SlackBench: Benchmarking Agents on Collaborative Projects Grounded in Real Code Repositories
Paper • 2611.05512 • Published • 1
Environment data for paper SlackBench: Benchmarking Agents on Collaborative Projects Grounded in Real Code Repositories (accepted to NeurIPS 2026 Evaluation and Dataset track).
Link to paper: https://arxiv.org/abs/2611.05512.
We will be publishing the slackbench dataset in harbor framework shortly.
Due to reward hacking concerns (see section 6.3 of paper), we will not be publishing all the queries and reference answer online. We will make them available to researchers upon request. Please email huanzhimao@berkeley.edu or mingxuanl@uchicago.edu.