Fleet Intelligence Agent Skills

View as Markdown

Fleet Intelligence ships three portable Agent Skills for harnesses such as Claude Code, Cursor, and Codex:

SkillUse it for
nvfleetintAd hoc fleet questions and CLI authentication
fleet-health-reportA fleet-wide, standalone HTML health snapshot
node-rca-rccaAn evidence-backed HTML RCA/RCCA for one node

Each skill uses the shared SKILL.md format. Harness-specific installation paths and optional UI metadata do not change the core workflow.

Install

Install all skills for Claude Code, Cursor, and Codex:

$npx skills add NVIDIA/fleet-intelligence-client \
>  --skill '*' \
>  --agent claude-code \
>  --agent cursor \
>  --agent codex

Add --global to install them for the current user instead of the current project. Select individual skills with repeated --skill <name> flags.

Install and authenticate nvfleetint before using any skill. Start a new agent session if an installed or updated skill is not discovered immediately.

Report skills

The two report skills are reference implementations with strict evidence and completeness rules. Adapt their report sections and presentation to the task, but preserve the live-data, read-only, no-secrets, and evidence-versus-inference requirements.

Both produce standalone, print-friendly HTML documents using their bundled references/html-report-template.md guidance:

  • fleet-health-report covers fleet distribution, operational signals, trends, issue concentration, and machines needing attention.
  • node-rca-rcca covers one node’s timeline, impact, root cause, contributing factors, corrective actions, and validation plan.

Use nvfleetint to explore ad hoc questions before or after generating either report.