oh-my-claude
By 2lab-ai
Save and load cross-session work contexts including files, plans, and todos. Run Ralph deep-work loops that automate multi-AI agent workflows: delegate codebase exploration, docs research, architecture advice, PR summaries, and ruthless code reviews with quality gates blocking poor code.
Component Overview
Install
npx claudepluginhub 2lab-ai/oh-my-claude --plugin oh-my-claudeComponent Details
Commands (16)
Cancel active Ralph Wiggum loop
Verify completion status of archived saves in .claude/omc/tasks/archives/
Start Deep work loop with gated AI review + multi-agent delegation
Search THIS codebase using Explore agent (Gemini). Find implementations, patterns, code flow.
Search external docs, best practices, library APIs using Librarian agent (Opus 4.5)
List all saved work contexts
Load saved work context
Ask Oracle for architecture advice, design decisions, or failure analysis
Multi-agent work coordinator. Delegates to Oracle/Explore/Librarian. Runs in current context (can use AskUserQuestion).
Generate multi-perspective PR Executive Summary for stakeholders
Save current work context, plan, and todos.
Install and verify required dependencies (codex, gemini-cli)
Use when user provides a SPEC file, feature description, or URL and wants to clarify requirements through interviewing. Triggers on "interview me about", "spec interview", "refine this spec", "create coding plan", "feature planning", or when user shares a specification file and wants detailed discussion.
Show current Ralph loop status (iteration, time elapsed, completion promise)
Advanced summarization
ultra workflow with multi-agent delegations
Agents (5)
Internal codebase exploration agent. Use for finding implementations, patterns, code flow in THIS codebase.
External documentation and open-source codebase understanding. Use for official docs, best practices, library APIs, GitHub source analysis. MUST provide GitHub permalinks as evidence. Background execution.
Strategic technical advisor with deep reasoning. Use for architecture decisions, after 3 failed fix attempts, unfamiliar patterns, security/performance concerns. Read-only consultant - BLOCKING execution.
Multi-agent work coordinator. Delegates to Oracle/Explore/Librarian. Use as subagent for autonomous task execution. NO user interaction.
Uncompromising code reviewer inspired by Linus Torvalds. Applies Occam's Razor and First Principles thinking. Use as 3rd reviewer in /deepwork for ruthless quality gate. BLOCKING execution.
Hooks (1)
MCP Servers (3)
README
oh-my-claude
____ _ _ __ ____ __ _____ _ _
/ __ \| | | | | \/ \ \ / / / ____| | | |
| | | | |__| |_____| \ / |\ \_/ /_____| | | | __ _ _ _ __| | ___
| | | | __ |_____| |\/| | \ /______| | | |/ _` | | | |/ _` |/ _ \
| |__| | | | | | | | | | | | |____| | (_| | |_| | (_| | __/
\____/|_| |_| |_| |_| |_| \_____|_|\__,_|\__,_|\__,_|\___|
[!CAUTION] EXPERIMENTAL PROJECT - This plugin runs autonomous AI loops that can consume significant tokens. Mind your token usage!
/ultraworkand/deepworkwill iterate until completion. Use--max-iterationsto set limits. You have been warned.
Claude Code plugin for AI-powered iterative development loops.
Inspired by oh-my-opencode and Ralph Wiggum.
Installation
/plugin marketplace add 2lab-ai/oh-my-claude
/plugin install oh-my-claude@oh-my-claude
/plugin install powertoy@oh-my-claude
Required: Run
/oh-my-claude:setupafter installation.
Authentication
oh-my-claude uses your Gemini CLI and Codex CLI subscriptions (not API keys).
# 1. Run setup to check dependencies
/oh-my-claude:setup
# 2. Login to each service
gemini auth login # Opens browser for Google auth
codex auth # Opens browser for OpenAI auth
# 3. Verify everything works
/oh-my-claude:setup
Once authenticated, agents use your existing subscriptions:
- Explore agent → Your Gemini subscription
- Oracle agent → Your OpenAI/Codex subscription
- Librarian agent → Your Claude subscription (already authenticated via Claude Code)
Main Features
/ultrawork - Multi-Agent Work Loop
Ralph loop with 3 specialized AI agents for autonomous development.
[!IMPORTANT] Requires YOLO mode:
claude --dangerously-skip-permissions
/ultrawork "Build REST API for users"
Agents:
- Oracle (Codex GPT-5.2): Architecture decisions, failure analysis (blocking)
- Explore (Gemini): Internal codebase search (parallel)
- Librarian (Opus 4.5): External docs, GitHub source analysis (parallel)
Loop exits when task is genuinely complete.
/deepwork - Reviewed Work Loop
Same as ultrawork, but requires 9.5+ score from all three reviewers (GPT-5.2, Gemini-3, Opus-4.5) before completion.
/deepwork "Critical security fix" --max-iterations 50
/save & /load - Cross-Tool Context Management
Save work context in one tool and resume in another.
# In Claude Code: save your work
/save
# In Gemini CLI or Codex: resume the work
/load
Cross-tool workflow example:
- Start work in Claude Code,
/savewhen done - Open Gemini CLI or Codex,
/loadto continue - Work across tools seamlessly - they share
./docs/tasks/save/
To enable cross-tool commands:
./install-cross-session-commands.sh
This installs /save, /load, /list-saves, /check to:
~/.gemini/commands/(Gemini CLI)~/.codex/prompts/(Codex)
Commands Reference
oh-my-claude
| Command | Description |
|---|---|
/ultrawork | Multi-agent autonomous work loop |
/deepwork | Work loop with triple AI review gate (GPT-5.2 + Gemini-3 + Opus-4.5, all ≥9.5) |
/save | Save work context |
/load <id> | Load saved context |
/list-saves | List all saved contexts |
/check [all|id] | Verify archived saves completion |
/cancel-work | Cancel active loop |
/setup | Verify dependencies |
powertoy
| Hook | Description |
|---|---|
| auto-title.sh | Auto-generate session titles (Claude Haiku) |
| play-sound.sh | Play sound on session end (macOS) |
MCP Servers
Bundled MCP servers:
| Server | Package |
|---|---|
| gemini | @2lab.ai/gemini-mcp-server |
| claude | @2lab.ai/claude-mcp-server |
| codex | codex mcp-server |
Credits
- oh-my-opencode: code-yeongyu
- Ralpy Wiggum plugin: https://github.com/anthropics/claude-code/tree/main/plugins/ralph-wiggum
- Ralph Wiggum technique: Geoffrey Huntley
FAQ
Q: Will this burn my tokens?
A: Yes. That's the feature, not a bug.
/\_/\
( o.o ) "I'm helping!"
> ^ < - Your AI, iteration 47
Changelog
| Version | Changes |
|---|---|
| 1.0.22 | Context window optimization - total plugin under 4k tokens |
TODO
- LSP support
License
MIT
Similar Plugins
ui-design
Comprehensive UI/UX design plugin for mobile (iOS, Android, React Native) and web applications with design systems, accessibility, and modern patterns
Uses power tools
Uses Bash, Write, or Edit tools