Oh My ClaudeAgent

oh-my-claudeagent

Multi-agent system for Claude Code. Specialist agents for planning, execution, review, debugging, research — with persistence, parallel execution, and natural language activation.

Installation

claude plugin marketplace add UtsavBalar1231/oh-my-claudeagent
claude plugin install oh-my-claudeagent@omca

Or from inside a Claude Code session:

/plugin marketplace add UtsavBalar1231/oh-my-claudeagent
/plugin install oh-my-claudeagent@omca

Team Setup

Add to your project's .claude/settings.json so team members get the plugin automatically:

{
  "extraKnownMarketplaces": {
    "omca": {
      "source": {
        "source": "github",
        "repo": "UtsavBalar1231/oh-my-claudeagent"
      }
    }
  },
  "enabledPlugins": {
    "oh-my-claudeagent@omca": true
  }
}

For GitHub Enterprise Server, use full git URLs in the source:

{
  "extraKnownMarketplaces": {
    "omca": {
      "source": {
        "source": "git",
        "url": "git@github.example.com:org/oh-my-claudeagent.git"
      }
    }
  }
}

Update

/plugin marketplace update omca

Local cache rebuild

Rebuild the locally-installed plugin cache from your dev tree:

bash scripts/package-plugin.sh ~/.claude/plugins/cache/omca/oh-my-claudeagent/$(jq -r .version .claude-plugin/plugin.json)/

Use --dry-run first to preview the file list. The script excludes dev artifacts (.omca/, .mypy_cache/, UPGRADE.md, tests/, etc.) that should not ship.

Uninstall

/plugin uninstall oh-my-claudeagent@omca

Quick Start

Run /oh-my-claudeagent:omca-setup to configure and verify dependencies. Then:

• "create plan for [your task]" — planning pipeline
• "ralph don't stop" — persistence mode (works until done)
• "ultrawork" — parallel agents (up to 5)
• /oh-my-claudeagent:start-work — execute a ready plan

What You Get

Specialist agents, skills via slash commands or keyword triggers, bundled MCP servers (omca: structural search + state, grep.app: public code search, context7: library docs), hooks for persistence, context injection, and auto-approval.

Heads-up — worktree.baseRef and unpushed commits

When you run anything in an isolated git worktree — /oh-my-claudeagent:start-work --worktree, or any agent you've given isolation: worktree — each invocation spawns in a fresh worktree. Since Claude Code v2.1.133, the default base ref for new worktrees is origin/<default-branch> (i.e. "fresh") — unpushed local commits are not visible inside that worktree. (OMCA's own explore and librarian agents run in your main checkout, not a worktree, so they always see uncommitted work.)

If you have WIP commits that haven't been pushed yet and you want worktree-isolated work to see them, set this in your user ~/.claude/settings.json:

"worktree": { "baseRef": "head" }

As of v2.1.154, "head" correctly resolves to the current worktree HEAD (the branch you have checked out), not the main checkout HEAD. Earlier versions had a bug where it resolved to the wrong HEAD in some setups.

Otherwise, push to your remote before delegating exploration of recently-changed code.

Requirements

• Claude Code CLI
• jq
• uv
• python3 3.10+
• ast-grep CLI (ast-grep or sg)

Documentation

• OMCA.md — Complete guide: agents, skills, workflows, MCP tools, runtime state, troubleshooting
• CLAUDE.md — Contributor internals: hook map, cross-file patterns, adding components

Acknowledgments

Based on oh-my-openagent by @code-yeongyu.

License

MIT