Skills

Overview

Choose a coding-agent workflow for running a demo, setting up a checkout, building an application, or reviewing repository changes. Each skill supplies the instructions for that workflow; installation and invocation examples are below.

Skills are small SKILL.md modules under the open Agent Skills spec that a coding agent installs and follows.

Run the browser demo

Choose run-demo to try xr-ai. Paste this to your coding agent:

Run the xr-ai Simple VLM browser demo for me. Reuse the checkout and ref
selected in this conversation or the current xr-ai working directory;
otherwise ask whether to use the latest release
(newest stable tag; a prerelease only if no stable exists) or main. Fetch
skills/run-demo/SKILL.md at that ref from
https://raw.githubusercontent.com/NVIDIA/xr-ai/. If it is absent at a
release tag, get my OK before using main for the skill, checkout, and docs.
Install the skill (or just follow it), handle any setup it requires, and
leave the demo running. Return the browser link, interaction verification
status, and stop instructions.

run-demo handles preparation with setup, leaves the Simple VLM browser demo running, and returns its link, interaction verification status, and stop instructions. Local model servers remain running independently of the sample; model shutdown is a separate action.

The demo uses the deprecated Simple VLM sample, which remains available for trying the stack. For new voice-and-vision application work, use Conversation example as the recommended foundation.

Prefer to do it by hand? Follow Quickstart.

Set up for development

To prepare a checkout and its prerequisites with setup, paste this to your agent:

Set up xr-ai for me: ask me whether to build against the latest release
(newest stable tag; a prerelease only if no stable exists) or main, then
fetch skills/setup/SKILL.md at that ref from
https://raw.githubusercontent.com/NVIDIA/xr-ai/. If it is absent at the
selected release, explain that and get my OK to use main for the skill,
checkout, and docs. Report other retrieval failures. Follow the skill and
prepare the checkout and prerequisites without launching a sample. Report
the checkout path, selected ref and model environment, and any setup blockers.

The skill has the agent ask two setup questions (latest release or main; models on the local GPU or a hosted endpoint), clone the repo, and route itself to the working contract, the docs, and the reference sample. An agent with no skills mechanism can follow the SKILL.md contents directly.

Build your application

For application work, build-your-app is also a complete entry point. It uses the published build guide before the checkout exists and continues with the same guide from the selected repository ref after cloning.

To start directly with an application, describe its goal in this prompt and paste it to your coding agent:

Build an xr-ai application that <describe what it should do>. Ask me whether
to use the latest stable release (or the latest prerelease if no stable release
exists) or main, then fetch skills/build-your-app/SKILL.md at that ref from
https://raw.githubusercontent.com/NVIDIA/xr-ai/. If the file does not exist at
that ref, get my approval before using main for the skill, guide, and checkout.
Install the skill (or just follow it), then build and verify the application.

Available skills

Skill	What it does
run-demo	Runs the Simple VLM browser demo and returns its connection link and verification status
setup	Prepares a checkout and prerequisites for development, with pointers to the working contract, docs, and reference sample
build-your-app	Guides your coding agent from checkout setup through building an application
gh-review-xr-ai	Reviews xr-ai PRs with strict scope discipline and comment-only feedback
gh-manage-xr-ai-reviews	Tracks an XR-AI review inbox with overall status, next actions, and safe batching

The bank lives at skills/ in the repository; new skills land there with the features they cover.

Manual installation

Set REF to the repository ref you will build against, choose a SKILL from the table, and download it into the directory used by your coding agent:

SKILLS_DIR=/path/to/your/agent/skills
REF=main  # or a release tag such as v0.4.0-beta
SKILL=run-demo
curl -fsSL --create-dirs -o "$SKILLS_DIR/$SKILL/SKILL.md" \
  "https://raw.githubusercontent.com/NVIDIA/xr-ai/$REF/skills/$SKILL/SKILL.md"

gh-manage-xr-ai-reviews composes gh-review-xr-ai; install both review skills together. The review skills also ship agents/openai.yaml, optional Codex interface metadata for their display name, short description, and default prompt. After setting SKILL to either review skill, install that metadata with:

curl -fsSL --create-dirs -o "$SKILLS_DIR/$SKILL/agents/openai.yaml" \
  "https://raw.githubusercontent.com/NVIDIA/xr-ai/$REF/skills/$SKILL/agents/openai.yaml"

If the requested skill is absent at the selected release, ask the user before using main for the skill, checkout, and documentation. An agent without a skill installation mechanism can read the downloaded SKILL.md and follow it directly in the current session.