New: 98,000+ skills indexed from GitHub repos with 100+ stars · every skill safety-checkedBrowse top skillsTop skills
Esc to closeBrowse all skills
Submit a skillBrowse skills
Submit a skillBrowse skills

Official agent skill

Port External Models

by NVIDIA in NVIDIA/structured-data-models

Implement or review a port of an external model in SDM, from upstream inference behavior and checkpoints through public prediction parity and reviewable PRs.

OfficialApache-2.0Auto-check passed

Install Port External Models

skills CLI
$ npx skills add NVIDIA/structured-data-models --skill port-external-models -a claude-code

Project install by default; add -g for ~/.claude/skills/.

GitHub CLI
$ gh skill install NVIDIA/structured-data-models port-external-models --agent claude-code

Project scope by default; add --scope user for a personal install. Needs GitHub CLI 2.90.0 or later (public preview).

Manual copy
$ git clone --depth 1 https://github.com/NVIDIA/structured-data-models.git skills-src && mkdir -p .claude/skills && cp -r skills-src/.agents/skills/port-external-models .claude/skills/port-external-models && rm -rf skills-src

Use ~/.claude/skills/ instead of .claude/skills for a personal install. The folder must contain SKILL.md.

Claude Code skills documentation · loads skills from .claude/skills/

Facts

Skill name
port-external-models
GitHub stars
316
Token cost
~1.3k tokens
SKILL.md length
703 words
Files
1
Skills in repo
4
Repo updated
First seen
Licence
Apache-2.0

At a glance

Implement or review a port of an external model in SDM, from upstream inference behavior and checkpoints through public prediction parity and reviewable PRs.

  • Works in 5 steps: Skeleton (optional): package and… → Blocks: one numerical piece per PR… → Working path: assemble the core, load… → …
  • SKILL.md covers Establish the reference, Build the SDM path, Plan reviewable PRs and Verify and review
  • Instructions only: no scripts, shell commands, URLs or credentials in SKILL.md

What it does

Port External Models is an agent skill from NVIDIA/structured-data-models, published by the product's own GitHub organization. Implement or review a port of an external model in SDM, from upstream inference behavior and checkpoints through public prediction parity and reviewable PRs.

Its SKILL.md is about 1.3k tokens, which your agent loads only when the skill is triggered. It is a single SKILL.md file with no bundled scripts.

The repository describes itself as: Foundation Models for Structured Data. The licence is Apache-2.0.

Example prompts

  • “/port-external-models”

Workflow steps

5 steps, taken from the first numbered list in SKILL.md.

  1. Skeleton (optional): package and ICLModel subclass, imported only from the submodule. Zeros are allowed here.
  2. Blocks: one numerical piece per PR (sdm.nn or model component) with a focused tensor test. Add only what the working path needs.
  3. Working path: assemble the core, load the pinned checkpoint, and run one public forward/predict for one supported input kind (related…
  4. Capabilities: extra input types, auxiliaries, missing values, size limits, recipe details. Keep each capability's implementation, tests…
  5. Index (last): add the model to sdm.models and the docs index.

What it can do on your machine

Read from SKILL.md and the folder at commit 842c408. It shows what the files ask for, not the result of running them.

  • Tool permissions

    Pre-approves nothing: there is no allowed-tools line, so your agent's usual permission prompts apply.

    From allowed-tools in the SKILL.md frontmatter.

  • Runs code

    No scripts in the folder and no shell commands in SKILL.md.

    From the folder's file list and the shell code blocks in SKILL.md.

  • Network

    No URLs in SKILL.md.

    From URLs in SKILL.md, links to its own repository left out.

  • Credentials

    Names no API keys, tokens, secrets or passwords.

    From names ending in _API_KEY, _TOKEN, _SECRET, _KEY or _PASSWORD in SKILL.md.

Context cost

Port External Models loads about 1.3k tokens when it runs. Until then it costs about 45 tokens; SKILL.md has 703 words of instructions outside code blocks.

