Generated banner for Holoscan Module Lifecycle
Generated logo for Holoscan Module Lifecycle

Holoscan Module Lifecycle

NVIDIAOfficialCreated by Holoscan TeamDistributed by NVIDIA

  • Skill
  • Workflow
  • Software development
  • Apache-2.0
  • Medium risk

From the author

Develop external Holoscan Modules with the Holoscan CLI and use them in Holoscan applications: scaffold, test, Python bindings, DEB/WHEEL packaging, and clean-consumer proof.
— the description by Holoscan Team

AutoGineer summary

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

This skill provides a structured workflow for developing external Holoscan Modules, which extend the Holoscan SDK with custom operators and resources. It covers the entire lifecycle from scaffolding a new module repository using the Holoscan CLI template to implementing public APIs, running tests, and packaging the module as DEB or WHEEL artifacts. The process emphasizes maintaining consistent identities, validating metadata, and proving consumer compatibility through clean-room installation checks.

What it does

  • Scaffolds new standalone Holoscan Module repositories using the official CLI template.
  • Configures module metadata, identities, and SDK version requirements.
  • Implements and validates public operator APIs and deterministic tests.
  • Builds and tests modules within containerized environments.
  • Packages modules into DEB or WHEEL formats for distribution.
  • Verifies module installability and consumer compatibility in clean environments.

When to use it

  • When creating a new reusable Holoscan Module outside of the core SDK.
  • When adding Python bindings to an existing C++ Holoscan Module.
  • When preparing a module for distribution via DEB or WHEEL packages.
  • When verifying that a consuming application can correctly link against a custom module.

Advantages

  • Ensures consistency across module development through standardized CLI scaffolding.
  • Promotes clean-room testing to verify package installability and dependency resolution.
  • Separates reusable module logic from specific application implementations.
  • Provides a clear path for integrating Python bindings with C++ Holoscan operators.

Usage examples

“Scaffold a new Holoscan module named 'image-processor' with C++ and Python support targeting SDK 5.0.0.”
The agent will initialize a new repository, configure the metadata with the specified SDK floor, and generate the necessary project structure for C++ and Python bindings.
“Package my existing Holoscan module as a DEB file and verify it in a clean environment.”
The agent will build the DEB artifact, simulate an installation in a fresh environment, and perform a smoke test to ensure the module is correctly linked and importable.

Install

  1. Download the skill as a ZIP:

    https://autogineer.com/agent-skills/nvidia/holoscan-module-lifecycle/download
  2. Unzip it into the cross-client skills folder of your project:

    mkdir -p .agents/skills && unzip holoscan-module-lifecycle.zip -d .agents/skills

Documentation by Holoscan Team

Holoscan Module lifecycle

Purpose

Take a reusable Holoscan Module from layout and API contract through producer, consumer, and clean artifact-install proof. Use uv run holoscan in a standalone module, or holoscan in its activated pip environment.

A Holoscan Module extends holoscan::core with operators, resources, and helpers for a domain capability, plus examples and tests. Core modules ship with the SDK; external modules live in their own repositories, and applications, including those in nvidia-holoscan/holoscan-applications, consume their installed packages. This skill covers external Holoscan Module development and consumption. For SDK core-module implementation, follow the selected SDK checkout’s modules/DEVELOPER.md and module-specific guide; do not apply the external CLI template or packaging workflow to SDK core modules.

Inputs

Required: the API contract, workspace, identities, SDK floor, license, maintainer, package formats, and acceptance checks, from the request or an authoritative repository source. Optional: a consuming application. If consequential release metadata remains unknown, stop before scaffolding or packaging rather than guessing.

  • a reusable producer/consumer API contract rather than an ordinary app;

  • the user-provided standalone module repository, or the parent directory for a new one;

  • language, module/operator/package identities, minimum SDK version, license, maintainer details, and requested package formats;

  • the consuming application, if any, such as one in a holoscan-applications checkout;

  • finite producer, consumer, and artifact acceptance checks.

