Ordewell 发布：把一个目标变成有序的编码智能体任务计划

Task orchestration for coding agents.
One goal becomes an ordered plan of tasks, each on its own runner, model and branch.

Website · Documentation · News

Ordewell turns a goal into a plan you can read and change before anything runs. A planner researches your repository and asks about whatever you left vague. It then hands back a dependency graph of tasks, and each task names the coding agent, model and thinking effort it will use. Independent tasks run in parallel, each in its own git worktree. A task only counts as done when its completion marker appears in the agent's output, and nothing reaches your branch until you have reviewed the result.

Claude Code, Codex and OpenCode are supported out of the box, and can be mixed freely within one plan.

Features

• An editable plan. Change any task's prompt, runner, model, effort or mode, add or remove tasks, and rewire dependencies without another round trip to the model.
• The right model for each task. A security refactor and a README update get different models, and you see every assignment before a token is spent.
• Isolated execution. Every task that changes code works on its own branch. Passing work lands on one integration branch in plan order, and you choose when to merge it.
• Operations in the right place and order. A deploy, a cloud CLI call or a push runs as an ops task in your own checkout, and waits until the change it depends on is merged into your branch.
• Verdicts from evidence. Completion is decided by a marker in the runner's output, never by a model's opinion of its own work.
• A planner that cannot write. It reads, asks, and plans. Commands that would change your repository are refused.
• No extra API key. A coding agent you already pay for can be the planner. An API key from any of 25 providers works too.
• Use it where you work. A terminal UI, a VS Code extension, a CLI for scripts and a local API, all sharing one core.

Installation

npm install -g ordewell

For VS Code, install Ordewell from the Marketplace or run code --install-extension ordewell.ordewell. The extension bundles its own core and needs nothing from npm.

Requirements: Node.js 20 or newer, at least one of Claude Code, Codex or OpenCode, and git for task isolation. tmux is optional: only the terminal transport (/transport terminal) uses it, to give each task a terminal window you can open. On Windows, run the terminal UI under WSL.

Quick start

Run ordewell in your project to open the terminal UI, then type a goal. The first run lets you pick a planner and runners with /planner and /runners.

The same workflow is available from the command line:

export AI_PROVIDER=claude-code      # plan with Claude Code, Codex or OpenCode

ordewell plan --goal "Add rate limiting to the public API"
ordewell run
ordewell handoff review             # read the diff
ordewell handoff merge              # bring it onto your branch

Between plan and run, the plan is yours to edit:

ordewell task-runner 2 opencode     # move a task to another agent
ordewell task-model 3 sonnet        # or just change its model
ordewell task-deps 3 1,2            # make it wait for tasks 1 and 2

How it works

1. Plan. The planner explores your workspace without modifying it, asks clarifying questions, and produces an ordered list of tasks with dependencies.
2. Execute. Each task starts a fresh coding agent session — in its own worktree, or in your checkout for an ops task — given the results of the tasks it depends on. Independent tasks run concurrently, three at a time by default.
3. Verify. A task passes when its unique completion marker appears in the agent's output. The exit code is kept as supporting evidence.
4. Land. A passing task is merged into the run's integration branch. If the merge conflicts, the task gets a chance to resolve it; if that fails, it waits for you with the conflicting files named.
5. Hand off. When the run finishes, review the diff, then merge it, discard it, or clean up its worktrees.

Folders containing several repositories are handled as one workspace: each task gets a worktree of every repository and lands in all of them or none. See isolation and handoff for details.

Documentation

The documentation covers the planner options, skills, commands, configuration and platform notes. Design decisions, including the options that were rejected, are recorded as architecture decision records, and CONTEXT.md defines the project's vocabulary.

Where it's going

• Now: take-over, opening a structured task in its runner's own terminal UI (#58). The structured transport is the default and tmux is optional (#61).
• Next: the planner keeps watching a run after you approve it, and suggests fixes when a task fails or gets stuck.
• Later: the planner supervises a run on its own, within limits you set in advance.

The full roadmap, and where to help, is in #21.

Contributing

Contributions are welcome. Please read CONTRIBUTING.md for the build order and project layout, and report security issues through SECURITY.md rather than the public tracker.

Acknowledgements

The grilling, spec and architecture skills, along with TDD task augmentation, are adapted from Matt Pocock's skills (MIT).

License

Ordewell is licensed under the Apache License 2.0. The Ordewell name and logos are not covered by this license; see NOTICE.