benchflow-aiOfficial·34 Agent Skills Included

benchflow

Benchmark coding agents in sandboxed task environments

Runs coding agents against sandboxed benchmark tasks and scores their output through one shared contract. Eliminates ad-hoc eval scripts, inconsistent scoring, and manual trajectory review across different agents and models. Includes skills for authoring tasks, reviewing submissions, auditing run trajectories, and uploading results to a shared database.

317|40

↗

npx skills add benchflow-ai/benchflow --all -g -y

Available:

Tells the agent how to run batch benchmark experiments safely, avoid sandbox and API-key pitfalls, and which sub-skill to route each task to.

All Skills in This Repository (34)

Pure Emerald Level Indicators

📦 In Repo

benchflow-ai

pcap-triage-tshark

Analyze packet captures with tshark to filter and summarize HTTP traffic.

suricata-rules-basics

Construct Suricata network intrusion detection signatures with sticky buffers and DPI logic.

suricata-offline-evejson

Run Suricata offline on PCAP files and parse EVE JSON alerts.

contribution-analysis

Quantify relative factor contributions to a response variable using R² decomposition.

pca-decomposition

Perform Principal Component Analysis with varimax rotation on standardized input matrices.

meteorology-driver-classification

Classifies meteorological and environmental variables into driver categories for attribution analysis.

trend-analysis

Detect statistically significant trends in time series data using linear regression and Mann-Kendall tests.

state-space-linearization

Linearize nonlinear dynamical systems by computing Jacobian matrices and discretizing for control design.

mpc-horizon-tuning

Tune MPC prediction horizons and cost matrices for web handling systems.

finite-horizon-lqr

Solve finite-horizon LQR problems using backward Riccati recursion and forward simulation.

integral-action-design

Implement integral control logic with anti-windup constraints for MPC systems.

flood-detection

Compare water level time-series data against flood stage thresholds to identify flood events.

Official

Intermediate

Frequently Asked Questions

FAQPage Schema

How to install BenchFlow?▼

Run `npx skills add benchflow-ai/benchflow --all -g -y` in your terminal to install all skills in this suite globally.

What does BenchFlow do?▼

It runs coding agents like Claude Code, Codex, and Gemini against sandboxed benchmark tasks and scores their results through one consistent contract. You get rewards, token usage, and full trajectories for every run.

How do I compare different coding agents?▼

Run the same task suite once per agent with `bench eval run`, then use `bench eval metrics` to aggregate pass rates, tokens, and cost side by side.

Can I create my own benchmark tasks?▼

Yes. The task-creator skill walks you from an idea to a submission-ready task with a Dockerfile, verifier tests, and an oracle solution that must score 1.0.

Do I need API keys to run benchmarks?▼

No. You can use a ChatGPT or Claude subscription via `codex login` or `claude auth login`, and BenchFlow passes that login into the sandbox automatically.