bundle/README.md

CherryTree computer app

Exported by REA's export-desktop-application from a capture of the cherrytree package (capture Evidence ev_6d194ec1e9120b0e3eea507aaf60cf6fea6b34ead242762bca5b9676f145e254). It was checked against agentenv-framework 0.9.1431.

• envs/cherrytree/ is a computer app on the ubuntu-24.04 base with the reference driver, for amd64 and arm64. It installs cherrytree from the distribution archive, launches sh -c cd "$HOME" 2>/dev/null; exec 'cherrytree' '--new_window' at boot and is ready when a window of class Cherrytree 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.
• The app opens no document at launch, since the seed holds 2 documents (~/.cache/dconf/user, ~/Documents/home-notebook.ctd) and the app opens one only when it holds exactly one.
• Owned paths: ~/.cache/dconf, ~/.config/cherrytree, ~/Documents. 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.
• Neither owned nor seeded, so the app starts without what it saved there: these files or folders, which changed during the capture, ~/.config/enchant. To own them, capture again naming them with state_paths on start_build_capture, or --state-path on capture-desktop-application.
• Seeded as captured, though the app keeps them for its own running, so a stale one can change how it starts: caches ~/.cache.
• 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/cherrytree/seed/ before publishing the bundle.

Task add-chore-line

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

In CherryTree, add a line saying Clean the gutters to the Weekend chores note in my notebook Documents/home-notebook.ctd and save it.

Checks, each of which must pass:

• new-line: {"kind":"text_contains","file":"Documents/home-notebook.ctd","pointer":"/root/children/[name=node][@name=Weekend chores]","value":"Clean the gutters"}
• kept-water: {"kind":"text_contains","file":"Documents/home-notebook.ctd","pointer":"/root/children/[name=node][@name=Weekend chores]","value":"Water the tomato plants"}
• kept-sweep: {"kind":"text_contains","file":"Documents/home-notebook.ctd","pointer":"/root/children/[name=node][@name=Weekend chores]","value":"Sweep the back porch"}
• kept-bulb: {"kind":"text_contains","file":"Documents/home-notebook.ctd","pointer":"/root/children/[name=node][@name=Weekend chores]","value":"Replace the hallway bulb"}
• kept-library: {"kind":"text_contains","file":"Documents/home-notebook.ctd","pointer":"/root/children/[name=node][@name=Weekend chores]","value":"Return the library books"}
• three-notes: {"kind":"member_count","file":"Documents/home-notebook.ctd","pointer":"/root/children","matching":"[name=node]","equals":3}

REA kept the task because its checks fail on the seed (new-line) 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-chore-line --agent claude-code --model MODEL