I4h Workflow

NVIDIAOfficialCreated by Isaac for Healthcare TeamDistributed by NVIDIA

Skill
Meta (authoring, orchestration, planning, memory)
Automation and integrations
Apache-2.0
Medium risk

From the author

Orient users to the i4h workflow runtime and route them to the correct stage skill. Use for architecture, support, or where-to-start questions; do not execute a known stage.

— the description by Isaac for Healthcare Team

AutoGineer summary

Written by AutoGineer from the skill’s own files, not by the author

i4h-workflow serves as the primary entry point for users interacting with the i4h robotics framework. It provides architectural orientation by summarizing key design principles and dynamically discovers available workflows and modes from the local repository. Rather than executing tasks itself, it acts as a navigation tool that routes users to the appropriate specialized skill based on their specific goal, such as environment setup, dataset management, training, or validation.

What it does

Resolves and verifies the local i4h repository checkout.
Summarizes core architectural concepts like Scenes, Tasks, Workflows, and the Engine.
Discovers and lists available workflows and run modes using live repository data.
Routes users to the correct specialized skill for tasks like dataset recording, RL training, or validation.
Provides a standardized interface for stopping all active processes.

When to use it

When you are unsure where to start with an i4h robotics project.
When you need an architectural overview of how Scenes, Tasks, and Workflows interact.
When you need to identify which specialized skill to use for a specific development task.
When you need to verify the current status of workflows or modes in your local environment.

Advantages

Ensures users follow the correct architectural patterns by providing context-aware guidance.
Prevents manual errors by routing users to the narrowest, most appropriate tool for their task.
Maintains consistency by relying on live repository discovery rather than static documentation.
Simplifies the onboarding process for new users of the i4h framework.

Usage examples

“What does the i4h workflow include, and where should I start?”: The agent inspects the local repository, summarizes the architecture (Scenes, Tasks, Workflows), lists available modes, and recommends the next skill (e.g., i4h-workflow-setup).
“I need to record demonstrations for a new task. What should I do?”: The agent identifies the request and routes the user to the i4h-workflow-dataset-teleop skill.
“Stop all running i4h processes.”: The agent executes the repository's stop script and reports the number of processes terminated.

Install

Download the skill as a ZIP:
https://mp-staging.autogineer.com/agent-skills/nvidia/i4h-workflow/download

Unzip it into the cross-client skills folder of your project:

mkdir -p .agents/skills && unzip i4h-workflow.zip -d .agents/skills

Documentation by Isaac for Healthcare Team

i4h Workflows

Purpose

Orient the user from live repository facts, then hand execution to the narrowest stage skill.

Instructions

Run the base-checkout resolver.
Read live support and DESIGN.md.
Use only current architecture facts in the answer.
Use the narrowest stage skill for execution.

Resolve the checkout

export I4H_WORKFLOWS_REPO_URL="${I4H_WORKFLOWS_REPO_URL:-https://github.com/isaac-for-healthcare/i4h-workflows}"
I4H_REPO_DIR_NAME="${I4H_WORKFLOWS_REPO_URL%/}"
I4H_REPO_DIR_NAME="${I4H_REPO_DIR_NAME##*/}"
I4H_REPO_DIR_NAME="${I4H_REPO_DIR_NAME##*:}"
I4H_REPO_DIR_NAME="${I4H_REPO_DIR_NAME%.git}"
[ -n "$I4H_REPO_DIR_NAME" ] || { echo "Cannot derive a checkout name from I4H_WORKFLOWS_REPO_URL" >&2; exit 2; }
ROOT="${I4H_WORKFLOWS:-$(git rev-parse --show-toplevel 2>/dev/null)}"
if [ ! -d "$ROOT/workflows/i4h_workflows" ]; then
  ROOT="${I4H_WORKFLOWS:-$HOME/$I4H_REPO_DIR_NAME}"
  [ -d "$ROOT/workflows/i4h_workflows" ] || git clone "$I4H_WORKFLOWS_REPO_URL" "$ROOT"
fi
export I4H_WORKFLOWS="$ROOT"
cd "$ROOT"

Treat this resolver as part of the skill contract: a hosted copy may run outside the base repository, so never assume the current checkout contains workflows/i4h_workflows. I4H_WORKFLOWS_REPO_URL selects the clone source. When I4H_WORKFLOWS is unset, derive the fallback directory from that URL; set I4H_WORKFLOWS only to reuse or choose a specific destination. Never replace an existing checkout.

