[NVIDIA #1] Install NVIDIA Skills into opencode-owl skill registry (P0)

[AugustChaoTW]

Summary

NVIDIA published an official agent skills repository at nvidia/skills using the identical SKILL.md format that opencode-owl's skill_install already supports. This issue tracks installing all relevant NVIDIA skills into the local registry with zero code changes.

GPU Environment

• GPU: NVIDIA GB10 (Grace Blackwell) — 121.6 GB VRAM, SM 12.1
• CUDA: 13.0
• Ollama: running with qwen3.5-35b, qwen3.6-reasoning, gpt-oss:20b, etc.

Skills to Install

Run the following skill_install MCP calls (or equivalent CLI) for each skill:

Skill	Path	Description
aiq-deploy	skills/aiq-deploy	Deploy NVIDIA AI-Q Blueprint infrastructure
aiq-research	skills/aiq-research	Run deep research via AI-Q backend
cuopt-install	skills/cuopt/cuopt-install	Install cuOpt (GPU-accel optimizer)
cuopt-developer	skills/cuopt/cuopt-developer	cuOpt development workflows
cuopt-routing-api-python	skills/cuopt/cuopt-routing-api-python	VRP/TSP routing optimization
cuopt-numerical-optimization-api-python	skills/cuopt/cuopt-numerical-optimization-api-python	LP/MILP/QP optimization

# Via MCP (from Claude Code with opencode-owl MCP active)
skill_install({ repo: "nvidia/skills", path: "skills/aiq-deploy" })
skill_install({ repo: "nvidia/skills", path: "skills/aiq-research" })
skill_install({ repo: "nvidia/skills", path: "skills/cuopt/cuopt-install" })
skill_install({ repo: "nvidia/skills", path: "skills/cuopt/cuopt-developer" })
skill_install({ repo: "nvidia/skills", path: "skills/cuopt/cuopt-routing-api-python" })
skill_install({ repo: "nvidia/skills", path: "skills/cuopt/cuopt-numerical-optimization-api-python" })

Compatibility Notes

• aiq-research requires AIQ_SERVER_URL env var (or a local AI-Q server at localhost:8000)
• cuopt-* requires CUDA ≥ SM 7.0 — GB10 (SM 12.1) fully satisfies this requirement
• All skills are Apache-2.0 licensed

Acceptance Criteria

• All 6 skills installed and visible in skill_list()
• skill_check_compatibility passes for cuopt skills on this GPU
• skill_search("cuopt") returns cuopt skills
• skill_search("deep research") returns aiq-research
• Skills injected into AHE eval prompts when relevant tasks are run

Effort: 10 minutes | Priority: P0 (no code change required)