First honor any request for planning only (for example, “do not run commands”); preserve that restriction when routing. Check the session’s available skill catalog for the destination skill. Route a concrete failing or wrong holoscan command to holoscan-app-debug-build-run. If the matching skill is unavailable, preserve the handoff context and name the skill to install instead of broadening this workflow. An ordinary application without a reusable module API is outside this skill; follow the target repository's own guidance, such as README.md and CONTRIBUTING.md in holoscan-applications.

Prerequisites

The module's own README.md, DEVELOPER.md, and AGENTS.md, a consuming checkout's guidance, local help, schemas, templates, tutorials, and nearby examples are the live technical authority where they do not conflict with user, system, or safety constraints.

Instructions

For a planning-only request, return the proposed workflow and proof requirements without executing commands or changing files. Otherwise continue below.

At any step, a failing effect-bearing CLI command ends this happy path; follow Troubleshooting with its exact context. Parse read-only diagnostic results and stop only when a failed capability is required by the module, consumer, packaging environment, or requested proof. For example, a failed Docker capability blocks a container build; an unavailable display need not block an explicitly headless consumer.

  1. Preserve and orient. Record full HEAD, concise status, selected CLI version/environment, and discovery output. Preserve the checkout revision and unrelated work.

  2. Choose the layout. Use a standalone holoscan-<name> repository with its own demo applications. holoscan-applications keeps applications only and directs reusable operators and Modules to independent repositories.

  3. Fix identities before files. Keep display name, snake_case module/namespace, holoscan-<kebab-case> repository/package, and snake_case_op operator slug consistent.

  4. Scaffold with the packaged template. Follow the uv-first creation sequence in the development reference, from a parent directory outside any application checkout. Preview and run the same non-interactive holoscan create command with explicit language (cpp, or cpp python to add Python bindings) and agreed metadata. Always specify --context holoscan_version=<agreed-sdk-floor> to override configured or template defaults; the template requires 5.0.0 or later, and its preview does not check that floor. Examples explicitly select 5.0.0; use the agreed SDK floor for the actual module. No launcher or holoscan-applications checkout is required. Preserve existing Git state; reject populated destinations. Inspect generated files and Git status before continuing.

  5. Implement the smallest public contract. Validate metadata, expose one useful operator/API, add deterministic tests, and add a finite demo that consumes the same public surface as another project. Keep demo policy out of the reusable operator.

  6. Build and test honestly. Preview and run container-first module and demo commands, passing --language cpp or --language python when a project declares both. The module's test runs its repository-wide suite; combine it with focused pytest or CTest targets and report the scope.

  7. Prove a real consumer. An application consumes the installed package with find_package(holoscan-<name> CONFIG REQUIRED) and holoscan::<module_slug>. In holoscan-applications, a metadata dependency alone neither fetches nor builds the module; follow the application consumer procedure and do not change the shared root CMake without maintainer agreement. For Python bindings, import from the exact build tree or install the holoscan-<name>-python component into a task-owned prefix; follow the Python consumer cleanup procedure: record the interpreter, build directory, prefix, and installed paths, always remove the task-owned prefix after the consumer attempt, and verify it is absent. The packaged template stages no editable-import hooks, so do not substitute install --dev. If cleanup fails, stop using that environment and report incomplete cleanup; a missing debugging skill does not permit continuing to package or publish. Discard a task-owned disposable environment only under existing cleanup authorization; a shared environment needs the owner’s repair decision.

  8. Package only after behavior proof. Preview and build the requested artifacts, inspect identity/metadata/dependencies/payload, and install each format in a clean artifact-only consumer with an import/link and finite smoke check. The packaged template produces a C++-only DEB and no wheel; request WHEEL only for a module whose own pyproject.toml builds one. For DEB, verify the declared holoscan Debian dependency in the consumer’s package database: a successful Python import or C++ build is insufficient. If it is absent, report DEB installation unverified and follow the DEB prerequisite guidance; do not force dependency checks or claim archive extraction proves installability.

