tile-aiOfficial·10 Agent Skills Included

TileOPs

Spec-driven GPU operator development, testing, and PR review

Builds and maintains high-performance GPU operators for large language models from a single declarative spec. Generates operator code, tests, and benchmarks automatically, then validates every result against PyTorch references and performance models. Automates pull request reviews, review-thread resolution, and follow-up issue creation to remove manual code review bottlenecks.

168|55

↗

npx skills add tile-ai/TileOPs --all -g -y

Available:

Tells the agent that the manifest spec is the source of truth and directs it to load the correct domain rule files before modifying tests, kernels, benchmarks, or manifest entries.

All Skills in This Repository (10)

Pure Emerald Level Indicators

📦 In Repo

tile-ai

lifecycle-pull-request

Automate GitHub pull request lifecycle from draft creation to human review.

creating-issue

Create validated GitHub issues with structured templates and gh CLI labels.

lifecycle-issue-fixer

Resolve GitHub issues via TDD, worktree isolation, and pull request creation.

tune-multiplication

Optimize GEMV, GEMM, and grouped GEMM kernels on NVIDIA Ampere and Hopper GPUs.

committing-changes

Enforce Git commit and branch workflows for the TileOPs repository.

tune

Profile TileOPs GPU kernel latency and bottlenecks with CUDA-event timings.

check-kernel-format

Validate kernel and op format conformance for T.prim_func and TileLang macros.

kernel-debug

Diagnose TileLang GPU kernel correctness failures across configurations and shapes.

migrating-new-op

Migrate operators into TileOPs with a phased kernel-to-op workflow.

creating-pull-request

Create draft GitHub pull requests with validated titles, bodies, and labels via gh CLI.

Official

Intermediate

Frequently Asked Questions

FAQPage Schema

How to install TileOPs?▼

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

What does TileOPs do?▼

TileOPs is a GPU operator library where every operator is declared in a manifest spec first, and code, tests, and benchmarks are generated and validated against that spec.

How do the TileOPs skills automate operator development?▼

Skills like align-op and scaffold-op read a manifest entry, generate the operator code, write PyTorch-referenced tests, run benchmarks, and flip the spec status once everything passes.

Can TileOPs skills review pull requests automatically?▼

Yes. The review-tileops and resolve-tileops skills review PRs under a separate GitHub identity, post inline comments, and drive multi-round resolution until approval.

Do I need a GPU to use TileOPs?▼

The library itself requires a CUDA-capable NVIDIA GPU to run operators and benchmarks, but the spec editing, manifest, and review skills work without one.