bundle/README.md

Planner computer app

Exported by REA's export-desktop-application from a capture of the planner-data package (capture Evidence ev_f19d8cf71cb49bddd6a2d14dc10a1180ea80b3249ea492c12fe2042102e7ea34). It was checked against agentenv-framework 0.9.1431.

• envs/planner/ is a computer app on the ubuntu-24.04 base with the reference driver, for amd64 and arm64. It installs planner-data from the distribution archive, launches sh -c cd "$HOME" 2>/dev/null; exec 'planner' "$HOME"/'Documents/website-launch.planner' at boot and is ready when a window of class Planner appears.
• planner, which owns the program the desktop entry runs, is installed with it.
• At launch the app opens ~/Documents/website-launch.planner, the seed's one document.
• Owned paths: ~/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/planner/seed/ before publishing the bundle.

Task add-user-testing

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

Add a 'User testing' task to my website launch project plan in Documents/website-launch.planner and save it.

Checks, each of which must pass:

• user-testing: {"kind":"value_equals","file":"Documents/website-launch.planner","pointer":"/root/children/[name=tasks]/children/[name=task][@name=User testing]/attributes/name","value":"User testing"}
• requirements-kept: {"kind":"value_equals","file":"Documents/website-launch.planner","pointer":"/root/children/[name=tasks]/children/[name=task][@name=Requirements]/attributes/name","value":"Requirements"}
• design-kept: {"kind":"value_equals","file":"Documents/website-launch.planner","pointer":"/root/children/[name=tasks]/children/[name=task][@name=Design]/attributes/name","value":"Design"}
• development-kept: {"kind":"value_equals","file":"Documents/website-launch.planner","pointer":"/root/children/[name=tasks]/children/[name=task][@name=Development]/attributes/work","value":"144000"}
• launch-kept: {"kind":"value_equals","file":"Documents/website-launch.planner","pointer":"/root/children/[name=tasks]/children/[name=task][@name=Launch]/attributes/name","value":"Launch"}

REA kept the task because its checks fail on the seed (user-testing) 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-user-testing --agent claude-code --model MODEL