loom-workflow

What This Skill Does

The Loom Workflow Analyzer is an AI-native tool designed to bridge the gap between human process documentation and technical automation. It ingests video walkthroughs—specifically Loom screen recordings—and systematically decomposes them into structured, executable logic. By leveraging a multi-stage pipeline involving yt-dlp for video acquisition, ffmpeg for intelligent frame capture, Whisper for transcription, and vision-capable LLMs for analysis, the skill identifies granular steps, active tools, and critical decision points within a process. It ultimately produces an executable Lobster workflow file, transforming passive observation into actionable automation code.

Installation

To install this skill within your OpenClaw environment, ensure you have the necessary system dependencies: yt-dlp, ffmpeg, and a local or API-based Whisper instance. Run the following command in your terminal:

clawhub install openclaw/skills/skills/g9pedro/loom-workflow

Ensure your configuration allows for local file-system access, as the skill generates a significant number of temporary frame files and output artifacts in your workspace.

Use Cases

• Automating Routine SaaS Tasks: Convert informal screen-share training videos into robust automation scripts.
• Process Auditing: Identify ambiguities, "tribal knowledge" gaps, and dangerous manual intervention points in existing company workflows.
• Rapid Prototyping: Accelerate the creation of complex workflows by simply recording them on-screen rather than hand-coding JSON schemas.
• Knowledge Transfer: Maintain up-to-date documentation that is inherently linked to executable code, ensuring that when the process changes, the automation can be quickly re-generated.

Example Prompts

1. "Analyze this Loom video link: https://loom.com/share/abc123 and extract the steps into a draft Lobster workflow, flagging any points where human approval is required."
2. "I have a screen recording of our Q3 report generation process. Use the loom-workflow skill to transcribe, analyze frames, and help me identify where the automation currently fails due to tool dependencies."
3. "Run the full extraction pipeline for this video. Focus specifically on identifying the decision-making logic used when determining if a customer support ticket needs escalation."

Tips & Limitations

• Frame Quality Matters: Ensure the Loom recording is high-resolution with a clear cursor; fuzzy text or rapid, erratic mouse movements can decrease the accuracy of the vision model's analysis.
• Ambiguity Detection: The tool is best at flagging what it doesn't know. Pay close attention to the generated workflow-summary.md regarding 'implicit knowledge'—these are usually the areas requiring human refinement before a workflow is production-ready.
• Compute Resources: This skill is resource-intensive due to the vision model analysis and video processing. Run it on machines with adequate RAM and CPU if processing long videos (over 10 minutes).