code-yeongyu/oh-my-openagent

oh‑my‑openagent (OmO) – an AI‑powered “team‑mode” development assistant

What it is – A TypeScript‑based plug‑in (and a standalone beta) that lets you run a fleet of LLM agents as a coordinated development team. It works as a plugin for the OpenCode editor, as a set of components for the Codex CLI, or as a native command‑line tool (omo-ai@beta). The core provides a single command (ultrawork / ulw) that launches a pre‑configured set of 11 agents, each specialised for tasks such as code generation, linting, searching, LSP integration, and “discipline” (goal‑oriented) execution.

---

Key concepts & features

Feature	Edition(s)	What it does
Team Mode	Ultimate	Up to 8 parallel agents with a lead “architect” agent; visualised in a tmux‑style pane.
ultrawork / ulw	Both	One‑word entry point that activates the full agent stack and runs until the declared goal is satisfied.
Discipline Agents	Ultimate	Built‑in workflow that orchestrates specialist agents (Librarian, Explorer, etc.) to keep work moving forward.
IntentGate	Ultimate (Light only sees ultrawork)	Keyword‑based trigger detection for commands like ultrawork, team, hyperplan.
Hash‑anchored edit tool	Ultimate (opt‑in)	Allows line‑level edits identified by LINE#ID tags, eliminating stale‑line errors.
LSP integration	Both	Provides diagnostics, navigation, symbol lookup, and workspace rename for the agents.
AST‑Grep	Both	Language‑aware code search/rewrite across ~25 languages via the sg tool.
Built‑in MCPs (websearch, context7, grep_app, LSP)	Ultimate	Ready‑to‑use micro‑service connectors for internet, documentation, GitHub, and editor data.
Goal / /goal	Ultimate	Persistent thread‑level objective; agents keep working on it until the flag is cleared.
Todo Enforcer	Ultimate	Detects idle agents and forces them back onto the task.
Comment Checker	Both	Ensures comments stay clean of AI‑generated artefacts.
Telemetry	Both	Anonymous daily‑active‑user ping (opt‑out via config or env vars).

---

Editions & how to get them

Edition	Host	Install command	What lands on disk
Ultimate (OpenCode)	OpenCode editor	bunx oh-my-openagent install (TUI)	opencode.json entry, ~/.omo/omo.jsonc config, provider auth prompts, 11 agent definitions.
Light (Codex CLI)	Codex command‑line tool	npx lazycodex-ai install (or npx lazycodex-ai install --no-tui --codex-autonomous)	~/.codex/plugins/cache/sisyphuslabs/omo/, ~/.codex/config.toml blocks, binaries in ~/.local/bin.
Senpi (beta)	Stand‑alone	npm i -g omo-ai@beta then run omo	A global omo binary that bundles the OMO extension; no host editor required.

The @beta tag is mandatory – the plain npm i -g omo-ai package is intentionally blocked.

---

Typical workflow (Ultimate edition)

1. Run installer – bunx oh-my-openagent install. The wizard asks for your model subscriptions (Anthropic, Gemini, Kimi, etc.) and creates the agent‑to‑model matrix.
2. Set a goal – In the OpenCode console type /goal "refactor the authentication module".
3. Launch – Type ultrawork (or the short alias ulw). The lead architect agent breaks the goal into subtasks, spawns specialist agents, and streams results back into the editor.
4. Monitor – If you enabled Team Mode, a tmux‑style pane shows each agent’s live output.
5. Iterate – Use /goal again or let the Todo Enforcer keep the agents busy until the job is done.

---

Use cases

• Full‑stack code generation – Turn a high‑level description into a working codebase (e.g., “convert a 45k‑line Tauri app to a SaaS web app”).
• Automated linting & fixing – Run the built‑in eslint‑style agent to clear thousands of warnings in minutes.
• Search‑and‑replace across a repo – Use the AST‑Grep skill to rewrite patterns safely in many languages.
• Documentation & web‑search augmentation – The websearch and context7 MCPs fetch up‑to‑date docs and feed them to the coding agents.
• Team‑style brainstorming – With Team Mode, multiple agents critique each other (e.g., “hyperplan” with hostile critics) to improve solution quality.

---

Documentation & community

• Docs site – https://omo.dev/docs
• Discord – https://discord.gg/PUwSMR9XNk
• Twitter/X – https://x.com/justsisyphus
• Roadmap – https://github.com/code-yeongyu/oh-my-openagent/blob/dev/ROADMAP.md
• License – SUL‑1.0 (see LICENSE.md).

---

Quick‑start cheat sheet

# Ultimate (OpenCode)
bunx oh-my-openagent install   # runs interactive TUI
# Light (Codex CLI)
npx lazycodex-ai install      # installs plugin into ~/.codex
# Stand‑alone beta
npm i -g omo-ai@beta && omo   # gives you the `omo` command

# After install (any edition)
ultrawork   # or `ulw`
# Set a persistent goal (OpenCode only)
/goal "add unit tests for the payment service"

---

Bottom line

oh‑my‑openagent is a real, actively maintained project that bundles a collection of LLM‑driven agents into a plug‑in/CLI ecosystem. It abstracts away the messy configuration of multiple model providers and gives developers a single entry point (ultrawork) to let a coordinated AI “team” write, refactor, and debug code.