LancerLab

DSL-specific tuning contract for CuTe DSL (Python JIT) kernels. Loaded by croq-tune when dsl=cute-dsl.

DSL-specific tuning contract for CuTe/CUTLASS C++ template kernels. Loaded by croq-tune when dsl=cute-cpp.

Launch an infinite AI-driven kernel optimization loop for GPU kernels. Use when the user asks to "tune", "ai-tune", "optimize", "auto-tune", or "perf-tune" a kernel. Profiles with ncu, iterates optimizations indefinitely until interrupted.

Cursor-only kernel tuning entrypoint for GPU kernels. Use when the user asks to /croq-tune, continue tuning, resume tuning, ai-tune, auto-tune, optimize, or perf-tune a kernel in this repository. Profiles with ncu, iterates optimizations indefinitely until interrupted.

Reference collection of Choreo GPU kernel implementations. Use when studying kernel patterns, tiling strategies, warp specialization, TMA/DMA forms, or pipeline staging before editing .co files.

Launch an infinite AI-driven kernel optimization loop for GPU kernels. Use when the user asks to "tune", "ai-tune", "optimize", or "auto-tune" a kernel. Profiles with ncu, iterates optimizations indefinitely until interrupted.

Durable wrapper for GPU kernel tuning. Use when the user wants to combine durable-request with croq-tune, asks for a checkpointed tuning workflow, or wants human confirmation between autonomous tuning invocation leases while keeping the base /croq-tune protocol unchanged.

Durable wrapper for kernel tuning in this repository. Use when the user asks for /croq-tune-iterative, wants to combine durable-request with croq-tune, wants checkpointed resume-or-continue control around autonomous tuning runs, or wants human confirmation between autonomous tuning invocation leases while keeping the base /croq-tune protocol unchanged.

Run operator benchmarks and single-run executables in Kebab for GEMM and elementwise add. Use when asked to execute performance baselines, compare implementations, or run quick correctness/perf checks through Make targets.

Build and bootstrap the Kebab project on Linux with CUDA, CMake, and Makefile workflows. Use when asked to set up dependencies, detect CUDA/GPU architecture, configure builds, or fix compilation environment issues.

Navigate and modify the Kebab CUDA codebase safely. Use when asked to locate GEMM, CuTe, CUDA baseline, benchmark, microbenchmark, profiling, or build logic in this repository, and when deciding where a change should be implemented.

Tune GEMM experiment settings in Kebab using config.yaml and optimization docs. Use when asked to adjust matrix sizes, precision, layout modes, tile sizes, kernel versions, or to design reproducible performance experiments.

Execute Kebab microbenchmarks and generate PTX/SASS dumps for kernel-level analysis. Use when asked to run mbench targets, inspect assembly output, compare copy or MMA kernels, or analyze low-level code generation.

Profile Kebab kernels with Nsight Compute and export readable reports. Use when asked to run tune/mtune targets, collect NCU reports, convert .ncu-rep to text summaries, or troubleshoot performance bottlenecks.

Rewrites Croqtile tutorial chapters (ch01–ch09) to match the storytelling, figure, and structure standards established in ch02 and ch03. Use when rewriting, reviewing, or restructuring any tutorial chapter, or when the user mentions chapter quality, chapter structure, or tutorial consistency.

Universal end-of-task continuation gate. Forces every agent turn that completes work to present an interactive checkpoint (AskQuestion or conversational) instead of silently finishing. Applies to ALL tasks — coding, writing, debugging, research, analysis, refactoring, any prompt. Use always, on every task completion, before ending any turn where work was performed.