Inspect before answering

Read ./DESIGN.md for architecture and skills/i4h-workflow/references/repo-map.md for ownership. Discover current support instead of copying a static table:

./run.sh list

If discovery fails because setup is incomplete, report that limitation and route to i4h-workflow-setup.

Explain the design

Keep the summary precise:

A Scene owns the simulated world, assets, embodiment, cameras, randomization, adapters, and reset hooks.
A Task owns one reusable capability. It reads ctx.scene, writes ctx.act, and never advances the simulator.
A Workflow selects one Scene, exposes run-mode-specific TaskGraph builders, and owns goal semantics. A run mode answers how that workflow should run; code and CLI use the shorter term mode.
The Engine schedules graph nodes; the shared SimulationRunner alone resets, steps, renders, records, retries whole episodes, and prints run summaries.
Online RL is a separate training lifecycle: its trainer owns vectorized stepping and returns a checkpoint to the normal policy Task and SimulationRunner validation path.
Simulator-compatible exported RSL-RL actors may run as in-process Tasks; incompatible foundation-model policy stacks remain remote.
Remote policy stacks run out of process and communicate over Zenoh; offline dataset tools remain independent of the simulator.
Python owns behavior. Manifests carry facts across dependency boundaries.

Do not describe retired environment YAMLs, per-mode runners, or separate policy/Arena launchers.

Route the next action

Goal	Skill
Install, sync, or repair dependencies	`i4h-workflow-setup`
Create a new workflow/environment	`i4h-workflow-create`
Edit an existing scene, camera, task, or success rule	`i4h-workflow-scene-edit`
Record demonstrations	`i4h-workflow-dataset-teleop`
Replay HDF5	`i4h-workflow-dataset-replay`
Augment HDF5	`i4h-workflow-dataset-mimic`
Grade/filter HDF5 with a VLM	`i4h-workflow-dataset-annotate`
Convert HDF5 to LeRobot	`i4h-workflow-dataset-convert`
Inspect LeRobot in a browser	`i4h-lerobot-viz`
Fine-tune a manifest-backed policy task	`i4h-workflow-finetune`
RL post-train a supported policy in simulation	`i4h-workflow-train-rl`
Run policy or rule-based rollouts	`i4h-workflow-validate`
Run the maintained complete pipeline	`i4h-workflow-e2e`

For Stop all, do not load a stage skill. Run ./stop.sh all from the repository root and report the stopped process count.

Troubleshooting

If discovery fails, verify the resolved checkout and run setup. If a mode is absent, report it as unsupported.

Prerequisites

Require a readable base checkout or network access to clone it.

Limitations

This router does not install, author, simulate, process data, train, or evaluate.

Examples

What does the i4h workflow include, and where should I start? → inspect live support, summarize DESIGN.md, and recommend one stage skill.

Completion gate

Answer with the live workflow/mode list, a short architecture summary, and one concrete next skill. If the requested workflow or mode is absent from run.sh list, say it is unsupported instead of inventing a command.

Files

ATTRIBUTION.mdAdded by AutoGineer913 B
BENCHMARK.md6.7 KiB
LICENSELicence file added by AutoGineer11.2 KiB
SKILL.md5.1 KiB
evals/evals.json1.4 KiB
references/repo-map.md3.5 KiB
skill-card.md4.4 KiB
skill.oms.sig4.7 KiB

BENCHMARK.md

Skill Benchmark: i4h-workflow

✅ Overall verdict: PASS — Recommended for publication

Publication Recommendation

Recommended for publication based on the completed evaluation evidence in this report.