Always · name and description, kept in context so the agent knows when to use it
~45
When it runs · the whole SKILL.md, loaded when a task matches
~1.3k

Estimates: characters ÷ 4, the usual rule of thumb; real counts depend on the model's tokenizer. Scripts and assets cost tokens only if the agent reads them.

Safety

Auto-check passed

The automated check found no risky patterns in SKILL.md.

Automated static check — not a guarantee. Review scripts before installing. It scans the text of SKILL.md for risky patterns (piping downloads into a shell, reading credential files, hidden Unicode, destructive commands); files beside SKILL.md are not scanned.

SKILL.md

The full file from NVIDIA/structured-data-models at commit 842c408, republished under its Apache-2.0 licence (© NVIDIA). 703 words, ~1,307 tokens.

Download SKILL.mdSave it as .claude/skills/port-external-models/SKILL.md (or your agent's skills folder).
name
port-external-models
description
Implement or review a port of an external model in SDM, from upstream inference behavior and checkpoints through public prediction parity and reviewable PRs.

Port an External Model to SDM

Follow the repository-root AGENTS.md. When a port adds reusable processors or changes a Recipe, also follow the processor-development or recipe-development skill, respectively. Use docstring when writing or reviewing public docstrings.

Establish the reference

  • Which revision: Pin the upstream git SHA, Hub revision, model config, and source vs weight licenses. Put them in the PR description; when merging, also in THIRD_PARTY_LICENSES.md, the model docstring, and the checkpoint load path.
  • Which behaviour in this revision: Trace an upstream public inference call through input preparation, numerical execution, and output restoration. Identify the functions responsible for each stage; a low-level forward method alone does not establish public behavior.
  • Which public contract: Record accepted forms and shapes, feature and target roles, auxiliary inputs, batching and size limits, missing and non-finite values, masks or padding, preprocessing and output restoration, and output ordering where applicable. Include any task-specific semantics that affect predictions. Distinguish public boundary behavior from numerical model computation, and state which capabilities the SDM port supports.

Build the SDM path

  • Before porting blocks, specify the SDM forward: which public inputs become which core tensors, and how core outputs become the returned TableTensor.
  • Keep the model wrapper responsible for SDM inputs, outputs, and lifecycle. Put reusable numerical operations in sdm.nn, model-specific composition and checkpoint handling in model components, and independently reusable preprocessing in sdm.processing.
  • ICLModel._forward takes TableTensor (plus related tables when used). Unwrap to Tensor for numerical work and keep that tensor's device and dtype. sdm.nn takes Tensor only; keep TableTensor only while column or relation schema is still required.
  • Start from existing sdm.nn components. Adapt one to the required semantics when the change is reusable across models; otherwise add a reusable component for a distinct operation. Prefer an efficient, numerically equivalent formulation over copying upstream computations, and verify it against the reference.
  • Design components for clear SDM semantics, not for strict loading of an upstream state dict. Remap checkpoint keys or tensors at the loading boundary as needed, and verify that all weights required by supported inference are loaded.
  • Write computations in execution order with short, typed methods and meaningful intermediate names. Extract helpers for distinct behavior or real reuse; omit upstream factories, configuration layers, and wrappers that SDM does not need.
  • Compose default recipes from existing public processors. Use a fitted recipe transform only when its state can be learned from permitted data and applied without leakage; ensure any inverse transform works on the prediction's actual shape.
Show full SKILL.md (294 more words)Show less

Plan reviewable PRs

Split the port into this stack. These are phases, not five PRs: Blocks and Capabilities are one PR each. Title PRs [Model N/n] with n the total count. Each PR has one reviewable outcome, names its parent PR, and keeps the branch working. Treat roughly 300 changed lines as a prompt to split, not a reason for incomplete PRs.