  9. Validate and hand off. Rerun producer and clean-consumer checks, git diff --check, and final status. In a dirty checkout, restrict auto-fixing lint to task paths; before a requested commit, validate the exact candidate change with the full lint required by that repository's guidance, in a clean disposable checkout. If the required commands cannot be determined, report lint unverified. Confirm no generated artifacts or install prefixes are staged. Do not commit, push, upload, publish, or release unless requested.

Troubleshooting

If a CLI command fails, stop the happy path and hand off its exact command, revision, dirty state, inputs, and observed result to holoscan-app-debug-build-run.

Examples

  • Scaffold and package a reusable operator with clean-consumer proof: use this skill.

  • Package an ordinary app or diagnose a failing build: route out of this skill.

Limitations

  • Preserve unrelated work. Do not reset, clean, remove unrelated files, change host configuration, broaden privileges, commit, push, upload, publish, or release without authorization.

  • Never prefix holoscan with sudo; obtain approval for holoscan setup, host packages, host-local execution, root containers, devices/capabilities, or permission changes.

  • Keep build, data, install prefixes, and package output out of commits.

  • Protect credentials, proprietary source, patient data, private media, and identifying metadata.

  • Claim support, reproducibility, installability, performance, or safety only for the producer, artifacts, and clean-consumer evidence actually collected.

Output

Return layout and identities, public API/demo, metadata and registrations, preview and real command results, honest test scope, consumer proof, artifact identity/contents/install results, licensing and publication limits, and final worktree state.

For a planning-only request, return the proposed order, authoritative metadata sources, approval boundaries, and proof requirements without claiming results.

Files

BENCHMARK.md

Skill Benchmark: holoscan-module-lifecycle

✅ Overall verdict: PASS — Recommended for publication

Publication Recommendation

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

Evaluation Metadata

  • Skill: holoscan-module-lifecycle

  • Evaluation date: 2026-10-20

  • Evaluator version: 1.5.6

  • Agents: Claude Code (aws/anthropic/bedrock-claude-opus-4-8), Codex (openai/openai/gpt-5.5)

  • Tasks: 8 evaluation tasks (7 positive, 1 negative)

  • Dataset digest: sha256:e4d70204e72ae37b649556e34cec480522ea05caa65ad65f6e267f4cbfc30ef5 (skill-evaluator-dataset-snapshot/1)

  • Attempts per task: 1

  • 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

MeasureClaude Code (Baseline → Skill Uplift)Codex (Baseline → Skill Uplift)
Overall95.1% — baseline ran, but no comparable score was available; uplift unavailable82.8% — baseline ran, but no comparable score was available; uplift unavailable
Security87.5% → 100.0% (+12.5 points)100.0% → 100.0% (±0.0 points)
Correctness67.5% → 100.0% (+32.5 points)55.0% → 80.0% (+25.0 points)
Discoverability100.0% — baseline ran, but no comparable score was available; uplift unavailable78.6% — baseline ran, but no comparable score was available; uplift unavailable
Effectiveness48.9% → 82.0% (+33.1 points)51.8% → 68.9% (+17.1 points)
Efficiency93.4% — baseline ran, but no comparable score was available; uplift unavailable86.5% — 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.

