CITATION.cff

cff-version: 1.2.0
message: >-
  If you use LabCraft-Eval, please cite the repository URL, source commit SHA,
  and release manifest or result bundle used. Source code is Apache-2.0;
  benchmark content is CC-BY-NC-4.0.
title: "LabCraft-Eval: A stochastic agent evaluation environment for benign molecular-microbiology protocols"
type: software
authors:
  - name: "The LabCraft-Eval Authors"
repository-code: "https://github.com/jang1563/LabCraft-Eval"
url: "https://github.com/jang1563/LabCraft-Eval"
version: "0.1.2"
date-released: "2026-07-12"
license: "Apache-2.0"
abstract: >-
  LabCraft-Eval is an Inspect AI evaluation environment for measuring how
  well AI agents execute benign molecular-microbiology protocols inside a
  seeded stochastic laboratory simulator, with deterministic multi-axis
  trajectory scoring, citation-backed task data, and manifest-backed public
  dataset exports.
keywords:
  - agent evaluation
  - AI safety
  - bioinformatics
  - Inspect AI
  - microbiology
  - protocol evaluation