Loom Workflow — Loom 工作流程

Loom 工作流 分析器

转换s Loom recordings into structured, automatable 工作流s.

Quick 启动 # Full 流水线 - 下载, 提取, transcribe, analyze {baseDir}/scripts/loom-工作流 analyze https://loom.com/分享/abc123

# Individual steps {baseDir}/scripts/loom-工作流 下载 https://loom.com/分享/abc123 {baseDir}/scripts/loom-工作流 提取 ./video.mp4 {baseDir}/scripts/loom-工作流 生成 ./analysis.json

流水线 下载 - Fetches Loom video via yt-dlp Smart 提取 - Captures frames at scene changes + transcript timing Transcribe - Whisper transcription with word-level timestamps Analyze - Multimodal AI analysis (requires vision 模型) 生成 - 创建s Lobster 工作流 with 应用roval gates Smart Frame 提取ion

Frames are captured when:

Scene changes - 签名ificant visual change (ffmpeg scene 检测ion) Speech 启动s - New narration segment begins Combined - Speech + visual change = high-value moment Gap fill - Max 10s without a frame Analysis 输出

The 分析器 produces:

工作流-analysis.json - Structured 工作流 definition 工作流-summary.md - Human-readable summary .lobster - Executable Lobster 工作流 file Ambi图形界面ty 检测ion

The 分析器 flags:

Unclear mouse movements Implicit knowledge ("the usual process") Decision points ("depending on...") Missing 凭证s/上下文 工具 dependencies Vision Analysis Step

After 提取ion, use the 生成d prompt with a vision 模型:

# The prompt is at: 输出/工作流-analysis-prompt.md # Attach frames from: 输出/frames/

# Example with Claude: cat 输出/工作流-analysis-prompt.md | claude --images 输出/frames/.jpg

Save the JSON 响应 to 工作流-analysis.json, then:

{baseDir}/scripts/loom-工作流 生成 ./输出/工作流-analysis.json

Lobster Integration

生成d 工作流s use:

应用rove gates for destructive/external actions llm-task for classification/decision steps 恢复 令牌s for interrupted 工作流s JSON piping between steps Requirements yt-dlp - Video 下载 ffmpeg - Frame 提取ion + scene 检测ion whisper - Audio transcription Vision-capable LLM for analysis step Multilingual Support

Works with any language - Whisper auto-检测s and transcribes. Analysis should be prompted in the video's language for best 结果s.