Video-to-Skill

Runs after youtube-learn completes. Reads the saved note. Makes one decision: is there a repeatable system here worth building into a skill?

Trigger

Automatically after youtube-learn saves a note to ~/.openclaw/workspace/knowledge/youtube/. Can also be triggered manually by Charlie dropping a note path or saying "turn this into a skill."

Workflow

Step 1 — Read the note

Read the most recently saved file in ~/.openclaw/workspace/knowledge/youtube/.

Step 2 — Evaluate: is this skill-worthy?

A video is skill-worthy if it contains at least one of:

• A repeatable workflow with clear steps
• A framework Charlie or Elon would apply more than once
• A system that maps to content creation, trading research, or AI tooling
• A mental model that should change how the agent responds to a recurring situation

A video is NOT skill-worthy if:

• It's a one-time news event or market update
• It's purely motivational with no operational content
• The insight is already covered by an existing skill
• The framework is too vague to turn into concrete agent behavior

If NOT skill-worthy → stay completely silent. Do not output anything.

Step 3 — Draft the skill

If skill-worthy, draft a full SKILL.md using this structure:

---
name: [kebab-case-name]
description: [One line — when to trigger this skill and what it does. Be specific.]
---

# [Skill Name]

## Source
Derived from: [video title] — [channel]
Saved note: [path to note]

## When to Use
[Specific trigger conditions — what Charlie says or does that activates this]

## Core Framework
[The actual system from the video, rewritten as agent instructions]
[Steps, rules, or decision tree — make it operational, not descriptive]

## Output Format
[What the agent should produce when this skill runs]

## Charlie-Specific Application
[How this maps to Charlie's world: trading, content creation, or AI tooling]
[Be concrete — not "apply this to your content" but exactly how]

## What to Avoid
[Common misapplications of this framework]
[What the video got wrong or oversimplified, if anything]

Step 4 — Surface for approval

Output exactly this and nothing else:

🧠 NEW SKILL DRAFT — [Skill Name]

Source: [video title] Would activate when: [trigger description]

[paste the full SKILL.md content]

Reply "install it" to add this to your skill library, or "skip" to discard.

Step 5 — Wait for Charlie's approval

Do not install, save, or register anything until Charlie explicitly says to install it. If approved, save to ~/.openclaw/workspace/skills/[skill-name]/SKILL.md and register it in openclaw.json under skills.entries.

Voice

• The draft should read like a real working skill, not a summary of the video.
• If the framework needs to be adapted to work as agent instructions, adapt it — don't copy-paste the video's language.
• Be skeptical. Most videos don't contain skill-worthy systems. Silence is the right output most of the time.
• When you do surface a skill, make it tight. One sharp skill beats three vague ones.