Agent Skills Setup

AI IDE Context Migration across Cursor, Claude Code, Codex, Cline, Windsurf, Copilot, Gemini CLI, and dozens of supported AI coding tools.

Languages: English · 中文 · 日本語 · Español

Version: 0.10.2. See the release changes. GitHub installation pins v0.10.2.

Offline, previewed, and rollback-safe migration of AI coding context (Skills, Rules, Instructions, and MCP) across Cursor, Claude Code, Codex, Cline, Windsurf, Copilot, Gemini CLI, and dozens of supported AI coding tools.

---

⚡ Quick Install

Install agent-skills-setup via ClawHub / OpenClaw:

openclaw skills install @luckycat133/agent-skills-setup

Or install directly from GitHub:

git clone --depth 1 --branch v0.10.2 https://github.com/Luckycat133/skills-repo.git

openclaw skills install \
  ./skills-repo/skills/agent-skills-setup \
  --as agent-skills-setup

---

💬 Say this to your agent

"Migrate Skills, rules, and MCP from Cursor in this project to Claude Code."

"I am switching computers. Back up all portable AI coding setup from my installed IDEs."

"Restore this ACB bundle to the installed IDEs on my new computer."

---

🛡️ Migration Boundary

Tier	Items	Behavior
Automatic	Skills, Instructions/Rules (CLAUDE.md, .cursorrules), Local stdio MCP	Verified copy for Skills, semantic conversion for instructions, and supported local stdio MCP subset conversion.
Explicit Opt-in	Plugin package copy (--include-plugins), Session handoff summary (--include-session)	Preserved with structured whitelisting and consent.
Manual Checklist	Remote MCP (HTTP/SSE), Cloud/UI configurations, Prompts, Commands, Agents, Hooks	Emits concrete reconstruction checklists; never writes unreviewed executable scripts.
Never Moved	API keys, OAuth tokens, credentials, trust state, raw chat history, generated memory	Strict secret scanning, subobject isolation, and fail-closed exclusions. Literal credentials are rejected or redacted before migration.

---

🚀 Capabilities

• migrate: Inspect the selected products, save a plan, apply it, and verify the results.
• snapshot: Capture an atomic, portable Agent Context Bundle (ACB) with 1:1 manifest file binding.
• restore: Dual-side restore planning from ACB bundles onto newly installed target tools with TOCTOU state guards.
• bundle-keygen, bundle-sign & bundle-verify: Generate Ed25519 keys, sign ACB bundles, and verify integrity with a trusted public key.
• doctor: Inspect bundle requirements and missing dependencies offline.

ACB captures ordinary file permissions, including executable bits, and still reads older bundles without permission metadata. Saved restore plans bind to their bundle identity; MCP previews list added, removed, and changed servers without credential values. Instruction/MCP conversions with reported losses are deferred until explicitly accepted. doctor reports package, extension, platform, and manual-install requirements alongside executable checks.

Use detect and inventory for discovery, plan for a saved preview, apply for the reviewed plan, and verify / rollback for evidence and recovery. See the command guide for complete examples and the compatibility matrix for each profile's tested scope. Remote MCP and cloud settings produce manual rebuild actions. Plugin and handoff transfer each require their explicit opt-in.

---

🌟 Support the Project

If Agent Skills Setup saved you setup or computer-switching time, please ⭐ Star this repository on GitHub to help other developers discover it!

You can also sponsor development through GitHub Sponsors or Afdian / Ko-fi.

---

Layout

skills-repo/
├── docs/                         # Maintainer guidance & release checklists
├── scripts/                      # Repository tooling
└── skills/agent-skills-setup/    # Canonical publishable skill
    ├── SKILL.md                  # Skill descriptor & entry point
    ├── references/               # Profile registry v2 & format adapters
    └── scripts/                  # Context migrator core & ACB engine

Develop & Validate

1. Edit skills/agent-skills-setup/, never the generated root repository pointer.
2. Regenerate the root pointer after canonical-skill edits: bash scripts/sync-root-mirror.sh.
3. Run bash validate-all.sh.
4. Merge only after all validation passes.

For isolated scale testing, run python3 scripts/benchmark-scale.py --sizes 10,100,1000. It checks migration, saved-plan restore, exact contents, and rollback; timings are observations, not CI thresholds. Add --report /path/to/new-report.json and --keep-workspace to retain evidence. No real agent configuration is used.

Project

• Changelog
• Release checklist
• Contributing
• Security
• Code of conduct
• License