earakely-scale/ZimFromRequestComputerEnv-REA

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

The real Zim 0.75 desktop wiki 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 27 minutes, with only REA's build skill and MCP tools; How it was made has the request and the steps. Zim opens a notebook, Notes, at launch, on its empty Home page. The notebook's Meeting Notes page holds a weekly sync's attendees, decisions and two action items written as unchecked checkboxes. The agent has to open Meeting Notes, add a third action item, "Dana to update the release checklist", as an unchecked checkbox, and save the page. 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

In my Zim notebook, add a new action item to the Meeting Notes page: "Dana to update the release checklist". Add it as an unchecked checkbox in the Action items list, like the others there, and save the page.

bundle/artifacts/add-action-item-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
dana-item	lines_present	Notebooks/Notes/Meeting_Notes.txt
dana-once	line_count	Notebooks/Notes/Meeting_Notes.txt
other-items-kept	lines_present	Notebooks/Notes/Meeting_Notes.txt

A Zim page is a text file in Zim's wiki markup, here Notebooks/Notes/Meeting_Notes.txt, where an unchecked checkbox is written [ ] at the start of its line. The checks read the page line by line: it must have the line "[ ] Dana to update the release checklist", only one line may mention Dana's item, and the two existing action items must still be there. REA's draft suggested the first check as written; Claude Code added the other two.

Results

3 of 3 runs passed, about $0.57 per run.

Run	Outcome	Turns	Cost	Video
1	passed	11	$0.71	video
2	passed	12	$0.25	video
3	passed	11	$0.74	video

Run it

You need Docker. agentenv-framework builds the computer: Ubuntu 24.04 with the app installed from Ubuntu's archive (zim 0.75.2).

pip install agentenv-framework==0.9.1431
hf download earakely-scale/ZimFromRequestComputerEnv-REA --repo-type dataset --local-dir zim-request
agent-env run zim-request/bundle --task add-action-item --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 Zim environment where the agent adds an action item to a meeting-notes page.

It asked only the skill's upfront questions, wrote no app-specific scripts, and finished in about 27 minutes. The steps were REA's usual ones:

1. Capture. REA launched Zim 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 item 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: the page has no line for Dana's item; the kept-items check passes
after_edit	pass
near_miss	fail, as expected: the new item's checkbox was ticked
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

• Zim starts on the notebook's Home page, which is empty, so the agent must open Meeting Notes from the index on the left. After data/reset, Zim reopens on the last page it showed instead: its ~/.cache/zim is not part of the seed.
• The new item's place on the page is not checked: the line passes anywhere, not only in the Action items list. A ticked checkbox ([*]), a bullet or different wording fails.
• The verifier reads the page as soon as the agent stops. Zim also saves an edited page by itself on a timer (autosave_timeout=15 in its preferences), so a run that leaves the saving to Zim can be graded on the unsaved page. The prompt asks for a save, and all three runs passed.
• The computer installs whatever zim Ubuntu 24.04's archive holds at build time (0.75.2 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. Zim itself is not included: the computer installs it from Ubuntu's archive (zim 0.75.2) at build time, under its own licence (GPL-2.0-or-later).