UmaDev CLI
A Rust CLI that runs development as a team of roles on top of an installed Claude Code, Codex or OpenCode: plan, checks and a final report
- 264 ★ on the repository
- Last commit
- MIT
Broad access
Why this level
- Controls Claude Code, Codex and other agents that write and run code in your project, including in `auto` mode
- Can install a pre-write hook in Claude Code and works with sources, call logs and `.umadev/` files
- The install script downloads a binary from the network, and the old unscoped npm package had a malicious version
Install
- Works in
In your terminal, with SkillFoxx CLI
npx skillfoxx add cli/umadevDetects the agents on your machine, shows the access level and pins the version.
Other ways to install
Access and checks
How we reviewBroad accessThe tool can change data, run commands or handle secrets.
- Controls Claude Code, Codex and other agents that write and run code in your project, including in `auto` mode
- Can install a pre-write hook in Claude Code and works with sources, call logs and `.umadev/` files
- The install script downloads a binary from the network, and the old unscoped npm package had a malicious version
This is third-party code. Review the repository files before installing.
What it does
UmaDev is a terminal coordinator for software development. It keeps no model of its own and asks for no keys: the work is done by an agent you have already installed and signed in to, chosen from five: Claude Code, Codex, OpenCode, Grok Build or Kimi Code. UmaDev starts with the `umadev` command as a terminal chat, and for scripts and CI there are subcommands like `umadev run "task" --backend claude-code`. The model of the chosen agent first classifies the request: chat, explanation, quick edit, debugging or a full build. For a large build the coordinator writes a dependency plan to `.umadev/plan.json`, shows it as a live checklist and steps through it. Writing roles, frontend and backend, work one at a time. Reviewing roles, product, architect, designer, QA, security and DevOps, get independent sessions and return verdicts. Quality is checked deterministically, without the model grading itself. Build, linter, types, tests, requirement coverage and frontend paths matching the API contract come first. A set of 113 content checks runs in the repository, for example for leaked secrets and hardcoded colors. They fire as a pre-write hook in Claude Code, as pre-commit and as part of the quality gate. Files remain at the end: a research note, PRD, architecture, UI/UX, a scorecard and an evidence archive. The `plan`, `guarded` and `auto` trust modes set the level of autonomy, and irreversible actions are confirmed on any tier.
Who it is for. For developers who already use Claude Code, Codex or OpenCode and want a planned project build with checks and a report.
Good fit when
- You need to build an app from scratch from a description and want a PRD, architecture, plan and a frontend and backend contract check
- You want the agent to keep going until the goal is met: the `/goal` command holds the goal to completion
- A team needs a decision trail: audit of tool calls and checks lands in `.umadev/audit/`
- You need to run in CI: `umadev run` works without the interface and `umadev ci` plugs in as a pre-commit
- An old repository must join the process: the `/adopt` command detects the stack and indexes the code
Not a fit when
- You need one small edit: it is faster and cheaper to work in Claude Code or Codex directly
- None of the five agents is installed and signed in: UmaDev does not log in to accounts and supplies no model
- You need to install without Node but `curl | bash` is forbidden: then you must build from Rust sources
- You need independent people to own decisions: the roles are one model in different sessions, not separate specialists
Example request
Start umadev and build a course-booking mini app: course list, time choice, booking and cancelling, an admin area for courses and bookings.Limitations
- Real work needs an installed and signed-in agent from the five: Claude Code, Codex, OpenCode, Grok Build or Kimi Code. You sign in inside the agent itself, and the model bill goes through its subscription or key.
- The roles are different sessions of one model and cannot replace independent specialists. Reviewer opinions are advisory, and the course of work is decided by the deterministic check.
- Local vectors need a build with `--features vector-local` and the `multilingual-e5-small` model of about 224 MB, which the npm package downloads on demand. Without it lexical BM25 search works.
- The old unscoped npm package `umadev` is retired and its version `1.0.74` was a malicious publish. Install only `@umatech/umadev`.
- Different agents expose different events, so some features (questions, approvals, goals) depend on what the chosen agent provides.
How to disable. Remove the package with `npm uninstall -g @umatech/umadev`, or if you used the script delete the binary from `~/.local/bin` (on Windows from `%LOCALAPPDATA%\Programs\umadev`). Project data lives in `.umadev/`, user settings in `~/.umadev/config.toml`. If you connected Claude Code hooks, remove them in the agent settings.
FAQ
How do I install UmaDev and which agents does it work with?
Run `npm i -g @umatech/umadev` or the script `curl -fsSL https://umadev.goder.ai/install.sh | bash`, then start `umadev`. It drives Claude Code, Codex, OpenCode, Grok Build and Kimi Code, which must be installed and signed in beforehand.
Does UmaDev need its own API key?
No. The program has no model of its own, and work goes through the chosen agent and its subscription or key. If a built app needs a key, the model and key of its runtime are set separately through environment variables.
How does UmaDev differ from running Claude Code directly?
UmaDev adds a coordinator above the agent: a plan, roles, deterministic checks, an API contract and final documents. The code is still written by the chosen agent.
What do the plan, guarded and auto modes mean?
They are a scale of autonomy set by the `/mode` command. Irreversible actions ask for confirmation on any tier, including obfuscated commands UmaDev cannot parse.
Why must I not install the unscoped umadev package?
The unscoped package is retired and its version `1.0.74` was malicious. Use `@umatech/umadev` instead and remove the old one with `npm uninstall -g umadev`.
Discussion
No questions or reviews yet.
Related
CLI toolsAn open-source coding agent in Rust with a persistent Python workspace, child agents and any model provider
Workflows for Claude Code: /do, /omo, /bmad-pilot and more commands, with Codex, Gemini or OpenCode writing the code through codeagent-wrapper
A team of AI agents from one request: it plans the task, writes code in parallel branches, tests, reviews and opens a pull request
A Rust CLI that builds a knowledge graph of code and decisions and feeds the agent the right slice through Claude Code hooks
A control plane that runs Claude Code, Codex and Gemini CLI agents in parallel with a shared task board and schedules
A Go coding agent for open models: it turns a chat into tasks on separate branches, checks and merges the result and runs all projects in one window
Comparison with similar tools
Entries that share the main search query, and how they differ.
| Entry | Official | Repository stars | Main difference |
|---|---|---|---|
| UmaDev (this entry) | no | 264 | |
| oh-my-claudecode | no | 40,377 | an add-on and agent set for Claude Code, while here a standalone CLI above five different agents with spec-based checks |
| oh-my-openagent | no | 70,474 | a plugin for OpenCode, while UmaDev works above several agents at once |
README in short
The page starts with install through npm or a script and a table of agents with sign-in commands. It continues with a quick start, a detailed example of building an app, the project's origin from super-dev, a feature list and a flow diagram.