benchflow 0.4.0

Multi-turn agent benchmarking with ACP — run any agent, any model, any provider.

BenchFlow

Multi-turn agent benchmarking — Scene-based lifecycle for any ACP agent

What

BenchFlow runs AI agents against benchmark tasks in sandboxed environments. Single-agent, multi-agent, and multi-round patterns share one Scene-based lifecycle.

• Any ACP agent — Gemini CLI, Claude Code, Codex, OpenCode, OpenHands, OpenClaw, Pi, or your own
• Single + multi + progressive — single-agent / multi-agent (coder + reviewer, simulated user) / multi-round with a Python BaseUser callback
• Sandboxes — Docker locally, Daytona for parallel cloud runs, Modal for serverless/GPU-backed task environments
• Hardened verifier — defaults block BenchJack/Meerkat-style reward-hacking; tasks opt out per-feature

Install

uv tool install benchflow

Requires Python 3.12+ and uv. Set DAYTONA_API_KEY for Daytona runs or configure Modal auth for Modal runs; export the relevant agent API key (GEMINI_API_KEY, ANTHROPIC_API_KEY, etc.) or run claude login / codex --login for subscription auth.

Documentation

Start with Getting started, then Concepts for the mental model. Then by goal:

If you want to…	Read
Run an eval on an existing task	Getting started
Understand Rollout / Scene / Role / Verifier	Concepts
Author a new task	Task authoring
Multi-agent: coder + reviewer, simulated user, BYOS, stateful envs	Use cases
Multi-round single-agent (progressive disclosure, oracle access)	Progressive disclosure
Skill evaluation (when the artifact is a skill, not a workspace)	Skill eval
Understand the security model	Sandbox hardening
CLI flags + commands	CLI reference
Python API surface	Python API reference

Notebooks and runnable example scripts live under docs/examples/ so examples stay versioned with the docs that explain them.

Benchmark task sources

Benchmark datasets live in external Git repos and are referenced with two fields:

# benchmarks/skillsbench-claude-glm51.yaml
source:
  repo: benchflow-ai/skillsbench   # GitHub org/repo
  path: tasks                       # optional subpath within repo
  ref: main                         # optional branch/tag
agent: claude-agent-acp
model: claude-sonnet-4-6

Run any benchmark via the CLI:

# From a YAML config
bench eval create --config benchmarks/skillsbench-claude-glm51.yaml

# Inline — mirrors the YAML source fields
bench eval create \
    --source-repo benchflow-ai/skillsbench --source-path tasks \
    --agent gemini --model gemini-3.1-flash-lite-preview --sandbox daytona --concurrency 64

Repos are cloned and cached locally under .cache/datasets/ on first use.

SkillsBench itself sources BenchFlow from GitHub main in its pyproject.toml. After a BenchFlow change lands, run uv lock --upgrade-package benchflow in SkillsBench when you need its lockfile to point at the newest BenchFlow commit.

Featured

• Progressive disclosure on SWE-bench Pro — the BaseUser abstraction drives a multi-round rollout: terse round-0 prompt → failing-test hints → full spec. 5/5 oracle on Daytona, runnable demo at docs/examples/swebench_pro_progressive_disclosure.ipynb. See Progressive disclosure.

Research artifacts

Two runnable labs validate the security story:

• labs/benchjack-sandbox-hardening/ — end-to-end demo that 0.2.1+ blocks three BenchJack exploits that flip 0.2.0's reward from 0.0 to 1.0.
• labs/reward-hack-matrix/ — full reward-hack sweep across real benchmarks comparing 0.2.0 vs 0.2.2.

Audience

• Eval researchers / paper writers → Getting started → Concepts → Use cases
• Task authors → Task authoring → Sandbox hardening
• Agent builders integrating with benchflow → Concepts → Python API reference → benchflow.agents.registry
• External benchmark adapters → Task authoring → Progressive disclosure

Contributing

PRs welcome. Open against main. CI runs ruff + tests on every PR; please run ruff check . and pytest tests/ locally first.

For a release: bump pyproject.toml to the next stable version, tag v<version> on main, push the tag — CI publishes to PyPI. Then bump main to the next .dev0.

License

Apache-2.0.