← 返回 Skills 市场
214
总下载
0
收藏
0
当前安装
1
版本数
在 OpenClaw 中安装
/install qqbot-stt
功能描述
在本地运行 Qwen3-ASR 模型，通过 HTTP 服务为 QQBot 提供精准的语音转文字功能。
安全使用建议
Before installing or running this skill:
- Treat it as suspicious until you verify sources: the SKILL.md and files refer to 'local-stt' while registry metadata says 'qqbot-stt' — verify you obtained the intended package.
- Do not run the server or scripts as root. Run them in an isolated environment (dedicated user account or container/VM).
- Inspect and confirm the file referenced by main.py (QWEN_ASR_SCRIPT = /Users/reks/.openclaw/skills/qwen-asr/scripts/main.py). Hard-coded absolute paths are unexpected and could point to mispackaging; ensure that path is what you expect or change it to the correct local script.
- The requirements.txt is incomplete. Expect to manually install heavy ML packages (torch/transformers or an 'mlx_qwen3_asr' package) and ffmpeg; these will pull code from PyPI/HuggingFace (including 'trust_remote_code=True' in suggested code), which increases supply-chain risk — review those packages' sources first.
- The code runs subprocesses (ffmpeg, model CLI) and spawns other Python modules. Review any third-party module (especially 'mlx_qwen3_asr' or 'qwen-asr') for network behavior before allowing model downloads.
- If you proceed, run first in an isolated VM or container, confirm it only binds to localhost and does not exfiltrate data, and verify logs and network connections during the initial model download and runtime.
If you want, provide the upstream repository URL or the contents of the referenced qwen-asr scripts; that additional context would raise or lower confidence in this assessment.
功能分析
Type: OpenClaw Skill
Name: qqbot-stt
Version: 1.0.0
The bundle provides a legitimate local Speech-to-Text (STT) service for the OpenClaw framework using the Qwen3-ASR model. It contains a FastAPI server (server.py) and CLI utilities (transcribe.py) designed to process audio files and return transcriptions, with specific support for Apple Silicon via the MLX framework. While main.py contains a hardcoded absolute path (/Users/reks/...) which is a portability flaw, the overall code logic is transparent, uses safe subprocess handling for ffmpeg, and lacks any indicators of malicious intent, data exfiltration, or unauthorized remote control.
能力评估
Purpose & Capability
The README / SKILL.md describe a 'local-stt' skill providing a Qwen3-ASR HTTP/CLI STT service for QQBot, which is coherent with the included code. However the package metadata says 'qqbot-stt' while files and instructions repeatedly refer to 'local-stt' (name mismatch). main.py hard-codes QWEN_ASR_SCRIPT to /Users/reks/.openclaw/skills/qwen-asr/scripts/main.py (a user-specific absolute path) which does not match the skill's own layout and is unexpected for a distributable skill.
Instruction Scope
SKILL.md asks you to clone/install a different skill (local-stt) and edit ~/.openclaw/openclaw.json — that is expected — but it also presumes external toolchains (HuggingFace model downloads, ffmpeg) while requirements.txt does not list the model/ML libs. The instructions and code reference an env var MLX_ASR_MODEL and rely on external network/model downloads; SKILL.md and requires.env declare no required env vars, so the runtime assumptions are under-specified. The docs also encourage running openclaw gateway and grepping logs (harmless) but give the agent broad leeway to run system commands during setup.
Install Mechanism
No install spec is provided (instruction-only), yet the package includes executable code files that must be run manually. requirements.txt only lists fastapi, uvicorn, python-multipart but the code imports mlx_qwen3_asr and uses model-serving libraries (transformers/torch) and ffmpeg; those dependencies are missing from requirements.txt. Running pip install -r requirements.txt will not install the packages actually required, and following the README will cause manual installs that fetch heavy third-party ML packages (and arbitrary code via 'trust_remote_code=True'). This increases risk because additional packages will be pulled from PyPI/HuggingFace at runtime.
Credentials
The skill declares no required env vars, but server.py and transcribe.py read MLX_ASR_MODEL and code expects ffmpeg on PATH. There are no API keys requested, which is proportionate, but the hidden dependency on MLX_ASR_MODEL (and possible use of huggingface credentials when downloading models) is not surfaced in the metadata.
Persistence & Privilege
always is false; the skill does not request forced inclusion or elevated platform privileges. It runs as a normal local service/CLI, so persistence/privilege requests are appropriate for its purpose.
如何使用
- 确保已安装 OpenClaw（本地或 Docker 部署）
- 在对话框中输入安装命令：
/install qqbot-stt - 安装完成后，直接呼叫该 Skill 的名称或使用
/qqbot-stt触发 - 根据 Skill 的参数说明提供必要输入，即可获得结构化输出
版本历史
v1.0.0
qqbot-stt 1.0.0
- Initial release of local STT (speech-to-text) integration for OpenClaw + QQBot.
- Provides a complete setup guide for running Qwen3-ASR locally as an HTTP transcription service.
- Includes configuration instructions for proper OpenClaw integration, emphasizing correct placement in `channels.qqbot.stt`.
- Offers troubleshooting steps for common setup and configuration issues.
- Supports Apple Silicon acceleration with mlx-qwen3-asr.
元数据
常见问题
qqbot-stt 是什么？
在本地运行 Qwen3-ASR 模型，通过 HTTP 服务为 QQBot 提供精准的语音转文字功能。 它是一个面向 Claude Code / OpenClaw 的 AI Agent Skill 插件，目前累计下载 214 次。
如何安装 qqbot-stt？
在 OpenClaw 或 Claude Code 对话框中运行命令「/install qqbot-stt」即可一键安装，无需额外配置。
qqbot-stt 是免费的吗？
是的，qqbot-stt 完全免费，采用 MIT-0 许可证，可自由下载、安装和使用。
qqbot-stt 支持哪些平台？
qqbot-stt 跨平台运行，可在任意部署了 OpenClaw / Claude Code 的环境中使用（cross-platform）。
谁开发了 qqbot-stt？
由 rekslee（@rekslee）开发并维护，当前版本 v1.0.0。
推荐 Skills