
Description
Use first for general NVIDIA NeMo Agent Toolkit coding-agent behavior, task routing, naming conventions, component discovery rules, and cross-skill guidance.
SKILL.md
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. - 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.
More skills from the NeMo-Agent-Toolkit repository
View all 11 skillsnat-agent-configuration
configure and compose NeMo Agent Toolkit agents
Jul 14AgentsNVIDIAOrchestrationnat-evaluation
run evaluations for NeMo Agent Toolkit
Jul 14DatasetsEvalsNVIDIAQuality Assurancenat-installation
install and configure NeMo Agent Toolkit
Jul 14AgentsDeploymentNVIDIAnat-mcp-and-serving
serve and configure NeMo agent workflows
Jul 14AgentsFastAPIMCPNVIDIAnat-optimization
optimize agent workflows with NeMo
Jul 14AgentsNVIDIAOptimizationPerformancenat-path-checks
fix NeMo Agent Toolkit documentation path errors
Jul 14DebuggingDocumentationNVIDIA
More from NVIDIA
View publishernemoclaw-user-guide
retrieve NemoClaw documentation and configuration
NemoClaw
Jul 20DocumentationMCPSearchmcore-build-and-dependency
manage Megatron-LM development environments
Megatron-LM
Jul 14ContainersDeploymentPythonmcore-bump-base-image
update NVIDIA PyTorch base images
Megatron-LM
Jul 14CI/CDDeploymentmcore-cicd
manage CI/CD pipelines for Megatron-LM
Megatron-LM
Jul 14CI/CDDeploymentGitHubmcore-create-issue
investigate CI failures and create issues
Megatron-LM
Jul 14DebuggingGitHubTriagemcore-linting-and-formatting
lint and format Megatron-LM code
Megatron-LM
Jul 14Best PracticesCode Analysis