README.md

metadata

license: apache-2.0
pretty_name: Planner computer-use environment from a one-line request (REA)
tags:
  - computer-use
  - cua
  - gui-agent
  - rl-environment
  - agentenv
  - desktop
  - linux
  - rea
  - project-management
  - gantt
  - planning
  - from-request

Planner: a computer-use environment built from a one-line request

The real Planner 0.14 project manager running on an Ubuntu 24.04 desktop, as an agentenv computer environment. An agent completes the task only through screenshots, mouse and keyboard, and a verifier grades the saved files with no model.

Claude Code built this environment from a one-line request in about 16 minutes, with only REA's build skill and MCP tools; How it was made has the request and the steps. Planner opens a small project plan, Documents/website-launch.planner, at launch: four tasks, Requirements, Design, Development and Launch, on a Gantt chart. The agent has to add a task named User testing and save, keeping the four tasks. It sees only the screen and acts only with the mouse and keyboard.

Watch

• Agent runs (Claude Code with Claude Sonnet 5.5): run 1, run 2, run 3
• Witness, which proves the task can be done through the computer: witness.mp4

Each video is the computer's own screen recording at 10 frames per second.

The task

Add a 'User testing' task to my website launch project plan in Documents/website-launch.planner and save it.

bundle/artifacts/add-user-testing-check/verify.py reads the app's files through the computer's data/get view and passes only if every check holds:

Check	Kind	File
user-testing	value_equals	Documents/website-launch.planner
requirements-kept	value_equals	Documents/website-launch.planner
design-kept	value_equals	Documents/website-launch.planner
development-kept	value_equals	Documents/website-launch.planner
launch-kept	value_equals	Documents/website-launch.planner

A .planner file is XML with one task element per task. Planner renumbers its tasks' ids by position when it saves, so the checks find each task by its name, never by id or position: a task named "User testing" must exist, Requirements, Design and Launch must still be there, and Development must keep its five days of work (work 144000, in seconds). REA's draft took the ids at their word, read the inserted task as a new Launch task plus task 4 renamed, and suggested checks on task 4's name and on the defaults of the "new" Launch. Claude Code kept the name check, finding the task by name instead of id, and wrote the rest.

Results

3 of 3 runs passed, about $0.71 per run.

Run	Outcome	Turns	Cost	Video
1	passed	13	$0.76	video
2	passed	17	$0.89	video
3	passed	10	$0.48	video

Run it

You need Docker. agentenv-framework builds the computer: Ubuntu 24.04 with the app installed from Ubuntu's archive (planner 0.14.91).

pip install agentenv-framework==0.9.1431
hf download earakely-scale/PlannerFromRequestComputerEnv-REA --repo-type dataset --local-dir planner-request
agent-env run planner-request/bundle --task add-user-testing --agent claude-code --model <model>

The agent needs a model provider configured for agentenv. --task smoke only boots the computer, with no agent. bundle/README.md describes the bundle in detail.

How it was made

Claude Code built this environment from a one-line request, given only REA's build skill and REA's MCP tools:

Build me a Planner environment where the agent adds a 'User testing' task to a small project plan and saves it.

It asked only the skill's upfront questions, wrote no app-specific scripts, and finished in about 16 minutes. The steps were REA's usual ones:

1. Capture. REA launched Planner on a Linux desktop and recorded its package, windows and opened files while the edit was made. It also recorded the app's state folders before and after.
2. Export. REA turned the capture into an agentenv computer_app. It installs the app, seeds the starting files, starts the app, and waits for its main window. It also serves typed views of the app's files for verifiers.
3. Task. The task's checks were validated against the capture: they fail on the starting state and pass on the captured end state.
4. Witness. In a fresh computer, the task was added through computer actions alone, and the checks were read at four stages. They held on the first try:

Stage	Checks
seed	fail, as expected: no task is named User testing; the four kept-task checks pass
after_edit	pass
near_miss	fail, as expected: the new task was renamed "User tesing"
after_reset	fail, as expected: data/reset restored the seed

The agent runs came after the build: Claude Code with Claude Sonnet 5.5 (--model anthropic/claude-sonnet-5-5 --no-emulate, on an amd64 host) solved the task 3 times, each run recorded.

What's inside

bundle/      the agentenv bundle: envs/ (app module, seed files), tasks/, artifacts/ (verifier), evals/
media/       witness and agent-run recordings (MP4) with short GIF previews
results.json the runs above, with each check's result

Limits

• The checks look for User testing among the plan's top-level tasks, and its name must match exactly. Added as a subtask, indented under another task, or named "User Testing", it fails.
• Only the new task's name is checked, not its duration or where it sits in the plan. Requirements, Design and Launch are checked by name only, so changing their durations still passes.
• The computer installs whatever planner Ubuntu 24.04's archive holds at build time (0.14.91 for these runs); it is not pinned.
• Three runs is a small sample: the solve rate shows the task is solvable, not how hard it is.

Licences

The bundle code, the task and the synthetic seed files are Apache-2.0. Planner itself is not included: the computer installs it from Ubuntu's archive (planner 0.14.91) at build time, under its own licence (GPL-2.0-or-later).