i4h-workflow-setup

par nvidia

Verify host requirements and run `workflows/agentic/setup.sh`. Use when asked to set up, install, or bootstrap the agentic workflow, or hits missing `.venv`,…

npx skills add https://github.com/nvidia/skills --skill i4h-workflow-setup

i4h Workflow — Setup

Purpose

Verify host requirements and run the idempotent workflows/agentic/setup.sh to bootstrap the agentic workflow. Use when asked to set up, install, or bootstrap the workflow, or when hitting a missing .venv, third-party checkout, or engine error.

Base Code

These steps drive the i4h-workflows base code (the workflows/agentic/ tree). To reuse an existing checkout, set I4H_WORKFLOWS to its path (no clone happens). Otherwise this resolves the current repo, or clones to ~/i4h-workflows — pick that default without prompting. Run every command below from the resolved root:

# Resolve the i4h-workflows base code (provides workflows/agentic/).
ROOT="${I4H_WORKFLOWS:-$(git rev-parse --show-toplevel 2>/dev/null)}"
if [ ! -d "$ROOT/workflows/agentic" ]; then
  ROOT="${I4H_WORKFLOWS:-$HOME/i4h-workflows}"
  [ -d "$ROOT/workflows/agentic" ] || git clone https://github.com/isaac-for-healthcare/i4h-workflows "$ROOT"
fi
export I4H_WORKFLOWS="$ROOT"; cd "$ROOT"

Basics

  • workflows/agentic/setup.sh is the idempotent setup entry point.
  • Cosmos setup is separate; invoke only when the user asks for Cosmos or video transfer.

Preflight

command -v uv
command -v git
nvidia-smi
df -h .

Required: Linux, uv, git, NVIDIA driver/GPU, disk space for third-party checkouts. Docker is optional unless using Cosmos or local VLM containers.

Run the steps below in order. Each step is a separate bash call; variables persist in the local agent's tmux session.

Step 1 — resolve repo and run dir

REPO_ROOT="${I4H_WORKFLOWS:-$(git rev-parse --show-toplevel 2>/dev/null)}"; [ -d "$REPO_ROOT/workflows/agentic" ] || REPO_ROOT="$HOME/i4h-workflows"
RUN_DIR="${REPO_ROOT}/workflows/agentic/runs/setup_$(date +%Y%m%d_%H%M%S)"
mkdir -p "${RUN_DIR}/logs"
ln -sfn "${RUN_DIR}" "${REPO_ROOT}/workflows/agentic/runs/.latest"

Step 2 — run setup

On a freshly cloned repo the harness may ask to approve setup.sh once. Set I4H_WORKFLOWS to a pre-existing trusted clone to skip the fresh-clone gate.

"${REPO_ROOT}/workflows/agentic/setup.sh" > "${RUN_DIR}/logs/setup.log" 2>&1

Watch progress with tail -f "${RUN_DIR}/logs/setup.log". For component-specific retries (also idempotent):

"${REPO_ROOT}/workflows/agentic/third_party/setup.sh"
"${REPO_ROOT}/workflows/agentic/policy/gr00t_n16/setup.sh"

Verify

"${REPO_ROOT}/workflows/agentic/policy/run.sh" --list-envs
"${REPO_ROOT}/workflows/agentic/arena/run.sh" --list-envs
"${REPO_ROOT}/workflows/agentic/policy/run.sh" --env scissor_pick_and_place --dry-run
"${REPO_ROOT}/workflows/agentic/arena/run.sh" --env scissor_pick_and_place --dry-run

Prerequisites

  • Linux host with uv, git, and an NVIDIA driver/GPU (check via the Preflight commands).
  • Disk space for the third-party checkouts (df -h .).
  • Docker only when using Cosmos or local VLM containers; otherwise optional.

Limitations

  • Linux only; requires an NVIDIA driver/GPU.
  • Cosmos setup is separate — setup.sh does not run it; invoke Cosmos only when the user asks for it.
  • Verify steps use --list-envs and --dry-run; they confirm envs register and build, not that a full rollout works.

Troubleshooting

  • Error: uv / git / nvidia-smi not found — Cause: missing host requirement. Fix: install it and re-run Preflight before setup.sh.
  • Error: a single component fails partway through setup — Cause: a checkout or component venv did not finish. Fix: re-run that component's script (e.g. third_party/setup.sh, arena/setup.sh, policy/setup.sh); all are idempotent.
  • Error: --list-envs or --dry-run fails after setup — Cause: setup incomplete or a component venv missing. Fix: inspect ${RUN_DIR}/logs/setup.log (or workflows/agentic/runs/.latest/logs/setup.log) and re-run setup.sh or the failing component script.
  • Error: setup.sh fails instantly redirecting to a missing log dir — Cause: Step 2 ran before Step 1, so ${RUN_DIR}/logs does not exist. Fix: run Step 1 first, then Step 2.

Final Response

Report setup status, failed component (if any), relevant log path, next recommended smoke test.

Plus de skills de nvidia

compileiq-debug
nvidia
Utilisez quand quelque chose ne va pas : Search() bloque, toutes les évaluations retournent INVALID_SCORE, les scores ne s'améliorent pas, chaque configuration retourne le même nombre, erreurs ptxas…
create-github-pr
nvidia
Créer des pull requests GitHub en utilisant l'interface en ligne de commande gh. Utiliser lorsque l'utilisateur souhaite créer une nouvelle PR, soumettre du code pour révision, ou ouvrir une pull request. Mots-clés de déclenchement -…
nemoclaw-maintainer-cross-issue-sweep
nvidia
Analyse les autres problèmes ouverts pour trouver ceux qu’une PR donnée pourrait également corriger ou casser accidentellement. Génère des opportunités de correctifs adjacents et des risques de contradiction avec fichier:ligne…
fhir-basics
nvidia
Apprend aux agents comment fonctionnent les API FHIR R4, quelles ressources sont disponibles, comment les interroger avec des paramètres de recherche, et comment analyser correctement tous les formats de réponse…
compileiq-validate-result
nvidia
Utiliser APRÈS qu'une recherche soit terminée et AVANT de réclamer un accélérateur ou d'expédier un ACF. Charge le CSV dump_results, extrait les K meilleurs candidats (mono-objectif)…
changelog-audit
nvidia
Auditer le CHANGELOG.md de Warp avant une publication : récupérer les entrées perdues, trier par impact utilisateur, affiner le langage des entrées, ajuster les retours à la ligne et (en mode branche de publication) mettre à jour la comparaison…
maintain-dynamic-plugins
nvidia
Maintenir les chargeurs de plugins dynamiques NeMo Relay, les manifestes, les SDK natifs Rust, le protocole worker gRPC, le SDK worker Python, la documentation, les tests et la couverture du workflow de publication
dgx-diagnose
nvidia
Diagnostiquer les problèmes courants du DGX Station GB300 — plantages CUDA, ciblage incorrect du GPU, bugs de conteneur vLLM/SGLang, problèmes d'état MIG, erreurs NVLink/Fabric Manager,…