earakely-scale/PoeditFromRequestComputerEnv-REA

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

The real Poedit 3.4 translation editor 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 29 minutes, with only REA's build skill and MCP tools; How it was made has the request and the steps. Poedit opens a French translation file, Documents/fr.po, at launch: the 34 messages of Python's argparse module, seven of them translated. One translation is wrong: "expected one argument" reads "un argument inattendu". The agent has to correct it to "un argument attendu" 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 Poedit, the French translation of "expected one argument" in Documents/fr.po is wrong. Fix it so it reads "un argument attendu", and save the file.

bundle/artifacts/fix-french-translation-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
fixed	lines_present	Documents/fr.po
wrong-gone	lines_absent	Documents/fr.po
others-kept	lines_present	Documents/fr.po

A .po file is gettext text, and REA reads it as lines. The checks need the line msgstr "un argument attendu", no line msgstr "un argument inattendu", and five other translations' msgstr lines kept. REA's draft suggested the first two checks, kept exactly as written, and a guard that held the edited entry's own source reference and msgid, which a translator can't change; Claude Code rebuilt that guard on the other translations.

Results

3 of 3 runs passed, about $0.82 per run.

Run	Outcome	Turns	Cost	Video
1	passed	15	$0.58	video
2	passed	15	$0.78	video
3	passed	14	$1.09	video

Run it

You need Docker. agentenv-framework builds the computer: Ubuntu 24.04 with the app installed from Ubuntu's archive (poedit 3.4.2).

pip install agentenv-framework==0.9.1431
hf download earakely-scale/PoeditFromRequestComputerEnv-REA --repo-type dataset --local-dir poedit-request
agent-env run poedit-request/bundle --task fix-french-translation --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 Poedit environment where the agent fixes a wrong translation in a French .po file and saves it.

It asked only the skill's upfront questions, wrote no app-specific scripts, and finished in about 29 minutes.

The first witness computer booted into a lock-permission error: REA had seeded Poedit's single-instance lock and its translation memory's lock, and the first had lost its private file mode. The build agent deleted both lock files from the bundle by hand and witnessed it again. The steps were REA's usual ones:

1. Capture. REA launched Poedit 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, with the lock files removed, the edit was made 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 translation reads "un argument inattendu"; the kept check passes
after_edit	pass
near_miss	fail, as expected: the translation was retyped as "un argument atendu"
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

• REA seeded two lock files, wxWidgets' single-instance ~/poedit.lock and the translation memory's ~/.local/share/poedit/TranslationMemory/write.lock, and seeding lost the first one's private file mode, so the first witness computer booted into a lock-permission error. The build agent deleted both from the bundle by hand; this is the bundle that was then witnessed and solved, published as built. The bug is being fixed in REA.
• The checks read the file as lines and can't tie a msgstr to its msgid: they rely on each translation line being unique in the seed.
• The seed also holds Poedit's translation memory (~/.local/share/poedit/TranslationMemory) and the compiled Documents/fr.mo beside the .po, which the capture picked up; the checks read only fr.po.
• The computer installs whatever poedit Ubuntu 24.04's archive holds at build time (3.4.2 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. Poedit itself is not included: the computer installs it from Ubuntu's archive (poedit 3.4.2) at build time, under its own licence (MIT).