i4h Agentic Workflow Guide
Navigate supported medical-robotics environments, policies, and workflow stages.
The skill clearly states that it only routes and orients, does not run pipeline stages, and does not request credentials or handle sensitive data. However, it may clone into $HOME/i4h-workflows without confirmation and depends on an external GitHub repository, with no least-privilege, data-flow, rollback, or pre-action confirmation guidance, so points are deducted.
Commands, fallback environment table, troubleshooting, and scope limits are reasonably clear, and BENCHMARK records external evaluation results. Static review did not execute the key command; behavior depends on repository, Git, and environment state, while abnormal-input coverage and failure diagnostics are limited, so the score remains conservative.
The audience, trigger questions, supported environments, and next-stage routing are clear, and the skill explicitly limits itself to overview. It does not define detailed non-fit cases, does not document Chinese-language support, and core initialization may depend on GitHub reachability that can be unreliable for mainland-China users, so points are deducted.
The skill provides a name, version 0.6.0, Apache-2.0 license, author, tags, references, limitations, skill index, and troubleshooting. It lacks separate Instructions and Examples sections, a changelog, and a clear maintenance/update path; repository license metadata is also NOASSERTION, so full marks are not justified.
It can provide orientation on environments, subprojects, skill routing, and setup recommendations; BENCHMARK reports 83% Codex correctness and 61% effectiveness. However, it is navigation-only, its supported-environment table may be a static fallback, its required output is rigid, and this file does not verify directly usable results for key paths, so points are deducted.
The skill content, version, and evaluation tasks are traceable, and BENCHMARK provides limited external metrics. But the evaluation has only two tasks with one attempt each and lacks committed tests or independently reproducible evidence covering the skill's key command, so the static score is conservative.
- Do not run the default clone command without confirming the workspace and network permissions; verify I4H_WORKFLOWS, repository provenance, and GitHub reachability first.
- The static supported-environment table may become stale; use the target repository's actual YAML files and --list-envs output before acting.
- This skill only provides overview and routing; it should not be treated as an execution guide for recording, training, or deployment.
- Add Chinese-language guidance, a mainland-China or offline access path, examples, a changelog, and a clear maintenance/update path.
What does this skill do, and when should you use it?
i4h-workflow is an orientation and routing skill for the Isaac for Healthcare agentic workflow. It explains supported environments, robots, policies, and the roles of the `workflows/agentic/` subprojects. The skill covers navigation across simulation, data, training, and validation stages, but does not itself record demonstrations, train policies, or run rollouts. Hands-on work requires an available i4h-workflows checkout and workflow setup.
Resolves or locates the i4h-workflows base code; checks for workflows/agentic/; runs workflows/agentic/policy/run.sh --list-envs to enumerate environments; and uses environment YAMLs, subproject descriptions, and the Skill Index to route users to the next stage skill. If the base code is unavailable, it identifies ~/i4h-workflows as the default clone location.
- A medical-robotics researcher needs to check whether a target environment, robot, and policy are supported.
- An engineer is starting a simulation-to-policy workflow and needs to identify the relevant arena, policy, dataset, mimic, or annotator project.
- A user is unsure whether to begin with setup, data collection, conversion, fine-tuning, or validation.
- An environment-listing command fails and the user needs a fallback support list and setup troubleshooting path.
What are this skill's strengths and limitations?
- Provides a single overview of supported environments, robots, policies, and subprojects.
- Indexes stage skills spanning data collection, processing, training, and validation.
- Includes a fallback environment table when environment discovery fails.
- Clearly limits itself to orientation and routing.
- Does not record demonstrations, train policies, or execute rollouts itself.
- Hands-on stages require separate setup and stage skills.
- Requires an existing or clonable i4h-workflows checkout; detailed platform and hardware requirements are not provided in the supplied material.
How do you install this skill?
Install the skill from NVIDIA's catalog with: npx skills add nvidia/skills --skill i4h-workflow --yes. To target Codex specifically, use: npx skills add nvidia/skills --skill i4h-workflow --agent codex --yes. No other installation method is documented in the supplied material.
How do you use this skill?
After installation, prompt the agent with something like: What is the i4h workflow, what environments are supported, and where should I start? For hands-on work, prepare an i4h-workflows checkout and complete i4h-workflow-setup; the skill then uses workflows/agentic/policy/run.sh --list-envs and routes you to a stage skill matching your goal.