Download bundle/README.md from earakely-scale/UmbrelloFromRequestComputerEnv-REA: direct link, hf CLI and curl.
- Browser
- Download file 4.06 kB
-
https://huggingface.co/datasets/earakely-scale/UmbrelloFromRequestComputerEnv-REA/resolve/main/bundle/README.md
- Command line
-
hf download hf://datasets/earakely-scale/UmbrelloFromRequestComputerEnv-REA/bundle/README.md
-
curl -L -o README.md https://huggingface.co/datasets/earakely-scale/UmbrelloFromRequestComputerEnv-REA/resolve/main/bundle/README.md
Umbrello computer app
Exported by REA's export-desktop-application from a capture of the umbrello package
(capture Evidence ev_ebffa2819a67223adce3ed669534631601016b20653d02c5dd858ce472a0aaeb). It was checked against agentenv-framework
0.9.1431.
envs/umbrello/is a computer app on theubuntu-24.04base with thereferencedriver, foramd64andarm64. It installsumbrellofrom the distribution archive, launchessh -c cd "$HOME" 2>/dev/null; exec 'umbrello5' "$HOME"/'Documents/library.xmi'at boot and is ready when a window of classumbrelloappears.- At launch the app opens
~/Documents/library.xmi, the seed's one document. - Owned paths:
~/.config/umbrellorc,~/Documents.data/getreturns their files and a typed JSON view of each;data/resetrestores 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 aslocalhost,user@hostasuser@localhostand 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 withstate_pathsonstart_build_capture, or--state-pathoncapture-desktop-application. tasks/smoke.jsondeploys 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/umbrello/seed/ before publishing the bundle.
Task add-member-email
tasks/add-member-email.json deploys the app, has the agent work through the computer from the prompt below, and grades
the outcome with artifacts/add-member-email-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/umbrello.toml names it, so a run of the bundle runs it.
In Umbrello, add an attribute named email of type string to the Member class in the library model Documents/library.xmi and save it.
Checks, each of which must pass:
email-string:{"kind":"value_equals","file":"Documents/library.xmi","pointer":"/root/children/[name=uml:Model][@xmi:id=m1]/children/[name=packagedElement][@name=Logical View]/children/[name=packagedElement][@name=Member]/children/[name=ownedAttribute][@name=email]/attributes/type","value":"uMUM2QvoIZjRj"}member-three-attributes:{"kind":"member_count","file":"Documents/library.xmi","pointer":"/root/children/[name=uml:Model][@xmi:id=m1]/children/[name=packagedElement][@name=Logical View]/children/[name=packagedElement][@name=Member]/children","matching":"[name=ownedAttribute][@xmi:type=uml:Property]","equals":3}name-kept:{"kind":"value_equals","file":"Documents/library.xmi","pointer":"/root/children/[name=uml:Model][@xmi:id=m1]/children/[name=packagedElement][@name=Logical View]/children/[name=packagedElement][@name=Member]/children/[name=ownedAttribute][@name=name]/attributes/type","value":"uMUM2QvoIZjRj"}memberid-kept:{"kind":"value_equals","file":"Documents/library.xmi","pointer":"/root/children/[name=uml:Model][@xmi:id=m1]/children/[name=packagedElement][@name=Logical View]/children/[name=packagedElement][@name=Member]/children/[name=ownedAttribute][@name=memberId]/attributes/name","value":"memberId"}
REA kept the task because its checks fail on the seed (email-string, member-three-attributes) 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-member-email --agent claude-code --model MODEL