earakely-scale/XournalppFromRequestComputerEnv-REA

Xournal++: a computer-use environment built from a one-line request

The real Xournal++ 1.2 handwriting notebook 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 18 minutes, with only REA's build skill and MCP tools; How it was made has the request and the steps. Xournal++ opens a one-page notebook, Documents/biology-notes.xopp, at launch: a lined page with three typed notes, Cell Biology - Week 3, Mitochondria make ATP and Ribosomes build proteins. The agent has to add a text note that says "Quiz on Friday: bring a calculator" and save. 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 Xournal++, add a text note that says "Quiz on Friday: bring a calculator" to the Cell Biology notebook page in Documents/biology-notes.xopp, and save it.

bundle/artifacts/add-quiz-note-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
note-text	text_contains	Documents/biology-notes.xopp
one-new-note	member_count	Documents/biology-notes.xopp
notes-kept	text_contains	Documents/biology-notes.xopp

An .xopp file is gzip-compressed XML, and REA reads the XML inside. Each typed note is a text element in the page's layer. The checks read the layer: its text must contain "Quiz on Friday: bring a calculator" and, with the notes' text run together, still contain the three original notes in order, and the layer must hold exactly four text notes (member_count), so the new text has to be a note of its own rather than added to an existing one. REA's draft suggested a check on the page's preview image, which Xournal++ redraws on every save, and one on the layer's whole text; Claude Code discarded the first, split the second into the new-note and kept-notes checks, and added the count.

Results

3 of 3 runs passed, about $0.56 per run.

Run	Outcome	Turns	Cost	Video
1	passed	10	$0.72	video
2	passed	10	$0.49	video
3	passed	9	$0.46	video

Run it

You need Docker. agentenv-framework builds the computer: Ubuntu 24.04 with the app installed from Ubuntu's archive (xournalpp 1.2.2, with librsvg2-common).

pip install agentenv-framework==0.9.1431
hf download earakely-scale/XournalppFromRequestComputerEnv-REA --repo-type dataset --local-dir xournalpp-request
agent-env run xournalpp-request/bundle --task add-quiz-note --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 Xournal++ environment where the agent adds a text note to a small notebook page and saves it.

It asked only the skill's upfront questions, wrote no app-specific scripts, and finished in about 18 minutes. The steps were REA's usual ones:

1. Capture. REA launched Xournal++ 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 note 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 layer holds three notes, none of them the quiz note
after_edit	pass
near_miss	fail, as expected: the note was changed to "Quiz on Friday: bring a calculater"
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 expect the notebook to keep one page with one layer: a note added on a new page or in a new layer makes their pointers match two and fail.
• The note must contain the sentence exactly as written, though words around it still pass; where it sits on the page is not checked.
• The bundle installs librsvg2-common, the SVG image loader Xournal++'s icons need, beside the app as an extra package. The build asked for it up front.
• The computer installs whatever xournalpp Ubuntu 24.04's archive holds at build time (1.2.2 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. Xournal++ itself is not included: the computer installs it from Ubuntu's archive (xournalpp 1.2.2, with librsvg2-common) at build time, under its own licence (GPL-2.0-or-later).