录制
Playwright 驱动无头 Chromium,按 story 点击、输入、滚动、等待。临时实例独占端口和数据,绝不动正在使用的服务。
Claude Code / Codex 技能 · 开源 · MIT
写一个 Python 分镜文件,自动完成录制、1080p60 渲染和逐项核验。旁白可有可无:语言(普通话、英语等十种)和引擎(本机 GPU、本机 CPU、远程 API)由你选。支持 Windows、macOS、Linux。
Playwright 驱动无头 Chromium,按 story 点击、输入、滚动、等待。临时实例独占端口和数据,绝不动正在使用的服务。
逐帧合成 1920×1080、60 fps 视频:推近镜头、框选元素、编号步骤、原文高亮、章节条和字幕。
旁白可选:本机 Qwen3-TTS(GPU)、本机 Kokoro(CPU)、远程 API(OpenAI 兼容 / ElevenLabs / 百炼)。Whisper 重听、CAM++ 核对音色、包哈希证明画面未被改动。
git clone https://github.com/OrrrGE/autoRecording \
~/.claude/skills/autoDemoRecordingSkill
cd ~/.claude/skills/autoDemoRecordingSkill
python3 scripts/demo_env.py setup
PY=~/.cache/auto-demo-recording/venv/bin/python
$PY scripts/demo_record.py --story examples/selftest/story.py --work /tmp/demo-selftest
$PY scripts/demo_render.py --work /tmp/demo-selftest --check
$PY scripts/demo_render.py --work /tmp/demo-selftest --out /tmp/demo-selftest/selftest.mp4
$PY scripts/demo_voice.py --work /tmp/demo-selftest
录制和渲染不需要显卡。最后一条配音可选:有 NVIDIA 显卡用 Qwen,没有就用 Kokoro(demo_env.py setup --kokoro)或远程 API;不要旁白就跳过。环境检查:python3 scripts/demo_env.py。
用另一个端口和独立数据目录启动副本,例如 DATA_DIR=/tmp/demo-data python -m myapp --port 8799。把数据目录写进 story 的 wipe,每次录制都从相同状态开始。
复制 story_template.py,填 TITLE、CHAPTERS、launch(),然后用 d.say、d.click、d.zoom、d.mark 写分镜。一条字幕一个意思,字幕同时就是解说稿。
demo_record.py --quick 快速抓选择器和状态问题;正式录制后运行 demo_render.py --check,按输出修正,直到没有 注意:。
demo_render.py --out 生成无声成片;demo_voice.py 加普通话解说。配音检查以 没有发现问题 结尾才算完成。
改字幕、标注、镜头:重录 → 重渲染 → 重配音。只改配音设置:demo_voice.py build。某一句念错:删掉它的 take,再 demo_voice.py takes N。
Claude Code / Codex skill · Open source · MIT
Write one Python story file. The pipeline records the app, renders a 1080p60 video and verifies it numerically. The narration is optional: pick its language (ten, Mandarin and English among them) and its engine (local GPU, local CPU, or a remote API). Windows, macOS and Linux.
Playwright drives headless Chromium through the story: clicks, typing, scrolling, waiting. The app runs as a throwaway instance with its own port and data, never the live service.
Every frame is composited at 1920×1080, 60 fps: camera moves, element outlines, numbered steps, original-text highlights, chapters and captions.
Optional narration by local Qwen3-TTS (GPU), local Kokoro (CPU) or a remote API (OpenAI-compatible / ElevenLabs / DashScope). Whisper re-listens, CAM++ checks the voice, packet hashes prove the picture is untouched.
git clone https://github.com/OrrrGE/autoRecording \
~/.claude/skills/autoDemoRecordingSkill
cd ~/.claude/skills/autoDemoRecordingSkill
python3 scripts/demo_env.py setup
PY=~/.cache/auto-demo-recording/venv/bin/python
$PY scripts/demo_record.py --story examples/selftest/story.py --work /tmp/demo-selftest
$PY scripts/demo_render.py --work /tmp/demo-selftest --check
$PY scripts/demo_render.py --work /tmp/demo-selftest --out /tmp/demo-selftest/selftest.mp4
$PY scripts/demo_voice.py --work /tmp/demo-selftest
Recording and rendering need no GPU. The last command is optional: Qwen with an NVIDIA GPU, Kokoro on any CPU (demo_env.py setup --kokoro), or a remote API — or skip it for a silent video. Check the machine with python3 scripts/demo_env.py.
Start a copy on another port with its own data directory, for example DATA_DIR=/tmp/demo-data python -m myapp --port 8799. Put that directory in the story's wipe list so every take starts from the same state.
Copy story_template.py, set TITLE, CHAPTERS and launch(), then direct with d.say, d.click, d.zoom and d.mark. One idea per caption; captions are also the narration script.
demo_record.py --quick catches bad selectors and page state fast. Record for real, then run demo_render.py --check and fix until it prints no 注意: problems.
demo_render.py --out writes the silent video; demo_voice.py adds the Mandarin narration. The narration is done only when its check ends with 没有发现问题.
Captions, marks or camera: re-record, re-render, re-narrate. Narration settings only: demo_voice.py build. One bad line: delete its takes, then demo_voice.py takes N.