Add PyTorch & FlashInfer as community project

[shljessie]

• Add PyTorch (12 skills) and FlashInfer (3 skills) to CI sync check as community projects with active NVIDIA contribution
• Add "Community Projects with NVIDIA Contribution" section to README

Wondering what your thought are on this @mosheabr? These are public open-source repos but we have active NVIDIA engineers & PMs contributing to them.

[mosheabr]

Thanks for putting this together! The README section and table formatting look great. However, a couple of blockers:

Must Fix

1. Action version bumps will break CI — actions/checkout@v5, upload-artifact@v6, download-artifact@v7, and peter-evans/create-pull-request@v8 don't exist. These need to be reverted to their current versions (v4, v4, v4, v7 respectively). Even if they did exist, version bumps should be a separate PR.

Scope

2. Community projects are out of scope for this catalog — For now, we're keeping this repo scoped to skills that NVIDIA owns and maintains. PyTorch and FlashInfer are great projects with active NVIDIA contribution, but since we don't own those repos or their skills, we can't guarantee their quality, count stability, or alignment with our signing/verification pipeline.

   This isn't a "never" — we may revisit community/partner tiers down the road — but for the initial catalog we need to keep it tight.

Happy to discuss further if you have thoughts on how community skills could fit in the future!

[mosheabr]

Thanks for putting this together @shljessie — love the idea of tracking community projects where NVIDIA engineers are actively contributing. PyTorch and FlashInfer are great candidates.

That said, we're going to hold off on adding community/third-party skills to the catalog until we finalize the signing and scanning process internally. Right now the catalog only includes NVIDIA-verified skills that go through our security pipeline (SkillSpector + Xray). Community skills would fall under a different trust tier and we need to have the scanning infrastructure in place before we start listing them — even if NVIDIA engineers are contributing to those repos.

This is on the roadmap and we'll circle back once the internal pipeline is production-ready. Please keep this PR open so we can revisit when the time is right.

[mosheabr]

Closing — out of scope for the current catalog.

The NVIDIA/skills catalog publishes skills only from NVIDIA-owned GitHub orgs (NVIDIA, NVIDIA-AI-Blueprints, NVIDIA-NeMo, NVIDIA-AI-IOT, NVIDIA-dev, ai-dynamo, nvidia-riva, nv-legate). PyTorch (pytorch/pytorch) and FlashInfer (flashinfer-ai/flashinfer) are valuable projects but live outside NVIDIA-controlled orgs, so they can't be onboarded through the same catalog gate that handles license/IP attestation and NVCARPS signing.

Two paths for community projects like these:

1. Skill content that targets PyTorch/FlashInfer can still live in NVIDIA repos — e.g., NVIDIA/TensorRT-LLM/skills/ could include a tensorrt-llm-with-pytorch skill that uses PyTorch as a dependency. The skill is NVIDIA-authored, just exercises the upstream project.

2. Anthropic's community catalog (anthropics/skills) or per-project repos are better venues for skills NVIDIA doesn't author.

Thanks for raising the question early — it helped clarify scope for the catalog. Closing this branch since neither project meets the current ownership gate, but the conversation is documented here for the next time this comes up.