What does this skill do, and when should you use it?
This skill prepares and guides setup of the i4h-workflows agentic workflow. It checks Linux, uv, git, the NVIDIA driver/GPU, and available disk space before running the idempotent setup.sh entry point. It also documents component-level retries, log inspection, and environment registration and dry-run checks. Cosmos setup is separate and is only invoked when explicitly requested.
Resolves an existing I4H_WORKFLOWS or Git-root checkout; clones the base repository to ~/i4h-workflows when needed; creates a timestamped run directory with logs; runs workflows/agentic/setup.sh with output redirected to setup.log; supports retries through third_party/setup.sh or policy/gr00t_n16/setup.sh; and verifies environments with policy/run.sh and arena/run.sh using --list-envs and --dry-run.
- A Linux developer installing or bootstrapping the i4h-workflows agentic workflow for the first time.
- A user diagnosing a missing .venv, third-party checkout, or engine error.
- A maintainer reusing a trusted existing i4h-workflows checkout instead of cloning another copy.
- A developer retrying one failed third-party, arena, or policy component after an interrupted setup.
- A user who needs setup logs and basic environment smoke checks after installation.
How do you install this skill?
- The default flow may clone and execute external code without explicit confirmation and does not pin the external repository revision; review the repository, scripts, and network permissions before execution.
- The skill card's API-key requirement conflicts with SKILL.md; clarify whether credentials are actually needed and how they are used.
- The workflow depends on Linux, an NVIDIA GPU, and GitHub reachability; cloning or dependency downloads may fail on mainland-China networks.
- Verification is limited to --list-envs and --dry-run and does not establish a successful full rollout; logs still require manual review.
- Shell / CLI
- Network access
- Local filesystem
LinuxuvgitNVIDIA driver/GPUDocker only for Cosmos or local VLM containers
Install the skill from NVIDIA/skills with the skills CLI: npx skills add nvidia/skills --skill i4h-workflow-setup --yes. The README does not specify the exact installation directory; the CLI handles destination selection.
tmp="$(mktemp -d)"
git clone --depth 1 https://github.com/NVIDIA/skills.git "$tmp"
mkdir -p ~/.claude/skills
cp -R "$tmp/skills/i4h-workflow-setup" ~/.claude/skills/
rm -rf "$tmp"Generated from the source repository and skill path; it copies only this skill's folder. If the author's install steps above differ, follow those first. To scope it to one project, replace ~/.claude/skills with that project's .claude/skills.
How do you use this skill?
Once installed, send your agent any of these to trigger it:
- set up, install, or bootstrap the agentic workflow,
Ask a skill-enabled agent to “set up, install, or bootstrap the agentic workflow,” or report a missing .venv, third-party checkout, or engine error. The documented flow first runs command -v uv, command -v git, nvidia-smi, and df -h ., then resolves the repository, creates the workflows/agentic/runs run and log directories, runs setup.sh, and checks environments with --list-envs and --dry-run.
What are this skill's strengths and limitations?
- The setup entry point is idempotent, so the workflow can be retried safely.
- The procedure covers repository resolution, run-directory creation, logging, and basic verification.
- It clearly separates ordinary agentic workflow setup from Cosmos setup.
- It documents common missing-prerequisite errors and the required run-directory order.
- Linux and an NVIDIA driver/GPU are required.
- uv, git, and sufficient disk space for third-party checkouts are required.
- The verification commands confirm environment registration and build, not a successful full rollout.
- The source provides no evidence of support for Windows, macOS, or GPU-less hosts.
How does this skill compare with similar options?
Side by side with related skills; every score comes from the same FSRS standard.
| Skill | FS score | Stars | Last updated | License |
|---|---|---|---|---|
| I4H Agentic Workflow Setup this page ✓ NVIDIA · Official | 36 · Not recommended | ★ 3.6k | 13d ago | Apache-2.0 |
| GEO-SEO Skill Updater | 46 · Use with care | ★ 11k | 22d ago | MIT |
| Vercel Preview Deploy ✓ OpenAI · Official | 40 · Not recommended | ★ 28k | 4mo ago | — |
| Cua Driver GUI Automation Skill | 61 · Recommended | ★ 30k | 13d ago | MIT |
| Dependency Refresh Playbook | 59 · Recommended | ★ 4.5k | 23d ago | MIT |
How did FollowSkills review this skill?
The skill specifies preflight checks and log capture and warns against exposing secrets; however, it can clone from GitHub and execute multiple external setup.sh scripts without an explicit user confirmation, writes under the user's home directory, does not pin the cloned revision, and provides no complete data-flow or rollback procedure. The skill card also claims an API key is required, conflicting with SKILL.md, so points are deducted.
The ordered steps, prerequisites, log paths, and troubleshooting guidance are reasonably coherent, and the scripts are described as idempotent. However, there are no reproducible tests for this skill's key paths, it depends on external repository scripts and a tmux-session persistence assumption, and verification only covers environment registration and dry runs. The score is therefore limited under static calibration.
The audience, trigger phrases, Linux/NVIDIA GPU requirements, and separate Cosmos scope are fairly clear. Boundaries are incomplete, Chinese-language support is not addressed, and the core GitHub clone dependency may be unreachable or unreliable for mainland-China users, so points are deducted.
The material includes a name, version 0.6.0, Apache-2.0 license, author, use case, dependencies, limitations, troubleshooting, and final-response requirements. The benchmark reports missing recommended Instructions and Examples sections; no changelog, explicit update path, or sufficiently clear maintenance responsibility is provided, and credential requirements are inconsistent. Points are deducted accordingly.
The skill provides a directly usable sequence for host checks, setup.sh execution, verification, and status reporting, and the committed evaluation report shows one positive task passing. That report is not independently verifiable from the supplied files, and outcomes depend on an external repository and hardware, so manual environmental judgment remains necessary and the static score is conservative.
The commands, expected failures, and log locations provide some auditability, and a committed evaluation report exists. However, there is no real CI and committed test suite covering this skill's key paths; the evaluation has only one task and cannot be independently reproduced from the supplied material. Only limited credit is warranted.
Open a dimension to read why it scored that way
Evidence confidence：Low — Mostly static review, author material or a limited demo; useful for discovery, not high-risk decisions.
See the full review method →