Evaluation Metadata

Skill: i4h-workflow
Evaluation date: 2026-09-21
Evaluator version: 1.5.6
Agents: Claude Code (aws/anthropic/bedrock-claude-opus-4-8), Codex (openai/openai/gpt-5.5)
Tasks: 2 evaluation tasks (1 positive, 1 negative)
Dataset digest: sha256:00d6d45e6392054493752474bd5ee6e1196f7ad41810fa6da95eb645ed1ac583 (skill-evaluator-dataset-snapshot/1)
Attempts per task: 3
Environment: k8s-sandbox
Tier 2 evidence: required for publication
Tier 3 evidence: required for publication

Each task attempt ran in its own isolated sandbox pod.

What This Report Answers

The three-tier evaluation checks whether the skill:

is safe to use;
produces correct answers;
is discovered and activated when needed;
helps the agent complete the user's goal and expected workflow; and
avoids wasted skill and tool usage.

Results at a Glance

Measure	Claude Code (Baseline → Skill Uplift)	Codex (Baseline → Skill Uplift)
Overall	67.5% — baseline ran, but no comparable score was available; uplift unavailable	72.1% — baseline ran, but no comparable score was available; uplift unavailable
Security	100.0% → 87.5% (-12.5 points)	100.0% → 87.5% (-12.5 points)
Correctness	6.7% → 25.0% (+18.3 points)	50.0% → 50.0% (±0.0 points)
Discoverability	95.0% — baseline ran, but no comparable score was available; uplift unavailable	90.0% — baseline ran, but no comparable score was available; uplift unavailable
Effectiveness	10.8% → 30.6% (+19.8 points)	26.3% → 41.7% (+15.4 points)
Efficiency	99.1% — baseline ran, but no comparable score was available; uplift unavailable	91.3% — baseline ran, but no comparable score was available; uplift unavailable

How to read this table: baseline is the same task attempted without the target skill. Scores are rounded to one decimal; threshold-adjacent values use additional precision so their displayed band matches the verdict. Uplift is derived from those displayed scores and shown in percentage points.

Example: 47.0% → 92.0% (+45.0 points) means the skill-assisted run scored 92.0%, 45.0 percentage points above its 47.0% no-skill baseline.

A partial dimension was calculated from only the available configured signals; review the detailed report before relying on it.

Token Usage

Actual Tier 3 execution usage is reported for every observed agent/case pair and both conditions.

Agent	Dataset case	With skill	Without skill	Delta	Change	Coverage
claude-code	All cases	395,062	703,477	N/A	N/A	skill 4/4; base 6/6
claude-code	i4h-workflow-readme-quick-overview	307,917	587,492	N/A	N/A	skill 1/1; base 3/3
claude-code	i4h-workflow-readme-quick-stop-all	87,145	115,985	-28,840	-24.87%	skill 3/3; base 3/3
codex	All cases	188,073	149,872	+38,201	+25.49%	skill 4/4; base 4/4
codex	i4h-workflow-readme-quick-overview	148,245	110,184	+38,061	+34.54%	skill 1/1; base 1/1
codex	i4h-workflow-readme-quick-stop-all	39,828	39,688	+140	+0.35%	skill 3/3; base 3/3
ALL AGENTS	Dataset aggregate	583,135	853,349	N/A	N/A	skill 8/8; base 10/10

Prompt tokens include cached reads, so total tokens are prompt + completion (cached is not added twice). The Efficiency score uses (prompt - cached) + completion. N/A means the relevant trajectory counters were not available; coverage is never estimated.

