oh-my-openagent: AI coding agent for OpenCode & Codex
/oh-my-openagent
oh-my-openagent is a plugin suite that turns OpenCode or Codex CLI into a multi-agent coding setup, built by developer code-yeongyu. Typing one command, ultrawork, hands a task to a full agent lineup instead of one model answering alone. Reach for the Ultimate edition if you already run OpenCode and want that automation; skip it if you just want a plain single-agent CLI with no extra config to learn.
The Challenge of AI Agent Orchestration
Running Claude Code, Codex CLI, and assorted open models side by side means separate configs, separate hook systems, and no shared way to hand a task to more than one agent at once. Switching providers mid-project usually means redoing prompts and rules by hand. oh-my-openagent's README frames its purpose around that gap: install once, then let ultrawork coordinate whichever agents and models are configured, instead of wiring each host-and-provider combination yourself.
Understanding oh-my-openagent
oh-my-openagent is a TypeScript project that ships in three editions: Ultimate, a plugin for the OpenCode host with 11 agents and built-in MCP servers; Light, a set of components for Codex CLI installed through LazyCodex; and Senpi, a standalone beta CLI. Each edition adds agent orchestration, hooks, and rules injection on top of a coding-agent host you already run.
Key Capabilities and Editions
- ✓Ultimate edition installs 11 agents, 54+ lifecycle hooks, and 5 built-in MCP servers into OpenCode, per the README's installation table.
- ✓Discipline Agents in the Ultimate edition: Sisyphus orchestrates Hephaestus, Oracle, Librarian, and Explore as a parallel dev team.
- ✓Team Mode landed in v4.0 as an opt-in feature: a lead agent plus up to 8 parallel members with tmux visualization and dedicated team_* tools, powering preset modes such as hyperplan (which puts 5 agents into a critic role) and security-research (3 vulnerability hunters paired with 2 proof-of-concept engineers).
- ✓ultrawork (or ulw) is a single command, available in both Ultimate and Light editions, that activates every configured agent and keeps going until the task is finished.
- ✓The Hash-Anchored Edit Tool (Ultimate only) validates every change against a LINE#ID content hash before applying it, an idea the README credits to the oh-my-pi project.
- ✓LSP integration, AGENTS.md rules injection, and the Ulw Loop persistent-orchestration system ship in both the Ultimate and Light editions.
- ✓Built-in MCP servers for web search (Exa), official docs (Context7), and code search (Grep.app) run always-on in the Ultimate edition.
- ✓IntentGate (Ultimate only) reads actual user intent before acting; the Light edition falls back to recognizing just the literal ultrawork/ulw keyword.
Installation Methods
oh-my-openagent comes as three separate editions, and each one has its own install command. Ultimate (full OpenCode plugin): `bunx oh-my-openagent install`, which walks you through setup in a TUI. Light (Codex CLI components): `npx lazycodex-ai install`, or non-interactively `npx lazycodex-ai install --no-tui --codex-autonomous`; this writes into `~/.codex/plugins/cache/sisyphuslabs/omo/` and updates `~/.codex/config.toml`. Both platforms at once: `bunx oh-my-openagent install --platform=both`. Senpi (standalone beta): `npm i -g omo-ai@beta`, then run `omo`; a bare `npm i -g omo-ai` without the beta tag fails by design. The README warns against running `bunx omo` or `npx omo`: that command resolves to a completely different tool maintained by someone else on npm. On npm, the CLI binary keeps the older name `oh-my-opencode`, and it's published a second time under `oh-my-openagent` while the rename is in progress; the README recommends letting an LLM agent read the full installation guide and handle per-provider auth for you rather than doing it by hand.
Automated Coding with `ultrawork`
Day to day, the workflow is one word: type `ultrawork` (or the short form `ulw`) and every agent configured for your edition activates, working until the task is done rather than stopping after one pass. The README says this works well even with modest subscriptions — it names ChatGPT ($20), Kimi Code ($19), and the GLM Coding Plan ($10) as examples, and is explicit that these are personal recommendations, not sponsorships. Behind ultrawork, the Ulw Loop tracks multiple goals at once in `.omo/ulw-loop/` and checks its own results against an evidence audit instead of stopping after a single pass. Other documented modes include `search`, `analyze`, `team`, and `hyperplan`, plus slash commands and an `/init-deep` command that generates hierarchical AGENTS.md files across a project.
