Skip to main content
Use Pluxx in Claude Code when you want a native Claude plugin bundle plus optional Claude-driven refine and review passes.

​
What Claude Code Consumes

Pluxx builds the Claude target into dist/claude-code/. The Claude bundle includes:
  • .claude-plugin/plugin.json
  • skills/
  • commands/
  • agents/
  • hooks/hooks.json
  • .mcp.json
  • CLAUDE.md
Claude Code reads those files as a local plugin bundle. MCP wiring lives in .mcp.json, hook wiring lives in hooks/hooks.json, and shared instructions live in CLAUDE.md.

​
Build And Install

npx @orchid-labs/pluxx build --target claude-code --install
If you already built the bundle, install it directly: 
npx @orchid-labs/pluxx install --target claude-code
Pluxx installs Claude plugins through a local Claude marketplace:
  • local marketplace root: ~/.claude/plugins/data/pluxx-local-<plugin>/
  • install command Pluxx runs: claude plugin install <plugin>@pluxx-local-<plugin> --scope user
If the plugin defines local hook commands, add --trust or confirm the trust prompt during install.

​
Agent-Driven Refinement

Build and install are deterministic. Claude only does model work when you explicitly run an agent pass: 
npx @orchid-labs/pluxx agent prepare
npx @orchid-labs/pluxx agent run review --runner claude --no-verify
Use Claude for:
  • taxonomy refinement
  • instruction rewrites
  • scaffold review passes

​
Claude-Specific Caveats

  • If Claude Code is already open, run /reload-plugins after install so the session picks up the new plugin immediately.
  • When a skill name collides with a command name, Pluxx keeps the command visible and rewrites the colliding skill as a hidden internal skill with user-invocable: false.
  • If you configure Claude platform-managed MCP auth, the generated .mcp.json keeps the server definition but does not inject local auth headers at install time.