bundle/README.md

QElectroTech computer app

Exported by REA's export-desktop-application from a capture of the qelectrotech package (capture Evidence ev_a890610feaedecf87f205f1c85fcda1608bc852f3a2c5c9718d8a2e6496168be). It was checked against agentenv-framework 0.9.1431.

• envs/qelectrotech/ is a computer app on the ubuntu-24.04 base with the reference driver, for amd64 and arm64. It installs qelectrotech from the distribution archive, launches sh -c cd "$HOME" 2>/dev/null; exec 'qelectrotech' "$HOME"/'Documents/motor-starter.qet' at boot and is ready when a window of class QElectroTech appears.
• Installed beside the app because the export was asked for them with --extra-package, as a build computer asks for the packages its request named: qelectrotech-data, libqt5sql5-sqlite.
• At launch the app opens ~/Documents/motor-starter.qet, the seed's one document.
• Owned paths: ~/.local/share/stalefiles/QElectroTech, ~/.qet, ~/Documents. data/get returns their files and a typed JSON view of each; data/reset restores the seed.
• Seed: 3 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.
• Neither owned nor seeded, so the app starts without what it saved there: these files or folders, which changed during the capture, ~/.config/QtProject.conf. To own them, capture again naming them with state_paths on start_build_capture, or --state-path on capture-desktop-application.
• 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/qelectrotech/seed/ before publishing the bundle.

Task rename-contactor-km1

tasks/rename-contactor-km1.json deploys the app, has the agent work through the computer from the prompt below, and grades the outcome with artifacts/rename-contactor-km1-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/qelectrotech.toml names it, so a run of the bundle runs it.

In QElectroTech, change the label of the contactor coil K1 in the diagram Documents/motor-starter.qet to KM1 and save it.

Checks, each of which must pass:

• coil-label-km1: {"kind":"text_equals","file":"Documents/motor-starter.qet","pointer":"/root/children/[name=diagram]/children/[name=elements]/children/[name=element][@uuid={1b6b5cb6-8885-4056-8ac4-869b57f29eac}]/children/[name=elementInformations]/children/[name=elementInformation][@name=label]","value":"KM1"}
• coil-function-kept: {"kind":"text_equals","file":"Documents/motor-starter.qet","pointer":"/root/children/[name=diagram]/children/[name=elements]/children/[name=element][@uuid={1b6b5cb6-8885-4056-8ac4-869b57f29eac}]/children/[name=elementInformations]/children/[name=elementInformation][@name=function]","value":"Pump contactor"}
• three-components: {"kind":"member_count","file":"Documents/motor-starter.qet","pointer":"/root/children/[name=diagram]/children/[name=elements]/children","matching":"[name=element]","equals":3}

REA kept the task because its checks fail on the seed (coil-label-km1) 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 rename-contactor-km1 --agent claude-code --model MODEL