openai-whisper
Local speech-to-text transcription for Apple Silicon machines using the MLX Whisper engine.
Install
mkdir -p .claude/skills/openai-whisper-qianleigood && curl -L -o skill.zip "https://agentskills.codes/api/skills/download/19451" && unzip -o skill.zip -d .claude/skills/openai-whisper-qianleigood && rm skill.zipInstalls to .claude/skills/openai-whisper-qianleigood
Activation
This is the description your AI agent reads to decide when to run this skill — the better it matches your request, the more reliably it fires.
Use when transcribing local audio or video on Apple Silicon without an API key, especially when MLX Whisper should be the default general-purpose local speech-to-text path.Key capabilities
- →Transcribe local audio and video files
- →Generate JSON or text transcripts
- →Support multiple Whisper model sizes
- →Execute transcription locally on Apple Silicon
How it works
The skill uses a shell script to invoke a Python-based MLX Whisper runtime, which processes media files locally on Apple Silicon hardware.
Inputs & outputs
When to use openai-whisper
- →Transcribe local audio recordings
- →Convert video file speech to text
- →Generate meeting transcripts locally
- →Process local media without cloud APIs
About this skill
MLX Whisper (Apple Silicon)
在 Apple Silicon（如 M2 Pro）上，优先将 mlx-whisper 视为默认的通用本地转写方案。
当前定位
- 通用主方案：默认本地音频/视频转写入口
- 平台特化：优先适配这台 Apple Silicon Mac
- 非短视频专用封装：如果只是要小红书/短视频链路里的窄场景包装，可再考虑更窄的专用 skill
运行方式
推荐通过技能目录下的 run.sh 调用：
./run.sh /path/to/audio-or-video.mp4
./run.sh /path/to/audio.m4a mlx-community/whisper-small
./run.sh /path/to/audio.m4a mlx-community/whisper-turbo --output-format json --output ./result.json
默认模型：
mlx-community/whisper-turbo
可选模型：
mlx-community/whisper-basemlx-community/whisper-small
输出
- 默认输出：纯文本（stdout）
--output-format txt：输出纯文本--output-format json：输出完整 JSON--output /path/file：写入文件而不是打印到终端
文件
run.sh：使用项目安装期准备的~/.crawclaw/runtimes/skill-openai-whisper/venv执行转写；可用CRAWCLAW_OPENAI_WHISPER_VENV覆盖 venv 路径transcribe_mlx.py：最小 MLX Whisper 调用入口
注意事项
run.sh会先检查 macOS Apple Silicon；其他平台请不要使用这个 skill- 如果 runtime 缺失，重新运行项目安装，或执行
crawclaw runtimes repair - 首次运行会下载模型到本地缓存，速度取决于网络
- 视频文件可直接输入；底层会处理音频解码
- 推荐默认使用
whisper-turbo，在 M2 Pro 上通常速度和效果更均衡 - 如果只是追求更轻量，可以改用
whisper-base
子代理执行策略
- 默认执行模式：优先使用子代理独立执行；主会话负责接单、补齐必要上下文、控制范围，并在最后整合结果回报给用户。
- 适合交给子代理的任务：预计超过 30 秒的处理、多步骤流水线、多文件/多产物生成、批处理、审计、转写、总结、离线分析。
- 主会话保留职责：只在必要时追问关键缺口、确认边界、挑选最终结果，并把输出改写成面向用户的完成答复。
- 不建议默认交给子代理的情况：一次性极短任务、需要高频来回确认的对话、强依赖当前聊天即时上下文的动作。
- 回报节点：默认只保留“已受理 / 已开始 / 已完成 / 已失败”四类关键节点，避免刷屏。
When not to use it
- →Non-Apple Silicon platforms
- →One-time extremely short tasks
- →Tasks requiring high-frequency conversational confirmation
Prerequisites
Limitations
- →Requires macOS Apple Silicon
- →Initial model download depends on network speed
- →Not suitable for non-Apple Silicon platforms
How it compares
Unlike cloud-based APIs, this approach performs transcription entirely offline on local hardware without requiring an API key.
Compared to similar skills
openai-whisper side by side with the closest alternatives in the catalog.
| Skill | Installs | Updated | Safety | Difficulty |
|---|---|---|---|---|
| openai-whisper (this skill) | 0 | 2mo | Review | Beginner |
| robotics-code-generator | 14 | 7mo | No flags | Advanced |
| modal | 5 | 7mo | Review | Intermediate |
| hugging-face-cli | 3 | 6mo | Review | Intermediate |
Try saying
Example prompts that trigger this skill in your AI assistant.
You might also like
robotics-code-generator
HumaizaNaz
Generates clean, runnable ROS 2, Gazebo, Isaac Sim, and VLA code for humanoid robotics
modal
davila7
Run Python code in the cloud with serverless containers, GPUs, and autoscaling. Use when deploying ML models, running batch processing jobs, scheduling compute-intensive tasks, or serving APIs that require GPU acceleration or dynamic scaling.
hugging-face-cli
patchy631
Execute Hugging Face Hub operations using the `hf` CLI. Use when the user needs to download models/datasets/spaces, upload files to Hub repositories, create repos, manage local cache, or run compute jobs on HF infrastructure. Covers authentication, file transfers, repository creation, cache operations, and cloud compute.
computer-use-agents
davila7
Build AI agents that interact with computers like humans do - viewing screens, moving cursors, clicking buttons, and typing text. Covers Anthropic's Computer Use, OpenAI's Operator/CUA, and open-source alternatives. Critical focus on sandboxing, security, and handling the unique challenges of vision-based control. Use when: computer use, desktop automation agent, screen control AI, vision-based agent, GUI automation.
machine-learning-ops-ml-pipeline
sickn33
Design and implement a complete ML pipeline for: $ARGUMENTS
ray-train
davila7
Distributed training orchestration across clusters. Scales PyTorch/TensorFlow/HuggingFace from laptop to 1000s of nodes. Built-in hyperparameter tuning with Ray Tune, fault tolerance, elastic scaling. Use when training massive models across multiple machines or running distributed hyperparameter sweeps.