  1. Skeleton (optional): package and ICLModel subclass, imported only from the submodule. Zeros are allowed here.
  2. Blocks: one numerical piece per PR (sdm.nn or model component) with a focused tensor test. Add only what the working path needs.
  3. Working path: assemble the core, load the pinned checkpoint, and run one public forward/predict for one supported input kind (related tables only if required). The prediction matches the pinned reference; a public behavior test covers it. Zeros, stubs, or block-only tests do not complete this phase.
  4. Capabilities: extra input types, auxiliaries, missing values, size limits, recipe details. Keep each capability's implementation, tests, and needed docs together.
  5. Index (last): add the model to sdm.models and the docs index.

Verify and review

  • Generate deterministic reference outputs from the pinned upstream revision. Compare observable SDM predictions through the public path, including checkpoint loading and output ordering. Claim end-to-end parity only after this comparison.
  • Test supported boundaries and edge cases, such as missing or constant inputs, multiple outputs, auxiliary data, and size limits where applicable. Assert public behavior rather than internal helper layout.
  • Benchmark performance-sensitive GPU changes with synchronization-aware timing.
  • When reviewing a proposed port, reconstruct the upstream public contract and intended SDM call path before evaluating the diff. Check the working public slice, reference evidence, leakage boundaries, checkpoint behavior, and parity claims against that contract; suggest removing machinery unrelated to the supported path.

© NVIDIA, Apache-2.0. Rendered from Markdown: HTML in the file is shown as text, images as links, and headings moved down two levels. Raw file

Files

Just SKILL.md in .agents/skills/port-external-models of NVIDIA/structured-data-models.

Open the folder on GitHubat commit 842c408

Compare with similar skills

Port External Models next to the 5 skills that share the most tags, products or categories with it. Stars are the repository's; “used in” counts other GitHub owners with a copy.

Port External Models compared with similar skills
SkillStarsUsed inTokensAuto-checkLicenceRepo updated
Port External Models this skillNVIDIA/structured-data-models316—~1.3kAutomated safety check: PassApache-2.0
Implementsickn33/agentic-awesome-skills47k5 repos~306Automated safety check: PassMIT
Implementcodewhale-hq/Codewhale41k—~190Automated safety check: PassMIT
Incremental Implementationaddyosmani/agent-skills103k1 repos~2.3kAutomated safety check: PassMIT
Implementbestofjs/bestofjs3.1k18 repos~109Automated safety check: PassMIT
ImplementAutomattic/simplenote-android1.9k—~1.1kAutomated safety check: PassGPL-2.0

Similar skills

  • Implement

    sickn33/agentic-awesome-skills

    Implement a piece of work based on a PRD or set of issues. An agent skill from sickn33/agentic-awesome-skills.

    47k GitHub starsUsed in 5 repos~306 tokens
    Product & Project ManagementAuto-check passed

  • Implement

    codewhale-hq/Codewhale

    Carry an authorized, defined request or approved plan through scoped edits and proportionate verification.

    41k GitHub stars~190 tokensUpdated today
    Auto-check passed

  • Incremental Implementation

    addyosmani/agent-skills

    Delivers a change in thin vertical slices, each implemented, tested, verified and committed before the next, using vertical, contract-first or risk-first slicing.

    103k GitHub starsUsed in 1 repo~2.3k tokens
    Agent WorkflowsAuto-check passed

  • Implement

    bestofjs/bestofjs

    Implement a piece of work based on a spec or set of tickets.

    3.1k GitHub starsUsed in 18 repos~109 tokens
    Auto-check passed

  • Implement

    Automattic/simplenote-android

    End-to-end implementation workflow: plan, implement, verify, commit, and open a draft PR.

    1.9k GitHub stars~1.1k tokensUpdated yesterday
    Testing & QAAuto-check passed

  • Sparc Implement

    ruvnet/ruflo

    Run the SPARC Pseudocode and Architecture phases (2 and 3) — write algorithm pseudocode, design module boundaries and API contracts, then implement

    74k GitHub stars~1.3k tokensUpdated today
    Backend & APIsAuto-check: notes

More from NVIDIA/structured-data-models

  • Docstring

    NVIDIA/structured-data-models

    Official

    Write or review docstrings for public modules, classes, and functions in sdm/.

