Agent Track

Agent Track evaluates coding agents that iteratively generate, verify, and optimize kernels.

• Setup
• Methods
• Commands
• Cost Analysis

What It Tests

Autonomous debugging and optimization capability with execution feedback.

When to Use

• Testing agent frameworks (Claude Code, OpenCode)

• Evaluating kernel-specialized agents (AutoKernel, AKO4ALL)

• Production-ready kernel generation

Quick Start

cd agent_bench
bash test_ops.sh add --device-count 1