bundle/README.md

Poedit computer app

Exported by REA's export-desktop-application from a capture of the poedit package (capture Evidence ev_a689e17edeb1eed905876ae6664f816ac921042e07370a1cb68641e74bb09a31). It was checked against agentenv-framework 0.9.1431.

• envs/poedit/ is a computer app on the ubuntu-24.04 base with the reference driver, for amd64 and arm64. It installs poedit from the distribution archive, launches sh -c cd "$HOME" 2>/dev/null; exec 'poeditor' "$HOME"/'Documents/fr.po' at boot and is ready when a window of class Poedit appears.
• At launch the app opens ~/Documents/fr.po, the seed's one document.
• Owned paths: ~/.local/share/poedit, ~/Documents, ~/poedit.lock. data/get returns their files and a typed JSON view of each; data/reset restores the seed.
• Seed: 12 file(s) made for the task, as saved at checkpoint seed. Text files and dconf keys spell the capture host's name as localhost, user@host as user@localhost and the home as /home/user, which the computer respells to its own at boot.
• tasks/smoke.json deploys the app.

Run the task with agent-env run <this folder>, or only the deploy with --task smoke.

The seed was made for this task, not taken from a user's own data; read envs/poedit/seed/ before publishing the bundle.

Task fix-french-translation

tasks/fix-french-translation.json deploys the app, has the agent work through the computer from the prompt below, and grades the outcome with artifacts/fix-french-translation-check/verify.py, which reads the app's data/get views and needs no model. It grades once the app's files have held for 3 s, so a save the app makes on its own within that time of the agent stopping counts; each result's settled is false when they were still changing 15 s after the first read. evals/poedit.toml names it, so a run of the bundle runs it.

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.

Checks, each of which must pass:

• fixed: {"kind":"lines_present","file":"Documents/fr.po","lines":["msgstr \"un argument attendu\""]}
• wrong-gone: {"kind":"lines_absent","file":"Documents/fr.po","lines":["msgstr \"un argument inattendu\""]}
• others-kept: {"kind":"lines_present","file":"Documents/fr.po","lines":["msgstr \"au plus un argument attendu\"","msgstr \"au moins un argument attendu\"","msgstr \"afficher ce message d'aide et quitter\"","msgstr \"afficher la version du programme et quitter\"","msgstr \"arguments positionnels\""]}

REA kept the task because its checks fail on the seed (fixed, wrong-gone) and all pass on the captured end state.

To measure a solve rate, run the task k times (3 by default) with the agent and model to measure, and count the runs whose summary line says passed; a run that says failed did not finish and is run again rather than counted:

agent-env run <this folder> --task fix-french-translation --agent claude-code --model MODEL