Mahal Systems OSUniverse
Fits teams that want a difficulty ladder for desktop agents. Borrow the five tiers (paper to gold) with step budgets and the checklist-based Gemini validator.
What it asks the agent to do
- click precisely on a desktop target
- complete browser, terminal, GIMP, LibreOffice and multi-app tasks
- play simple desktop games
What to borrow from it
taxonomygrading design
Caveats
Validator is an LLM judge over screenshots, not a state check; small; difficulty tuned so agents stay under 50 percent at release.
Details
| Workloads | computer use, games |
|---|---|
| Interaction | single turn |
| Environment | desktop |
| End state checked | No |
| Tools | Docker virtual desktop (AgentDesk) |
| Grading | llm judge, human |
| Size | 160 |
| Data | synthetic |
| Personal data | none |
| License | MIT |
| Harness | own repo |
| Status | active |