agent-skills-setup

Use when a user names two supported IDEs or agent products to plan, migrate, or inspect specific skills, instructions, and MCP. The skill inventories local paths and runs bundled Bash/Python; an approved apply or rollback may write targets, create backups/manifests, verify results, and scan or redact secrets. Network access is forbidden.

Install

openclaw skills install @luckycat133/agent-skills-setup

AI IDE Context Migration

Capabilities and authorization

  • detect, doctor, inventory, plan, snapshot, and bundle-verify read only named products and workspace; network access is forbidden.
  • A generic migration request authorizes planning only; separate explicit user approval (--yes) or explicit action verbs (apply, restore, 迁到) under --apply-safe authorize write.
  • Save the plan, review its diff/rebuild manifest, and apply that exact file.
  • The explicit legacy subcommand keeps lookup compatibility; legacy writes are disabled.

Route

  1. Resolve both product profiles through ide-registry.md / registry-v2.json.
  2. Read only references/ides/.md and references/ides/.md.
  3. Load reference by need:

Execution & Scope

  • High-level: bash scripts/smart-ide-migration.sh migrate --source <src> --target <dst> --workspace . --objects all-portable --yes
  • Step-by-step: plan --output <plan.json> -> apply <plan.json> --manifest <manifest.json> --yes -> verify --manifest <manifest.json> -> rollback --manifest <manifest.json> --yes.
  • Device handoff (ACB): snapshot --output-bundle <b.acb> -> bundle-verify --bundle <b.acb> -> restore --bundle <b.acb> --yes (offline self-contained archive).
  • Diagnostics: detect / doctor inspect local probes and installation states offline.
  • Surface scope: skills, instructions, MCP, prompts, commands, workflows, agents/droids, and hooks (executable agents/hooks default to draft-disabled).
  • Plugins & extensions: opaque binaries/plugins are non-executable and marked manual-rebuild.
  • Claude Desktop app MCP in Settings → Extensions and Settings → Connectors is UI-managed; do not infer or rewrite it from legacy JSON.
  • Never move secrets, OAuth/session state, runtime metadata, approval grants, chat history, or generated memory.