Developer Feedback and Benefits
- ✓Three editions instead of one bundle — Ultimate, Light, and Senpi — so the install matches whichever host (OpenCode, Codex CLI, or neither) you already run.
- ✓Telemetry is documented in specifics: a single daily event per machine, a SHA256-hashed installation ID rather than a raw hostname, no PostHog person profiles, and separate opt-out variables per product.
- ✓The README names concrete, inexpensive subscriptions it says work well with ultrawork instead of implying you need one expensive plan.
- ✓Team Mode's hyperplan and security-research presets give you a built multi-agent review setup rather than requiring you to assemble one from scratch.
- ✓LSP integration and rules injection (AGENTS.md / .omo/rules/**) are shared across both plugin editions, not exclusive to the pricier one.
Current Considerations
- △No license identifier is listed in the facts available for this review, so confirm the license on the repo page before using oh-my-openagent beyond personal experiments.
- △Package naming is mid-transition: npm still lists the CLI binary under the older name `oh-my-opencode`, with a parallel `oh-my-openagent` package covering the rename, and a same-named `omo` package on npm is a separate tool from someone else entirely — easy to install the wrong thing.
- △Team Mode, IntentGate, the Hash-Anchored Edit Tool, and most Discipline Agents are Ultimate-only; the Light edition drops agent orchestration and recognizes only the literal ultrawork/ulw keyword.
- △Anonymous telemetry ships on by default for both the main plugin and the Codex Light edition and has to be turned off explicitly, product by product.
- △The README's adoption evidence is a set of quoted social posts from named individuals rather than independent benchmarks or a published case study, so treat the enthusiasm as self-selected.
- △The project describes an in-progress 'Multi-Harness Agent OS Refactor' to support more hosts, meaning core architecture is still being restructured rather than settled.
Comparing with Other AI Coding Tools
Common Questions
oh-my-openagent ships as three editions: Ultimate (the full plugin for OpenCode, installed via `bunx oh-my-openagent install`), Light (portable components for Codex CLI, installed via `npx lazycodex-ai install`), and Senpi (a standalone beta CLI, `npm i -g omo-ai@beta`).
The installation guide for oh-my-openagent covers provider authentication for Anthropic, Gemini, GitHub Copilot, Z.ai, and OpenCode Zen.
Typing `ultrawork` (or `ulw`) activates every agent configured in oh-my-openagent, and it keeps working until the task is finished without further prompting.
oh-my-openagent lets you turn telemetry off: set `"telemetry": false` in the main plugin's config, or use `OMO_DISABLE_POSTHOG=1` / `OMO_SEND_ANONYMOUS_TELEMETRY=0`, with separate variables for the Codex Light edition.
LazyCodex is the install path for oh-my-openagent's Light edition inside Codex CLI: running `npx lazycodex-ai install` sets up the omo plugin under the sisyphuslabs marketplace in `~/.codex/`.
Team Mode is limited to oh-my-openagent's Ultimate edition, per the README's edition legend; it isn't included in Light or Senpi.
Best use cases
- •Running OpenCode and wanting a pre-built multi-agent team instead of hand-writing your own prompts and hooks — install the Ultimate edition.
- •Using Codex CLI and wanting the ultrawork command plus LSP integration without full agent orchestration — the Light edition via `npx lazycodex-ai install`.
- •Holding a few cheaper model subscriptions (ChatGPT, Kimi Code, GLM Coding Plan) and wanting one workflow that routes across them instead of standardizing on a single vendor.
- •Wanting an automated adversarial review pass on a change — Team Mode's hyperplan and security-research presets run multiple critic or hunter agents for you.
Who should try it — and who should skip
Try oh-my-openagent if you already run OpenCode or Codex CLI and want a maintained multi-agent layer — Team Mode, Discipline Agents, and ultrawork — instead of building your own hooks and prompts. Skip it if you want a single, simple agent with no extra install step, or if an undocumented license and an in-progress architecture refactor are dealbreakers for how you plan to use it.
Related repositories
Curious whether oh-my-openagent is right for you?
Let ChatGPT, Claude, or Perplexity look into it — click below and see what AI actually says about oh-my-openagent.