Mahal Systems BrowserGym / AgentLab
A harness, not a task set; fits teams that want to run several web benchmarks with one agent loop. Borrow the gym-style observation and action space and AgentLab's trace viewer.
What it asks the agent to do
- run MiniWoB, WebArena, VisualWebArena, WebArena-Verified, WorkArena, AssistantBench and WebLINX under one interface
- define new browser tasks with the same observation and action space
What to borrow from it
sandbox designbaseline
Caveats
Brings no tasks of its own; each wrapped benchmark keeps its own hosting needs and license.
Details
| Workloads | web browsing, workflow automation |
|---|---|
| Interaction | single turn |
| Environment | browser |
| End state checked | Yes |
| Tools | Playwright browser, accessibility tree, screenshots |
| Grading | exact match, programmatic check, state diff |
| Size | Not published |
| Data | mixed |
| Personal data | possible |
| License | Apache-2.0 |
| Harness | own repo |
| Status | active |