Catheter Navigation E2E Smoke Test
Runs the Isaac for Healthcare v0.7 catheter workflow from setup verification through digital-twin DRR rendering and unit tests.
The files show Apache-2.0 licensing, named authorship, and scoped run, log, and cache paths. However, the skill may clone remote code automatically and does not specify user confirmation, revision pinning, CT/patient-data handling, privilege isolation, rollback, or dependency-integrity checks, so points are deducted.
The ordered stages, shared RUN_DIR/CACHE, no-SUBJ partial path, and fallback behavior are described, and BENCHMARK reports external evaluation. However, referenced stage skills are not included here, key commands are not fully specified, variable persistence assumes tmux, and abnormal-exit or logging failures lack sufficient diagnostics; static evidence is limited.
The trigger, audience, workflow boundaries, viewport exclusion, GPU and SUBJ prerequisites, and partial-run behavior are relatively clear. Input validation, environment/version constraints, and non-fit ranges remain incomplete; Chinese support is not evidenced, and first-use cloning depends on GitHub, which may affect mainland-China reachability.
The document includes purpose, stage mapping, run steps, outputs, prerequisites, limitations, troubleshooting, version, and Apache-2.0 metadata, with an identifiable owner. However, BENCHMARK reports that the root version field is ignored, recommended Instructions and Examples sections are missing, and changelog, maintenance ownership, release path, and stable parameter contracts are not clearly documented.
The design plausibly connects setup, digital-twin generation, DRR rendering, and CPU tests while producing shared logs, a summary, and an image. The external report indicates some correctness and effectiveness uplift, but no independently reproducible outputs are included; the synthetic-DRR command is unspecified and referenced stage instructions are absent, so static calibration limits the score.
BENCHMARK supplies four evaluation tasks, external metrics, and static-validation findings, providing limited audit evidence. It includes no committed CI workflow, execution logs, or key-path test suite for this skill, and the reported results cannot be independently reproduced from the supplied files, so points are deducted.
- Confirm authorization before automatically cloning the workflow from GitHub, and pin or review the remote code and dependency versions.
- SUBJ may contain patient CT data; define de-identification, storage, logging, cache access controls, and safeguards against exposing sensitive data.
- Provide and validate the setup, digital-twin, render-drr, and smoke stage instructions; key commands, input validation, and failure diagnostics are currently incomplete.
- Validate GitHub reachability and GPU/container dependencies in mainland-China or GPU-constrained environments, and document offline or alternative paths.
What does this skill do, and when should you use it?
This skill runs an end-to-end smoke pipeline for the Isaac for Healthcare catheter-navigation workflow. It verifies setup, builds a digital twin from a TotalSegmentator subject, renders a cached DRR, and runs CPU unit tests. Without SUBJ, it skips digital-twin construction and produces a synthetic DRR instead. It does not launch the interactive viewport or cover reinforcement-learning and imitation-learning training loops.
Finds or clones the i4h-workflows repository, initializes a timestamped RUN_DIR and log directory, verifies the catheter-navigation setup, builds a digital twin when SUBJ is supplied, renders a cached DRR or creates drr_synthetic.png, runs CPU unit tests, and writes SUMMARY.txt with stage status, paths, and artifacts.
- A developer validating an Isaac for Healthcare catheter-navigation environment can run the setup and CPU smoke checks.
- A medical-AI engineer with an extracted TotalSegmentator subject can validate the full v0.7 digital-twin and cached-DRR workflow.
- A team preparing a non-interactive catheter-navigation demonstration can produce logs, a summary, and a DRR artifact.
- A user without CT data can check the partial workflow using synthetic DRR rendering.
What are this skill's strengths and limitations?
- Chains setup verification, digital-twin construction, DRR rendering, and unit testing in one workflow.
- Provides a synthetic-DRR fallback when CT data is unavailable.
- Preserves per-stage logs, a run summary, and rendered artifacts.
- Has an explicit 0.7.0 version and identifies the Isaac for Healthcare Team as author.
- The full workflow requires a TotalSegmentator subject and a GPU; without SUBJ it is only partial.
- It does not launch the display-dependent interactive viewport.
- It does not cover reinforcement-learning or imitation-learning training loops.
- The supplied SKILL.md does not specify a tested operating-system, driver, or hardware-version matrix.
How do you install this skill?
Install it through the repository's documented skills CLI flow: npx skills add nvidia/skills --skill i4h-catheter-navigation-e2e --yes. The README does not document additional skill-specific installation steps. At runtime, the skill may clone https://github.com/isaac-for-healthcare/i4h-workflows if the expected workflow directory is absent.
How do you use this skill?
Ask the agent to “run the full catheter workflow smoke” or “demo the v0.7 pipeline.” Set I4H_WORKFLOWS if the workflow repository is located elsewhere. For the full end-to-end path, set SUBJ to an extracted TotalSegmentator subject. If SUBJ is unset, the skill skips digital-twin construction and runs a synthetic DRR path. Review RUN_DIR/logs/SUMMARY.txt, stage logs, and the DRR output afterward.