Tier Status

Tier	Purpose	Status	Evidence
Tier 1	Static validation	PASSED WITH OBSERVATIONS	11 validator(s); 1 finding(s)
Tier 2	Semantic deduplication	PASSED	2 validator(s); 0 finding(s)
Tier 3	Live agent evaluation	PASS	2 agent(s); 2 task(s)

Findings and Observations

<details> <summary>Show detailed findings and successful checks</summary>

LOW SECURITY/Unknown (SQP-2): The skill instructs an AI agent to clone a remote Git repository (driven by the environment variable I4H_WORKFLOWS_REPO_ (SKILL.md:41)

</details>

Scoring Methodology

<details> <summary>Show dimension definitions, source signals, and thresholds</summary>

Dimension	Question	Scored signals
Security	Is it safe to use?	`security` (100%)
Correctness	Is the answer correct?	`accuracy` (100%)
Discoverability	Was the right skill loaded when needed?	`skill_execution` (100%)
Effectiveness	Did the skill help complete the task?	`goal_accuracy` (50%) + `behavior_check` (50%)
Efficiency	Did it avoid wasted tool calls and token usage?	`skill_efficiency` (50%) + `token_efficiency` (50%)

Dimension bands: PASS at 50% or above; NEUTRAL from 40% to below 50%; FAIL below 40%.
Overall Tier 3 lift: PASS at +5 points or more; FAIL at -10 points or less; values between those bands are NEUTRAL.
Overall verdict: PASS only when every configured dimension passes for at least one supported agent. Lift is reported as diagnostic evidence and does not override this gate.
The 50% attempt pass threshold is a separate per-task gate; it is not the dimension pass threshold.
Effectiveness is the equal-weight mean of goal completion (goal_accuracy) and expected workflow adherence (behavior_check).
Efficiency is 50% tool-call productivity (the backward-compatible skill_efficiency wire id) and 50% token_efficiency. Positive-case skill routing is scored under Discoverability, not Efficiency; a negative case without a routing target is N/A. N/A sources are omitted, remaining weights are renormalized, and the dimension is marked partial.

Signals present in this run:

security (Security): unsafe operations, secret leakage, and unauthorized access.
skill_execution (Skill Execution): whether the expected skill was selected, decoys were avoided, and the workflow executed.
skill_efficiency (Tool Productivity): tool-call productivity (legacy wire id; routing is scored under Discoverability).
accuracy (Accuracy): final-answer correctness against the reference answer.
goal_accuracy (Goal Accuracy): whether the user's goal was achieved.
behavior_check (Behavior Check): whether the expected workflow behavior was followed.
token_efficiency (Token Efficiency): actual uncached prompt plus completion usage (50% of Efficiency).

</details>

Freshness

Regenerate this benchmark when the skill, evaluation dataset, target agent/model, evaluator version, environment, or scoring policy changes.

Requirements and compatibility

Works with: Claude Code, Claude apps, OpenAI Codex, GitHub Copilot, Cursor, Gemini CLI, Kiro, OpenCode, Goose, Amp, Roo Code, JetBrains Junie, Factory Droid, Devin (ex-Windsurf), Mistral Vibe, OpenHands, Letta

“Works with” means the tool reads the open Agent Skills format; a skill written for one tool may use features others lack.

Security

What our scanner noticed· Medium risk

Large base64-encoded blobs

A static scan is not a guarantee. Skills are third-party content: review a skill before you let an agent use it.

Attribution and licence

Created by

Isaac for Healthcare Team

Publisher

NVIDIA (Official)

Distributed by

NVIDIA

Source

NVIDIA/skills · skills/i4h-workflow · commit fce3cca

Licence

Apache License 2.0 (redistribution allowed)

LICENSE (added by AutoGineer from LICENSE-APACHE)
ATTRIBUTION.md (written by AutoGineer)

Synced

2026-11-08

AutoGineer is not affiliated with or endorsed by Isaac for Healthcare Team or NVIDIA.

Download ZIP

The ZIP holds the skill’s files, its licence files and an ATTRIBUTION.md. Download i4h-workflow.zip