Metadata-Version: 2.5
Name: agentbridge-bench
Version: 0.1.0
Summary: AgentBench: reproducible evaluation of single- and multi-agent AI coding setups, built on AgentBridge.
Project-URL: Homepage, https://github.com/yoonjae26/AgentBridge
Project-URL: Documentation, https://github.com/yoonjae26/AgentBridge/blob/main/docs/benchmark-methodology.md
Author: yoonjae26
License: MIT
License-File: LICENSE
Keywords: agents,ai,benchmark,claude-code,evaluation
Classifier: Development Status :: 3 - Alpha
Classifier: Environment :: Console
Classifier: Intended Audience :: Developers
Classifier: Intended Audience :: Science/Research
Classifier: License :: OSI Approved :: MIT License
Classifier: Programming Language :: Python :: 3.12
Classifier: Topic :: Software Development :: Testing
Requires-Python: >=3.12
Requires-Dist: agentbridge-core<0.2,>=0.1.0
Requires-Dist: pydantic<3,>=2.7
Requires-Dist: pytest<9,>=8.2
Requires-Dist: pyyaml<7,>=6.0.1
Requires-Dist: rich<15,>=13.7
Requires-Dist: typer<1.0,>=0.12
Description-Content-Type: text/markdown

# agentbridge-bench

**AgentBench**, the evaluation harness of [AgentBridge](https://github.com/yoonjae26/AgentBridge).
It answers one question with data: *does multi-agent collaboration beat a single
agent on real software-engineering tasks, and when?*

```bash
pip install agentbridge-bench          # also installs agentbridge-core
agentbridge-bench list
agentbridge-bench run smoke -c mock-oracle -c mock-noop     # free harness check
agentbridge-bench run core  -c claude-single -c claude-multi -n 3
agentbridge-bench report <experiment_dir>
```

`agentbridge benchmark …` runs the same commands when this package is installed.

- 12 deterministic tasks in 8 categories, with acceptance tests hidden from the agent
  until it has finished.
- Single- and multi-agent configurations, harness controls (`mock-oracle`, `mock-noop`).
- JSON, CSV, Markdown and SVG reports; missing data is reported as `n/a`, never guessed.

Paid configurations run Claude Code with **your own** account. Method and limits:
[benchmark methodology](https://github.com/yoonjae26/AgentBridge/blob/main/docs/benchmark-methodology.md).
Licensed under MIT.
