bundle/README.md

Glade computer app

Exported by REA's export-desktop-application from a capture of the glade package (capture Evidence ev_15f3718c281812dc693d84a88bc75f40cd9c1e587f72fe5099c216ffeecbe2f7). It was checked against agentenv-framework 0.9.1431.

• envs/glade/ is a computer app on the ubuntu-24.04 base with the reference driver, for amd64 and arm64. It installs glade from the distribution archive, launches sh -c cd "$HOME" 2>/dev/null; exec 'glade' "$HOME"/'Documents/feedback.glade' at boot and is ready when a window of class Glade 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: librsvg2-common.
• At launch the app opens ~/Documents/feedback.glade, the seed's one document.
• Owned paths: ~/Documents. data/get returns their files and a typed JSON view of each; data/reset restores the seed.
• Seed: 1 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/glade/seed/ before publishing the bundle.

Task rename-submit-button

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

In Glade, change the label of the Submit button in the feedback dialog design in Documents/feedback.glade to Send Feedback and save it.

Checks, each of which must pass:

• submit-label: {"kind":"text_equals","file":"Documents/feedback.glade","pointer":"/root/children/[name=object][@id=feedback_dialog]/children/[name=child]/children/[name=object]/children/[name=child][@internal-child=action_area]/children/[name=object]/children/[name=child][>packing=TrueTrue1]/children/[name=object]/children/[name=property][@name=label]","value":"Send Feedback"}
• cancel-kept: {"kind":"text_equals","file":"Documents/feedback.glade","pointer":"/root/children/[name=object][@id=feedback_dialog]/children/[name=child]/children/[name=object]/children/[name=child][@internal-child=action_area]/children/[name=object]/children/[name=child][>packing=TrueTrue0]/children/[name=object]/children/[name=property][@name=label]","value":"Cancel"}
• two-buttons: {"kind":"member_count","file":"Documents/feedback.glade","pointer":"/root/children/[name=object][@id=feedback_dialog]/children/[name=child]/children/[name=object]/children/[name=child][@internal-child=action_area]/children/[name=object]/children","matching":"[name=child]","equals":2}

REA kept the task because its checks fail on the seed (submit-label) 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-submit-button --agent claude-code --model MODEL