Doca Flow

Use this skill for DOCA Flow on a supported NVIDIA NIC/DPU — both for DOCA Flow application knowledge and for writing Flow applications: defining match/action pipes, bringing up ports / representors, forwarding with FWD_PORT (or another fwd target), reading flow counters, validating a pipe before hardware programming, picking the right Flow version against an installed DOCA, or debugging DOCA_ERROR_* from the Flow API. Trigger even when the user does not explicitly mention "DOCA Flow" or "pipe" — typical implicit phrasings include "packets aren't reaching my representor", "rule isn't matching on the BF", "PMD reports init failed", "match/action drop on egress", "how do I steer this 5-tuple to a queue", "ConnectX hairpin routing on Linux", or any flow-steering / packet-classifier question where DOCA is installed. Refuse and route elsewhere for non-Flow pipelines (DPDK rte_flow without DOCA, kernel TC offload, OVS), BFB bring-up, or DPU OS install — those belong to other skills.

mkdir -p ~/.claude/skills/doca-flow
curl -L https://claudskills.com/skills/doca-flow/SKILL.md \
  -o ~/.claude/skills/doca-flow/SKILL.md

@misc{doca-flow-2026,
  author    = {NVIDIA},
  title     = {Doca Flow [Claude Code skill]},
  year      = {2026},
  publisher = {ClaudSkills},
  url       = {https://claudskills.com/skills/doca-flow/}
}