Oh My Claude Code (OMC): Agent Swarm Orchestration
Oh My Claude Code (OMC) is a plugin for the Claude Code terminal that turns the standard request-response loop into a multi-agent orchestration system. Named in the spirit of oh-my-zsh, it aims for a low learning curve: a library of specialized agents and skills that you drive in plain language. The point is to stop hand-coding each step and instead orchestrate a swarm of agents that plan, execute, test, and self-correct.
If the creator’s own workflow is the manual version of fleet-commander development, OMC is the packaged one.
The five execution modes
OMC’s structure is a set of modes, each tuned for a different scenario — from fully autonomous builds to cost-conscious runs.
1. Autopilot — full autonomous execution. You describe a high-level goal; OMC handles the whole lifecycle: planning, parallel execution with specialized agents, continuous testing, and self-correction until the task is done.
2. Ultrapilot — parallel autopilot. Autopilot with concurrency. It decomposes a large task into parallelizable sub-tasks and runs several workers at once — suited to big refactors or apps with independent components.
3. Swarm — coordinated agent teams. Spawns a set number of agents that pull from a shared task pool. Each claims an atomic task, completes it, and marks it done, so no two agents duplicate work — the model of a team burning down a sprint backlog.
4. Pipeline — sequential chaining. For work with a required order, Pipeline chains agents in sequence, feeding each stage’s output into the next. It ships presets for common flows like review, implement, and debug.
5. Ecomode — cost-optimized parallelism. Parallel execution tuned for cost. It routes each task to the cheapest model that can handle its complexity:
- Simple tasks → Haiku
- Standard work → Sonnet
- Complex reasoning → Opus
Natural-language control and routing
OMC’s main convenience is interpreting intent instead of demanding memorized commands. Phrasing shapes behavior — for example, adding “don’t stop until done” to a prompt engages a persistence mode.
It also assigns work to the right specialist automatically: “fix the UI colors” goes to a designer agent, “debug this crash” to an architect agent. Routing the task to the right agent also routes it to the right model, so reasoning depth (and cost) match the job. This is the same model-routing logic covered in the model-selection guidance in AI Models: A Comparative Overview — applied automatically per task.
MCP connectivity
OMC inherits Claude Code’s support for the Model Context Protocol (MCP), so its orchestration modes can operate against external tools and data sources — databases, repositories, and cloud services — through configured MCP connectors. For how to enable and scope MCP servers, see the MCP section of the Claude Code Operator’s Guide.
Assessment
OMC’s value is making multi-agent orchestration accessible: it formalizes the fleet-commander pattern so one developer can direct the output of a small team. The routing across Haiku, Sonnet, and Opus is what keeps that affordable on large projects, and the natural-language interface keeps it usable day to day. For anyone using Claude Code beyond simple one-shot prompts, it’s a real force multiplier.
