Task orchestration for coding agents.
Turn a goal into an editable plan. Run it with your coding agents. Review and merge.

Website · Documentation · News

Describe what you want to build. Ordewell explores your repository, asks clarifying questions, and turns your goal into a plan you can read and edit. When you're ready, run it: independent tasks work in parallel, each on its own branch. Review the combined diff and choose when to merge it into your branch. Nothing runs until you approve the plan.

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

• 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.
• Skills your team shares. Write down how your team works once, as a skill in your repository. The planner attaches it to the tasks it fits, whichever agent runs them.
• 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. When the planner can state the exact commands, it adds a script task instead: Ordewell runs them with no model, and you approve what will run as written, or pre-approve exact commands with the effect allowlist.
• Done means reported. A task finishes when its agent reports it done through Ordewell's own tools, never because a model thinks the work looks finished. You see which tasks passed, failed or need you, and can read each one's log live.
• Task handoffs. Each task leaves a summary and optional numbered findings with evidence. Dependents get whole findings and can pull the full handoff; a loop's check sees what its fix claims and verifies it against the code.
• A supervisor for the run. The planner keeps watching after you approve a plan: at Watch it writes you a note when a task fails or gets stuck, and at Act it may retry, stop or edit tasks itself, within limits. /settings sets the supervisor level, merging automatically and the rest of your run settings.
• Read-only tasks. A review or an audit runs without write access, so it cannot change your repository.
• Research tasks. A read-only task can be research: its report goes back to the planner, which wakes to read it while you keep talking. The planner decides how many to start, within the run's Parallel limit.
• A planner that cannot write. It reads, asks, and plans. Commands that would change your repository are refused, and an effect such as a push reaches the world only as a script task or an ops task in the plan.
• 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.

npm install -g ordewell

For VS Code, install Ordewell from the Marketplace or run code --install-extension ordewell.ordewell. In Cursor, Windsurf or VSCodium, get it from Open VSX. 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. On Windows, run the terminal UI under WSL.

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

Tried it? Tell us how your first run went.

_{A real run, sped up where marked. Watch the full 48 seconds.}

1. Describe your goal. The planner explores your workspace without modifying it and asks clarifying questions.
2. Review the plan. Read the tasks and their dependencies. Edit prompts, agent assignments and models before running. To skip this step, turn autorun on in /settings and each plan starts as soon as the planner submits it.
3. Run the tasks. Each task starts a fresh coding agent session. Independent tasks run in parallel; dependent tasks wait for the work they need. Code changes stay on separate branches, and you can follow each task's progress.
4. Review and merge. Ordewell combines the changes for your review. Read the diff, then merge it into your branch or discard the run. If a task or a merge needs your attention, Ordewell shows you where.

The planner only reads; tasks run as you, in the mode the plan assigns. For exactly what each part can touch and what to use for untrusted repositories, see SECURITY.md.

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.

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. To write your own planner or task skills, see the skills guide. If Codex's sandbox cannot start on your machine, see troubleshooting.

• Now: a planner that does small things itself. It runs an exact command you consent to as a script task (#107), reads with a sandboxed shell that can loop and script (#64, #109), and keeps talking while research tasks report back (#111).
• Shipped in 0.10: the supervisor, which watches a run and, if you let it, acts within limits you set.

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

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.

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

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