NVIDIA NeMo Agent Toolkit User Rules
Use this skill first when working in the NeMo Agent Toolkit repository. It routes tasks to focused skills and states rules that apply across all toolkit work.
Mandatory Rules
- Discover registered component
_typevalues withnat info componentsbefore writing workflow, evaluation, optimizer, logging, or tracing YAML. - Do not invent
_typenames or configuration keys from memory. - Use
natonly for technical identifiers such as the CLI, package name, Python namespace, paths, and environment variables. - Treat Git submodule contents as read-only unless the user explicitly requests a change within that submodule. Exclude submodule contents from repository-wide mutation tasks by default.
- In prose, use "NVIDIA NeMo Agent Toolkit" on first use, then "NeMo Agent Toolkit" or "the toolkit".
- Prefer existing examples and docs before creating new patterns.
- Keep generated examples runnable from the repository root unless the surrounding example uses another convention.
Task Routing
| Task | Skill |
|---|---|
| Installing or configuring the toolkit | skills/nat-installation/SKILL.md |
| Creating, editing, validating, or running workflow YAML | skills/nat-workflow-creation/SKILL.md |
| Choosing or composing agents | skills/nat-agent-configuration/SKILL.md |
| Writing custom tools, functions, or function groups | skills/nat-tools-and-functions/SKILL.md |
| Designing or running evaluation | skills/nat-evaluation/SKILL.md |
| Running optimizer workflows | skills/nat-optimization/SKILL.md |
| Adding tracing, logging, profiling, or telemetry exporters | skills/nat-telemetry/SKILL.md |
| Serving workflows or wiring MCP | skills/nat-mcp-and-serving/SKILL.md |
| Fixing documentation path-check failures | skills/nat-path-checks/SKILL.md |
| Creating or improving skills | skills/skill-evolution/SKILL.md |
Discovery Commands
uv run nat info components -t function
uv run nat info components -t llm_provider
uv run nat info components -t evaluator
uv run nat info components -t logging
uv run nat info components -t tracing
Skill Evolution
If a user corrects the skill routing, a command fails and the recovery is reusable, or a reference is stale, finish the user task first. Then read skills/skill-evolution/SKILL.md and update the relevant focused skill if the lesson should generalize.