Agent Track evaluates coding agents that iteratively generate, verify, and optimize kernels.
:maxdepth: 1
setup
methods
commands
cost-analysis
Autonomous debugging and optimization capability with execution feedback.
- Testing agent frameworks (Claude Code, OpenCode)
- Evaluating kernel-specialized agents (AutoKernel, AKO4ALL)
- Production-ready kernel generation
cd agent_bench
bash test_ops.sh add --device-count 1