Skill Migrator

Skill Migrator

[English](#english) | [简体中文](#简体中文)

---

English

This repository documents a skill for OpenClaw to learn.

Its core purpose is not to be a generic "migrate any skill from any platform" toolkit. The primary goal is:

• Teach OpenClaw how to convert Claude Code skills into OpenClaw skills
• Teach OpenClaw how to convert Hermes skills into OpenClaw skills

CloudCode is still kept as extended support, but it is not the first priority of this skill.

What OpenClaw Should Do After Learning This Skill

When a user provides a Claude Code or Hermes skill and asks to migrate it into OpenClaw, OpenClaw should:

1. Detect whether the source skill is Claude Code or Hermes.
2. Analyze the source structure, tools, state, dependencies, and migration risks.
3. Produce an OpenClaw-compatible output directory.
4. Generate valid SKILL.md, scripts/, references/, and state.yaml when needed.
5. Validate the migrated result.
6. Clearly explain what was migrated automatically and what still needs manual review.

In other words, this skill should help OpenClaw produce a usable OpenClaw skill scaffold, not just explain migration ideas.

Expected Usage

After OpenClaw learns this skill, it should prefer using it for requests like:

• "Convert this Claude Code skill into an OpenClaw skill"
• "Migrate this Hermes skill"
• "Analyze whether this skill can be migrated to OpenClaw"
• "Analyze first, then do a phased migration into OpenClaw"

Recommended strategy:

• Simple case: analyze and convert directly
• Complex case: prefer phased migration
• High-risk case: preserve manual review points instead of pretending the migration is fully automatic

Improvement Note

v4.5 - Keep boundaries honest, add back the decision tree, checkpoints, and minimal examples.

What Improved in v4.5

• Restored a short decision tree in SKILL.md
• Restored explicit review checkpoints
• Restored minimal scenario examples
• Moved detai