    316 GitHub stars~1.2k tokensUpdated today
    Auto-check passed

  • Processor Development

    NVIDIA/structured-data-models

    Official

    Create or modify reusable SDM processors and focused tests. An agent skill from NVIDIA/structured-data-models.

    316 GitHub stars~1.4k tokensUpdated today
    Auto-check passed

  • Recipe Development

    NVIDIA/structured-data-models

    Official

    Compose or restructure SDM model recipes. An agent skill from NVIDIA/structured-data-models.

    316 GitHub stars~156 tokensUpdated today
    Auto-check passed

Questions about Port External Models

What does Port External Models do?

Implement or review a port of an external model in SDM, from upstream inference behavior and checkpoints through public prediction parity and reviewable PRs. Port External Models is an agent skill from NVIDIA/structured-data-models, published by the product's own GitHub organization. Implement or review a port of an external model in SDM, from upstream inference behavior and checkpoints through public prediction parity and reviewable PRs.

How do I install Port External Models in Claude Code?

Run `npx skills add NVIDIA/structured-data-models --skill port-external-models -a claude-code`. Or copy the skill folder (.agents/skills/port-external-models in NVIDIA/structured-data-models) into .claude/skills/port-external-models in your project. Claude Code loads it when a task matches its description.

How do I install Port External Models in Codex?

Run `npx skills add NVIDIA/structured-data-models --skill port-external-models -a codex`. Or copy the skill folder (.agents/skills/port-external-models in NVIDIA/structured-data-models) into .agents/skills/port-external-models in your project. Codex loads it when a task matches its description.

Can I use Port External Models in Cursor, Gemini CLI or GitHub Copilot?

Cursor, Gemini CLI, GitHub Copilot and OpenCode also load SKILL.md folders. With the skills CLI, run `npx skills add NVIDIA/structured-data-models --skill port-external-models -a cursor` (or -a gemini-cli, github-copilot or opencode for the others). To copy it by hand, put the folder in .cursor/skills/port-external-models, .gemini/skills/port-external-models, .github/skills/port-external-models and .opencode/skills/port-external-models in your project.

What does Port External Models need to run?

SKILL.md names no scripts, command-line tools or credentials: Port External Models is instructions for the agent only.

Does Port External Models access the network?

SKILL.md contains no URLs. Any network use would come from the scripts or tools the agent runs. This is read from the text; nothing was executed.

Is Port External Models safe to install?

Our automated static check of SKILL.md found no risky patterns, such as piping downloads into a shell, reading credential files or hidden Unicode. It is not a guarantee. Review the folder before installing.

What licence does Port External Models use?

Port External Models is published under the Apache-2.0 licence (the repository's licence). It allows redistribution, so the full SKILL.md is shown on this page.

How many tokens does Port External Models use?

About 1.3k tokens (SKILL.md is roughly 5.2k characters). Agents keep only the skill's name and description in context until a task matches; then they load SKILL.md in full.

What are the alternatives to Port External Models?

Skills that share tags, products or a category with Port External Models: Implement (sickn33/agentic-awesome-skills, 47k stars), Implement (codewhale-hq/Codewhale, 41k stars), Incremental Implementation (addyosmani/agent-skills, 103k stars) and Implement (bestofjs/bestofjs, 3.1k stars). The comparison table on this page puts their stars, adoption, token cost, safety result and licence side by side.

Who maintains Port External Models?

NVIDIA (a GitHub organization, an official publisher) maintains it in NVIDIA/structured-data-models, which has 316 GitHub stars. The repository holds 4 skills in this directory. The repository was last updated on October 8, 2026.

Source: NVIDIA/structured-data-models on GitHub. Facts on this page come from the repository at the commit we read; the author's words are quoted as theirs.