cff-version: 1.2.0
title: Agent Retrieval Bench
message: "If you use Agent Retrieval Bench, please cite the benchmark and dataset."
type: dataset
authors:
  - family-names: Qin
    given-names: Bowen
    affiliation: National University of Singapore
  - family-names: Xie
    given-names: Yi
    affiliation: Peking University
url: https://github.com/eyuansu62/agent-retrieval-bench
repository-code: https://github.com/eyuansu62/agent-retrieval-bench
repository-artifact: https://huggingface.co/datasets/eyuansu71/agent_retrieval_bench
date-released: "2026-06-09"
abstract: >-
  Agent Retrieval Bench evaluates repository context retrieval for coding
  agents across code2test, comment2context, trace2code, edit2ripple, and
  selective no-gold retrieval. It contains 427 workflow-derived samples
  evaluated against frozen base-commit corpora.
keywords:
  - code retrieval
  - coding agents
  - software engineering
  - retrieval benchmark
  - agentic retrieval
license: MIT