bundle/README.md

Inkscape computer app

Exported by REA's export-desktop-application from a capture of the inkscape package (capture Evidence ev_0da6a9a44d419cae65e63e9313874ef8b262f45cf66c1c9e0d2d53b4efa84d18). It was checked against agentenv-framework 0.9.1431.

• envs/inkscape/ is a computer app on the ubuntu-24.04 base with the reference driver, for amd64 and arm64. It installs inkscape from the distribution archive, launches sh -c cd "$HOME" 2>/dev/null; exec 'inkscape' at boot and is ready when a window of class Inkscape appears.
• Owned paths: ~/.config/inkscape, ~/.local/share/Trash/files, ~/Documents. data/get returns their files and a typed JSON view of each; data/reset restores the seed.
• Seed: 6 file(s) from the user's home 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 is synthetic test data made for this showcase; it holds no one's real files.

Task poster-headline

tasks/poster-headline.json deploys the app, has the agent work through the computer from the prompt below, and grades the outcome with artifacts/poster-headline-check/verify.py, which reads the app's data/get views and needs no model. evals/inkscape.toml names it, so a run of the bundle runs it.

Change the poster's headline in Documents/poster.svg to Autumn Seed Exchange and save it.

Checks, each of which must pass:

• headline-text: {"kind":"value_equals","file":"Documents/poster.svg","pointer":"/root/children/[name=g][@inkscape:label=Layer 1]/children/[name=text][@id=headline]/children/[name=tspan]/children/0","value":"Autumn Seed Exchange"}
• date-unchanged: {"kind":"value_equals","file":"Documents/poster.svg","pointer":"/root/children/[name=g][@inkscape:label=Layer 1]/children/[name=text][@id=event-date]/children/[name=tspan]/children/0","value":"Saturday 12 April, 10 am to 3 pm"}

REA kept the task because its checks fail on the seed (headline-text) 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 poster-headline --agent claude-code --model MODEL