| name | spark-docs-skill |
| category | productivity |
| description | USE THIS SKILL for any question about NVIDIA DGX Spark hardware, DGX OS 7, system setup, first boot, recovery, dashboard, NGC, containers, or DGX-specific software. This skill has complete local documentation — do not use web search or WebFetch instead. |
| pairs_with | nvidia-dgx-research |
NVIDIA DGX Spark Documentation Skill
Intelligent documentation access for the NVIDIA DGX Spark (GB10) and DGX OS 7.
Path Resolution
Use ${CLAUDE_SKILL_DIR} to reference this skill's directory — it resolves automatically whether the skill is symlinked or local.
Overview
This skill bundles official NVIDIA documentation for two tightly coupled products:
- DGX Spark — the GB10 desktop / workstation system (hardware, software stack, dashboard, recovery, common use cases)
- DGX OS 7 — the Ubuntu-based operating system that ships on Spark (installation, networking, storage, additional software)
Sources are Sphinx-generated documentation trees on docs.nvidia.com, fetched and indexed for local lookup. No llms.txt exists for these products, so the skill maintains its own snapshot.
Documentation Update Requests
When user requests to update documentation (trigger: "update spark docs" / "update dgx documentation"):
"${CLAUDE_SKILL_DIR}/scripts/update-docs.sh"
What gets updated:
- Quick Reference — Tier 1 — key pages: overview, first boot, hardware, common use cases, known issues
- Section Index — Tier 2 — every page organized by
dgx-spark vs dgx-os-7-user-guide
- Full Documentation — Tier 3 — complete content with grep-friendly page markers
Duration: ~30 seconds (≈ 50 small pages with rate limiting)
Error Handling:
- Each page fetch status is logged
- Failed pages are skipped with warnings
- Aborts if <10 pages fetched (critical failure)
3-Tier Routing
Tier 1: Quick Reference
- File:
references/quick-reference.md
- Covers: System overview, first-boot setup, hardware spec, software stack, common use cases, known issues, DGX OS 7 entrypoint
- Use when: Initial Spark setup, hardware questions, "what is X on Spark?"
Example queries:
- "What hardware is in DGX Spark?"
- "How do I first-boot a DGX Spark?"
- "What common use cases does NVIDIA highlight for Spark?"
- "What are the known issues on DGX Spark right now?"
Tier 2: Section Index
- File:
references/section-index.md
- Covers: Every page in both doc trees, with title and short description
- Use when: Locating a specific topic, scoping which pages cover an area
Example queries:
- "Where is recovery documented for DGX Spark?"
- "What does the DGX OS 7 guide cover under additional software?"
- "List all pages in the DGX Spark section"
Access pattern:
grep -A 30 "^## DGX-SPARK" references/section-index.md
grep -A 30 "^## DGX-OS-7-USER-GUIDE" references/section-index.md
grep -i "recovery" references/section-index.md
Tier 3: Full Documentation
- File:
references/full-docs.txt
- Covers: Complete prose for every page with grep markers
- Use when: Detailed instructions, exact command names, troubleshooting deep dives
IMPORTANT: Never load the entire file. Use grep markers for targeted extraction.
Access patterns:
grep -A 200 "^PAGE: /dgx-spark/first-boot.html" references/full-docs.txt
grep "^PAGE:" references/full-docs.txt | grep "/dgx-spark/"
grep -n "nvidia-sync" references/full-docs.txt
sed -n '/^PAGE: \/dgx-spark\/system-recovery.html/,/^═\{80\}$/p' references/full-docs.txt
Companion Skill
For questions that fall outside the bundled documentation — community issues on the NVIDIA developer forum, recent posts on the NVIDIA developer blog, or adjacent product docs (CUDA, NeMo, Jetson, Brev, etc. via the llms.txt catalog at docs.nvidia.com/llms.txt) — use the nvidia-dgx-research skill.
This skill = snapshot. The companion = live search.
Response Format
Always cite tier and source:
- "From Quick Reference: …"
- "From Section Index (DGX-OS-7-USER-GUIDE): …"
- "From Full Docs (/dgx-spark/first-boot.html): …"
Provide context:
- Mention which doc tree the answer comes from (Spark vs OS 7)
- Link to the official URL when relevant:
https://docs.nvidia.com/dgx/<path>
- Note "known issues" page references if the topic is recent
Best Practices
DO
- Start with Tier 1 for overview / setup / hardware questions
- Use Tier 2 to find which page covers a specific topic
- Use targeted grep on Tier 3 for deep details (never load full file)
- Cite the page path in responses
DON'T
- Load entire
full-docs.txt
- Answer from training data — these are recent products, always check the local snapshot
- Use WebFetch / WebSearch for anything in scope of these docs
- Confuse DGX Spark (Grace Blackwell desktop) with the larger DGX servers — they're different products
File Locations
spark-docs-skill/
├── SKILL.md
├── docs-config.yaml
├── references/
│ ├── quick-reference.md
│ ├── section-index.md
│ └── full-docs.txt
└── scripts/
└── update-docs.sh
Maintenance
| Frequency | Reason |
|---|
| Monthly | DGX docs change slowly — Sunday day 1 of the month via launchd |
| On request | When user asks about a feature missing from the snapshot |
Manual refresh:
"${CLAUDE_SKILL_DIR}/scripts/update-docs.sh"
Automated refresh: ~/Library/LaunchAgents/com.flight505.spark-docs-refresh.plist
Version
Skill Version: 1.0.0
Documentation Sources:
Remember: Tier 1 first → Tier 2 to find topics → Tier 3 for deep details. For forum/blog/adjacent-product questions, hand off to nvidia-dgx-research.