AgentDataset caseWith skillWithout skillDeltaChangeCoverage
claude-codeAll cases1,471,5953,485,627-2,014,032-57.78%skill 8/8; base 8/8
claude-codeholoscan-module-lifecycle-005663,6601,737,070-1,073,410-61.79%skill 1/1; base 1/1
claude-codeholoscan-module-lifecycle-006115,76494,110+21,654+23.01%skill 1/1; base 1/1
claude-codeholoscan-module-lifecycle-008116,767915,082-798,315-87.24%skill 1/1; base 1/1
claude-codeholoscan-module-lifecycle-009146,427187,252-40,825-21.80%skill 1/1; base 1/1
claude-codeholoscan-module-lifecycle-010111,872235,986-124,114-52.59%skill 1/1; base 1/1
claude-codeholoscan-module-lifecycle-011107,150125,336-18,186-14.51%skill 1/1; base 1/1
claude-codeholoscan-module-lifecycle-012107,63164,810+42,821+66.07%skill 1/1; base 1/1
claude-codeholoscan-module-lifecycle-013102,324125,981-23,657-18.78%skill 1/1; base 1/1
codexAll cases532,333452,365+79,968+17.68%skill 8/8; base 8/8
codexholoscan-module-lifecycle-005127,752129,909-2,157-1.66%skill 1/1; base 1/1
codexholoscan-module-lifecycle-00614,23813,979+259+1.85%skill 1/1; base 1/1
codexholoscan-module-lifecycle-00898,62255,373+43,249+78.10%skill 1/1; base 1/1
codexholoscan-module-lifecycle-009115,42159,213+56,208+94.93%skill 1/1; base 1/1
codexholoscan-module-lifecycle-01089,627118,837-29,210-24.58%skill 1/1; base 1/1
codexholoscan-module-lifecycle-01130,22524,962+5,263+21.08%skill 1/1; base 1/1
codexholoscan-module-lifecycle-01241,89335,744+6,149+17.20%skill 1/1; base 1/1
codexholoscan-module-lifecycle-01314,55514,348+207+1.44%skill 1/1; base 1/1
ALL AGENTSDataset aggregate2,003,9283,937,992-1,934,064-49.11%skill 16/16; base 16/16

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

TierPurposeStatusEvidence
Tier 1Static validationPASSED11 validator(s); 0 finding(s)
Tier 2Semantic deduplicationPASSED2 validator(s); 0 finding(s)
Tier 3Live agent evaluationPASS2 agent(s); 8 task(s)

Findings and Observations

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

  • Schema & Repository Governance: Found skill manifest: SKILL.md

  • Semantic Version Validation: No semantic version label present; resource will use commit-hash history (opting back out of an existing label is allowed)

  • Security Scan: No security vulnerabilities detected (secrets, API keys, credentials)

  • PII Scan: Scanning 5 files for PII

  • Code Integrity & Hygiene: Checking 5 markdown files for dead links

  • Unicode Smuggling Detection: No invisible Unicode characters detected in 5 file(s)

  • QUALITY: Score: 100.0/100 (Grade: A)

  • SCRIPT_LINT: No scripts/ directory found

  • Context Deduplication: Collected 5 file(s)

  • Inter-Skill Deduplication: Parsed skill 'holoscan-module-lifecycle': 174 char description

  • AGENT_EVAL: Tier 3 evaluation complete: verdict PASS; best agent claude-code

</details>

Scoring Methodology

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

DimensionQuestionScored signals
SecurityIs it safe to use?security (100%)
CorrectnessIs the answer correct?accuracy (100%)
DiscoverabilityWas the right skill loaded when needed?skill_execution (100%)
EffectivenessDid the skill help complete the task?goal_accuracy (50%) + behavior_check (50%)
EfficiencyDid 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.

Needs: Needs Python

Security

What our scanner noticed· Medium risk

  • Large base64-encoded blobs
  • Instructions that use the network

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
Holoscan Team
Publisher
NVIDIA (Official)
Distributed by
NVIDIA
Source
NVIDIA/skills · skills/holoscan-module-lifecycle · commit fce3cca
Licence
Apache License 2.0 (redistribution allowed)
  • LICENSE (added by AutoGineer from LICENSE-APACHE)
  • ATTRIBUTION.md (written by AutoGineer)
Synced

AutoGineer is not affiliated with or endorsed by Holoscan Team or NVIDIA.

Download ZIP

The ZIP holds the skill’s files, its licence files and an ATTRIBUTION.md. Download holoscan-module-lifecycle.zip