bundle/README.md

Gnote computer app

Exported by REA's export-desktop-application from a capture of the gnote package (capture Evidence ev_e86b14229d6781bc12245fbd13055af395acf587f33975376921033b77b71bd5). It was checked against agentenv-framework 0.9.1431.

• envs/gnote/ is a computer app on the ubuntu-24.04 base with the reference driver, for amd64 and arm64. It installs gnote from the distribution archive, launches sh -c cd "$HOME" 2>/dev/null; exec 'gnote' at boot and is ready when a window of class gnote appears.
• Owned paths: ~/.config/gnote, ~/.local/share/gnote. data/get returns their files and a typed JSON view of each; data/reset restores the seed.
• Seed: 5 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/gnote/seed/ before publishing the bundle.

Task add-olive-oil

tasks/add-olive-oil.json deploys the app, has the agent work through the computer from the prompt below, and grades the outcome with artifacts/add-olive-oil-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/gnote.toml names it, so a run of the bundle runs it.

Add olive oil to my Shopping List note in Gnote.

Checks, each of which must pass:

• olive-oil: {"kind":"text_contains","file":".local/share/gnote/75573f01-009e-4771-acee-93df3e993db1.note","pointer":"/root/children/[name=text]/children/[name=note-content]","value":"olive oil"}
• keeps-cherry-tomatoes: {"kind":"text_contains","file":".local/share/gnote/75573f01-009e-4771-acee-93df3e993db1.note","pointer":"/root/children/[name=text]/children/[name=note-content]","value":"cherry tomatoes"}
• keeps-lemons: {"kind":"text_contains","file":".local/share/gnote/75573f01-009e-4771-acee-93df3e993db1.note","pointer":"/root/children/[name=text]/children/[name=note-content]","value":"lemons"}

REA kept the task because its checks fail on the seed (olive-oil) 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 add-olive-oil --agent claude-code --model MODEL