NemoClaw AI Agent User Guide
Directs AI coding assistants to authoritative NemoClaw docs for setup, operations, security, and troubleshooting.
The skill explicitly prefers a read-only documentation MCP, requires confirmation before running commands, forbids requesting or pasting secrets, uses redacted placeholders, and cites canonical NVIDIA documentation. Six points are deducted because external data-flow, network trust boundaries, rollback, and recovery procedures are not specified, while the command examples can still cause external effects.
Retrieval order, variant selection, fallback behavior, and troubleshooting routes are internally coherent, and the skill discourages guessing while requiring verification commands. Two points are deducted because it lacks concrete diagnostics for MCP failure, missing pages, URL transformation errors, or restricted networks; static review cannot establish runtime reliability.
The audience, scenarios, three agent variants, and trigger keywords are reasonably explicit, and the skill requests environment choices incrementally. Five points are deducted because non-fit boundaries, exclusion triggers, operating-system coverage, and Chinese-language support are unspecified; the core function depends on docs.nvidia.com without evidence of mainland-China reachability.
The document has readable sections for retrieval, configuration, starting pages, routing, and response requirements; the skill card adds owner, Apache 2.0 licensing, use case, dependencies, risks, references, and a version identifier. Six points are deducted for missing recommended Instructions and Examples sections, inconsistent version/revision signals, limited changelog and maintenance-path detail, and repository license metadata marked NOASSERTION.
The skill can route installation, variant, inference, network-policy, and troubleshooting requests to canonical Markdown pages while requiring citations and verification steps. Ten points are deducted because the benchmark reports N/A for every actual metric, the evaluation dataset is unavailable, and no representative output is verified, so only partial task value is supported statically.
The fixed revision, official documentation endpoints, evaluation cases, and benchmark report provide some auditability. Seven points are deducted because benchmark values are N/A, Tier 3 details and the dataset are unavailable, and no committed CI or test suite reproduces the skill's key paths; the evidence remains largely author-reported.
- The core workflow depends on the external docs.nvidia.com MCP/Markdown service; verify reachability in restricted or mainland-China networks and prepare an offline or manual fallback.
- Do not treat the official-docs claim or the benchmark PASS verdict as proof of runtime correctness; the benchmark has no actual values or reproducible experiment details.
- Before executing installation, configuration, or network-policy commands produced through this skill, review permissions, target environment, credential handling, and rollback steps.
- Add concrete MCP/network failure diagnostics, non-fit boundaries, examples, version change history, and an explicit maintenance/update path.
What does this skill do, and when should you use it?
This skill helps AI agents use NVIDIA’s canonical NemoClaw documentation in Fern Markdown form. It prioritizes the NemoClaw docs MCP server, with an llms.txt index fallback when MCP is unavailable. It routes questions about installation, configuration, inference, network policy, sandbox management, security, and troubleshooting to relevant documentation. It is intended for users deploying and operating NemoClaw with OpenClaw, Hermes, or Deep Agents.
It instructs an agent to configure https://docs.nvidia.com/nemoclaw/_mcp/server and search documentation with the read-only searchDocs tool; without MCP, it reads https://docs.nvidia.com/nemoclaw/llms.txt and then fetches the relevant Markdown page. It routes requests based on the selected agent variant, operating system, and inference provider to prerequisite, quickstart, inference, network-policy, sandbox-management, monitoring, security, or troubleshooting pages. It also requires citations to the Markdown documentation pages used.
- A user needs to install NemoClaw with OpenClaw, Hermes, or Deep Agents.
- A user is troubleshooting local inference, hosted providers, model switching, or tool calling.
- A user needs network-policy approvals or custom egress configuration.
- A user needs sandbox status, logs, rebuilds, upgrades, backup, restore, or messaging-channel guidance.
- A user needs guidance on credential storage, sandbox hardening, or security posture.
What are this skill's strengths and limitations?
- Uses live canonical documentation as the source of truth.
- Covers installation, operations, security, network policy, and troubleshooting routes.
- Explicitly supports OpenClaw, Hermes, and Deep Agents.
- Requires documentation citations and a verification command after setup or recovery steps.
- The SKILL.md does not contain OS-, inference-provider-, or model-specific procedures; actual instructions depend on the online docs.
- The MCP server and NemoClaw documentation URLs require network access.
- No test suite, offline documentation bundle, or concrete credential-configuration implementation is provided.
- The repository metadata lists the license as NOASSERTION, while this skill declares Apache-2.0.
How do you install this skill?
Install the specific skill from NVIDIA’s collection with:
npx skills add nvidia/skills --skill nemoclaw-user-guide --yes
The README does not document a standalone installation directory or another installation method.
How do you use this skill?
First tell the agent which variant you want to use: OpenClaw, Hermes, or Deep Agents. Then make a concrete request such as “Help me install NemoClaw with OpenClaw.” The agent should retrieve the canonical Markdown documentation and tailor steps to your operating system and inference provider. Do not paste API keys, bot tokens, or other secrets into chat.