Build Applications on NVIDIA BlueField Faster with NVIDIA DOCA Agent Skills
Source
Tanya Lenz
Author
Tanya Lenz
Date
Key takeaways · AI-distilled
Each DOCA agent skillA reusable instruction file that teaches an agent how to do one job well — the procedure, the tools, and what counts as done.Full definition → is a SKILL.md scoped to one component, holding real function signatures, pkg-config module names, build-container constraints and known failure modes. NVIDIA frames it as a machine-readable spec, not a docs summary.
Without skills, the most common failures in NVIDIA's 65-prompt evalA repeatable test for AI quality — a set of tasks plus scoring — used the way software teams use test suites, because model output is too variable to judge by eyeballing.Full definition → were API and flag misuse (59 prompts), unverified hardware capability (46), wrong tool routing (39), skipped smoke tests (34) and guessed versions (30).
For a firmware-level mlxconfig change on a production BlueField-3, NVIDIA says the with-skills AI agentAn AI system that doesn't just answer once but works toward a goal in a loop — taking actions, reading the results, and deciding what to do next.Full definition → covered preflight inventory, an out-of-band path, a maintenance window and rollback, and noted the write needs a cold power cycle, not a warm reboot.
In NVIDIA's side-by-side demo building a Go RDMA app on BlueField-3, both agents succeeded, but the one with skills wrote 189 lines of handwritten code versus 695 and ran 20 hardware commands versus 37.
Terms in this piece · Glossary
AI agent — An AI system that doesn't just answer once but works toward a goal in a loop — taking actions, reading the results, and deciding what to do next.
eval — A repeatable test for AI quality — a set of tasks plus scoring — used the way software teams use test suites, because model output is too variable to judge by eyeballing.
agent skill — A reusable instruction file that teaches an agent how to do one job well — the procedure, the tools, and what counts as done.
Why it matters
Shows that packaging verified API signatures, hardware constraints and preflight checks as agent skills lifted checklist pass rate from 19% to 100% on specialized infrastructure code.
articleAI Native by Design: Lessons Learned from Building NVIDIA TensorRT Model Connect
articleNVIDIA Open Agent Safety Platform: A Reference for Continuous In-Silicon Agent Monitoring
articleAccelerating a ROS 2 Node with an AI Agent and NVIDIA Isaac ROS
articleSimplifying Model Serving Across Multiple GPUs with NVIDIA TensorRT Multi-Device Integration in NVIDIA Dynamo-Triton
