bundle/README.md

HomeBank computer app

Exported by REA's export-desktop-application from a capture of the homebank package (capture Evidence ev_e0dde61fc1c0be8aca9878658fdcc47d16f2e4a2ce40157e9b881b37c76bb082). It was checked against agentenv-framework 0.9.1431.

• envs/homebank/ is a computer app on the ubuntu-24.04 base with the reference driver, for amd64 and arm64. It installs homebank from the distribution archive, launches sh -c cd "$HOME" 2>/dev/null; exec 'homebank' "$HOME"/'Documents/household.xhb' at boot and is ready when a window of class Homebank appears.
• At launch the app opens ~/Documents/household.xhb, the seed's one document.
• Owned paths: ~/.config/homebank, ~/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/homebank/seed/ before publishing the bundle.

Task record-grocery-expense

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

In HomeBank, record a new grocery expense in the Everyday Checking account of Documents/household.xhb: $64.18 paid to Fresh Market on November 4, 2026, category Groceries, memo "Weekly groceries". Then save the file.

Checks, each of which must pass:

• amount: {"kind":"value_equals","file":"Documents/household.xhb","pointer":"/root/children/[name=ope][@wording=Weekly groceries]/attributes/amount","value":-64.18,"tolerance":0.001}
• account: {"kind":"value_equals","file":"Documents/household.xhb","pointer":"/root/children/[name=ope][@wording=Weekly groceries]/attributes/account","value":"1"}
• date: {"kind":"value_equals","file":"Documents/household.xhb","pointer":"/root/children/[name=ope][@wording=Weekly groceries]/attributes/date","value":"739924"}
• payee: {"kind":"value_equals","file":"Documents/household.xhb","pointer":"/root/children/[name=ope][@wording=Weekly groceries]/attributes/payee","value":"1"}
• category: {"kind":"value_equals","file":"Documents/household.xhb","pointer":"/root/children/[name=ope][@wording=Weekly groceries]/attributes/category","value":"44"}
• rent-untouched: {"kind":"value_equals","file":"Documents/household.xhb","pointer":"/root/children/[name=ope][@wording=October rent]/attributes/amount","value":-1450,"tolerance":0.001}

REA kept the task because its checks fail on the seed (amount, account, date, payee, category) 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 record-grocery-expense --agent claude-code --model MODEL