name as the input file (e.g., /tmp/audio.txt).
Read the output file and present the transcription to the user.
If the transcription ends with repeated lines (a known Whisper hallucination artifact when audio trails off into silence), trim those from the output.
mlx_whisper must be installed. If not found, install it with:
uv tool install mlx-whisper