bundle/README.md

KmPlot computer app

Exported by REA's export-desktop-application from a capture of the kmplot package (capture Evidence ev_d8ffd3e13d18b512c3269712baf199a9ad9e5890e664ca6dd4d5822ca14f1122). It was checked against agentenv-framework 0.9.1431.

• envs/kmplot/ is a computer app on the ubuntu-24.04 base with the reference driver, for amd64 and arm64. It installs kmplot from the distribution archive, launches sh -c cd "$HOME" 2>/dev/null; exec 'kmplot' "$HOME"/'Documents/lesson_plots.fkt' at boot and is ready when a window of class kmplot appears.
• At launch the app opens ~/Documents/lesson_plots.fkt, the seed's one document.
• Owned paths: ~/.config/kmplotrc, ~/Documents. data/get returns their files and a typed JSON view of each; data/reset restores the seed.
• Seed: 2 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/QtProject.conf. To own them, capture again naming them with state_paths on start_build_capture, or --state-path on capture-desktop-application.
• 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/kmplot/seed/ before publishing the bundle.

Task change-h-formula

tasks/change-h-formula.json deploys the app, has the agent work through the computer from the prompt below, and grades the outcome with artifacts/change-h-formula-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/kmplot.toml names it, so a run of the bundle runs it.

In KmPlot, change the formula of the function h in Documents/lesson_plots.fkt to h(x) = 2x + 1 and save the plot file.

Checks, each of which must pass:

• h-new-formula: {"kind":"line_count","file":"Documents/lesson_plots.fkt","contains":">h(x) = 2x + 1<","equals":1}
• h-old-formula-gone: {"kind":"line_count","file":"Documents/lesson_plots.fkt","contains":">h(x) = 0.5x + 1<","equals":0}
• g-unchanged: {"kind":"line_count","file":"Documents/lesson_plots.fkt","contains":">g(x) = 2sin(x)<","equals":1}
• f-unchanged: {"kind":"line_count","file":"Documents/lesson_plots.fkt","contains":">f(x) = x^2","equals":1}
• three-functions: {"kind":"line_count","file":"Documents/lesson_plots.fkt","contains":"<function ","equals":3}

REA kept the task because its checks fail on the seed (h-new-formula, h-old-formula-gone) 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 change-h-formula --agent claude-code --model MODEL