Websites · PROJECT NOTES
KNOWS
A browser-agent benchmark for researching sources and producing Google Docs, Sheets and Slides artifacts through multi-step workflows.
What can you do with it?
Evaluate research-to-artifact workflows with task-specific structure, content, layout and browsing-trace checks plus model judges.
What should you know before using it?
Uses real Google Workspace, requiring accounts, enabled APIs and evaluator access to files; some tasks need fresh write targets. Code/data are Apache-2.0. The public Hub dataset contains prompts and rubrics, while scoring needs repository evaluators and release assets. Configure the judge and report the release tag; external pages can change. The Hub card describes v1.1.0, while the current GitHub release is v1.2.0; do not mix prompt and evaluator/gold-data versions. Official evaluation rules prohibit evaluated agents from accessing evaluator code, checkpoints.md or reference data.
What is verified here?
Runtime: The reference provisioner creates fresh Drive targets with anyone-with-link write access. The slides_17 instance 1 presentation copy requires OAuth; folder-only provisioning can use either supported credential type.
This entry was reviewed against original project material. We have not independently run the software or reproduced its results. AI-development claims are author-reported; check upstream for current versions, setup requirements, and licenses.
- Resource type
- Benchmark suite
- Provenance
- Development provenance not established
- Reviewed
- Platforms
- Web