aicr-creating-guided-demos

par nvidia

Génère un script de démonstration guidée interactive (demos/*.sh), en direct ou à son rythme, avec le modèle Frame → Tell → Show → Close. Se déclenche sur « script de démonstration », « guidé…

npx skills add https://github.com/nvidia/aicr --skill aicr-creating-guided-demos

Creating Guided Demos

Overview

A guided demo is a narrative, not a command dump — and it serves two readers equally: a presenter driving it live for a room, and a lone user stepping through it to learn or reproduce the flow. Either way, at every step they know why they're here, what's about to happen, and what just happened.

Four beats: Frame → Tell → Show → Close. The script is the guide — it paces itself, explains each step, and streams the real tool output so what's watched is the actual thing working.

When to use

  • Building a demos/*.sh to present a CLI/workflow live or to let a user self-walk it (onboarding, learning, reproducing a result).
  • Turning a bare sequence of commands into a self-paced, narrated walkthrough.
  • Not for: non-interactive CI scripts, one-off scratch scripts, or pure reference docs.

The four beats

  1. Frame — the why, for whoever's following (a room, or one reader). One or two sentences: what's missing/broken and why it matters. No commands.
  2. Tell — name the steps before running them, so they have a map. Flag what's slow, what needs the network/VPN, what to watch for.
  3. Show — run the real commands one at a time, pausing before each, streaming their full output. Pre-stage the slow/fragile parts; keep a fallback for anything that can fail.
  4. Close — one line on what just happened, plus pointers: docs/ADR, related demos, the source of truth.

Mechanics

Copy skeleton.sh. It gives you:

  • banner / note / pause / run helpers — numbered sections, dim explanations, an Enter-gated pause per step, and a run that echoes then streams output verbatim. Plus an optional run_expect_fail for steps where a non-zero exit is the expected result (uses || rc=$? so set -euo pipefail doesn't abort).
  • One env config block at the top — every input is an env var; required ones fail fast (validate producer-only inputs inside the producer branch so the skip flag still works), the rest default. Each runner overrides only what differs.
  • Re-runnabilityDEMO_NO_PAUSE=1 runs unattended; a skip flag (e.g. SKIP_SETUP=1) reuses pre-staged state. Skip contract: guard the producer, assert the artifact exists (fail loudly), then run the fast steps either way.

Rules

  • Show real output. Never >/dev/null, never fabricate. Watching the real tool is the point.
  • Don't fake success. Never weaken a gate/threshold to force green — an honest failure is a valid beat.
  • Respect time. Anything multi-minute (deploys, benchmarks, sign-in) is pre-staged; live steps are seconds.
  • Commit-ready by default. No home paths, kubeconfig paths, or private refs; generic defaults. (A throwaway personal demo can relax this.)

Worked example

demos/evidence-demo.sh — the split-leg recipe-evidence walkthrough: Frame (the reachability gap) → Tell (validate → publish → verify) → Show (each leg, streamed, pre-staged with fallbacks) → Close (ADR-007 + related demos). Reads the same whether presented live or run solo.

Common mistakes

  • Command dump, no Frame/Close → the runner never learns why, or what they saw.
  • Slow producer step run live → blows the time budget; pre-stage it and add a skip flag.
  • Hardcoded paths/criteria → not reusable; use env vars and fail fast on required ones.
  • Faked/suppressed output or a weakened gate → destroys the trust the demo exists to build.

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,…