CITATION.cff

cff-version: 1.2.0
message: "If you use benchflow in your research, please cite it as below."
title: "BenchFlow: framework for RL environments for LLM agents"
abstract: "BenchFlow is a framework for building RL environments to evaluate and train LLM agents. Built on the Agent Client Protocol (ACP), it provides Scene-based multi-turn, multi-agent, and multi-model evaluation in shared sandboxes — without Docker Compose or sidecar containers. Supported use cases include interactive user simulation, code-review loops, bring-your-own-skill (BYOS) skill generation, multi-turn iterative refinement, cross-model review (cheap coder + strong reviewer), and stateful service tasks against live mock APIs (Gmail, Calendar, Docs, Drive, Slack). See docs/use-cases.md."
type: software
authors:
  - name: "BenchFlow team"
    website: "https://github.com/benchflow-ai/benchflow"
repository-code: "https://github.com/benchflow-ai/benchflow"
url: "https://github.com/benchflow-ai/benchflow"
license: Apache-2.0
version: 0.3.2
keywords:
  - benchmark
  - llm-agents
  - acp
  - agent-evaluation
  - multi-turn
  - terminal-bench
  - skillsbench