Claude Code plugins and skills
The Cognitive3D plugin marketplace adds Cognitive3D knowledge to Claude Code. Each plugin installs a skill that Claude loads when you ask about its topic, so it can plan an SDK integration, call the REST API, or integrate the ROS 2 SDK without you looking up the details first.
Prerequisites
- Claude Code, installed and signed in.
- A Cognitive3D account.
- An organization API key if you want Claude to call the REST API for you. See API overview.
Add the marketplace
The marketplace is a public GitHub repository, CognitiveVR/c3d-marketplace-public. Add it by running this command in a Claude Code session:
/plugin marketplace add CognitiveVR/c3d-marketplace-public
Then install each plugin you want by name:
/plugin install cognitive3d-sdk-implementation@c3d-marketplace-public
/plugin install cognitive3d-public-api@c3d-marketplace-public
/plugin install cognitive3d-ros2-integration@c3d-marketplace-public
Available plugins
| Plugin | Skill command | What it helps with |
|---|---|---|
cognitive3d-sdk-implementation |
/cognitive3d-sdk-implementation |
Planning what to track and integrating the Cognitive3D SDK in Unity, Unreal Engine, Apple visionOS, Android XR, WebXR, and C++ projects. |
cognitive3d-public-api |
/cognitive3d-public-api |
Choosing REST API endpoints, building slicer queries, and parsing responses. |
cognitive3d-ros2-integration |
/integrate-cognitive3d-ros2 |
Adding the Cognitive3D SDK for ROS 2 to your robot's existing ROS 2 setup. |
SDK implementation
The cognitive3d-sdk-implementation skill takes you from an uninstrumented project to a phased tracking plan and the code that implements it. It works through these steps:
- It identifies which SDK your project targets and loads only that SDK's reference. If it cannot tell from the project, it asks.
- It asks discovery questions about your goals before it recommends any instrumentation.
- It builds a phased tracking plan of custom events, Dynamic Objects, ExitPoll surveys, and session properties.
- It screens the plan against what your SDK and platform can capture, because capability differs between SDKs.
- It keeps event properties readable and avoids recording data the platform already captures.
- It turns place-based questions, such as whether participants reached or looked at an area, into dashboard zones instead of custom events.
The skill also points you to the relevant SDK APIs and documentation pages.
Most of the plan is code that Claude writes in your project. Some steps happen outside your code and need a key or a role:
| Step | Who does it | What it needs |
|---|---|---|
| Custom events, session properties, sensors, and other code. | Claude, in your project. | The SDK set up in your project with its application key. No other key. |
| Scene and Dynamic Object mesh uploads. | You, in your engine's Cognitive3D tools or the Upload Web App. | Your developer key. |
| Objectives, ExitPoll question sets, and zones. | You on the dashboard, or Claude through the MCP Server. | On the dashboard, no key. Through the MCP Server, an organization-admin or project-admin role, and a write-enabled organization API key if you connect with a key instead of signing in. |
The skill asks which route you want for objectives, ExitPoll question sets, and zones before it changes anything. An organization API key applies to every project in your organization, so staying on the dashboard for these steps is a reasonable choice.
Public API
The cognitive3d-public-api skill gives Claude the endpoint catalog and the slicer query syntax for the Cognitive3D REST API. Use it to:
- Choose the right endpoint for session, gaze, event, sensor, Dynamic Object, objective, and ExitPoll data.
- Build slicer queries with session filters, event filters, aggregations, and slice-bys.
- Parse and interpret API responses.
- Write Python, JavaScript, or C# scripts for data pipelines and dashboards.
Scripts that call the REST API authenticate with your organization API key.
For example, you can ask Claude to:
- "Build a simple web dashboard that shows sessions per week and average session duration for my project."
- "Write a script that fetches objective results for each participant so I can send completions to our learning management system."
- "Export last month's sessions with their session properties to a CSV file every night."
- "Pull the responses to our end-of-session ExitPoll and summarize them by question."
- "Which Dynamic Objects did participants look at the longest in the Showroom scene?"
ROS 2 integration
The integrate-cognitive3d-ros2 skill adds the Cognitive3D node to a robot's existing tf tree, topics, and launch files. It follows these steps:
- It surveys the live robot, including the ROS 2 distribution, middleware, tf tree, topics, actions, and namespaces.
- It decides each parameter from what it measured rather than from examples.
- It writes the params file, adds the node to your robot's own launch file, and runs
doctor. - It records and verifies the first session, then hands you a report of every decision.
The skill asks you to export your keys as environment variables and never reads or writes a key itself.
doctor is a command that ships with the SDK. It checks your configuration, credentials, network path, and live ROS graph without opening a session or uploading anything. Many integration mistakes produce no error at run time, and doctor is where they become visible. The skill runs it before the first launch. Run it yourself again after every change to the params file. See Doctor and verification.
AI-assisted integration is experimental, so check the result yourself:
- In the params file, confirm that the scene, the pose convention, and the frames match your robot, and that the topic list includes the sensors you want to record.
- In the
doctoroutput, read every failure and note. The skill reports them to you word for word.
Then open the first session in Session Replay and compare it with what the robot did:
- The path lies on the floor plan, where the robot drove.
- The robot, or its gaze line, faces the direction of travel.
- The sensor series you chose are present under their own names.
- The events and action outcomes you expect are there.
- Each camera and lidar shows frames.
This plugin is served from the ROS 2 SDK repository at a release tag, so the skill matches that SDK version. The skill checks your installed SDK version first. To pin the skill in your own repository, see the agent skills README.
Use a skill
A skill activates on its own when you ask Claude about its topic. For example, ask "Plan the Cognitive3D tracking for this Unity project" in a Unity project, and Claude loads the SDK implementation skill.
To invoke a skill directly, type its command in Claude Code:
/cognitive3d-sdk-implementation
/cognitive3d-public-api
/integrate-cognitive3d-ros2
Use the skills with other AI tools
The skills are plain Markdown files, so you can use them with AI tools that do not support Claude Code plugins. The guidance is written for Claude Code, so results in other tools can vary.
Each skill is a folder that contains a SKILL.md file and supporting reference files. SKILL.md holds the instructions and tells the assistant which reference file to read for a given task.
| Skill | Folder |
|---|---|
| SDK implementation | plugins/cognitive3d-sdk-implementation/skills/cognitive3d-sdk-implementation |
| Public API | plugins/cognitive3d-public-api/skills/cognitive3d-public-api |
| ROS 2 integration | .claude/skills/integrate-cognitive3d-ros2 in the ROS 2 SDK repository |
To use a skill in another tool:
- Download or clone the repository that holds the skill.
- Copy the skill folder into your project, and keep
SKILL.mdand its reference files together. - Give the skill to your assistant in the way your tool supports:
- If your tool reads an instructions file from the repository, such as
AGENTS.md, add a line that tells the assistant to read the skill'sSKILL.mdbefore it works on Cognitive3D tasks. - If your tool only reads files that you attach to the chat, such as GitHub Copilot Chat, attach
SKILL.mdand each reference file that your task needs. For example, attachunity_sdk_reference.mdfor a Unity project.
- If your tool reads an instructions file from the repository, such as
The marketplace repository also publishes the public API skill in a layout for the Code Puppy agent. See the code_puppy_skills folder.
Update plugins
To refresh the marketplace so Claude Code sees the latest plugin versions, run this command in a Claude Code session:
/plugin marketplace update c3d-marketplace-public
Then update each installed plugin from your terminal:
claude plugin update cognitive3d-sdk-implementation@c3d-marketplace-public
Restart Claude Code to load the new version.
Migrate from the Unity plugin
The cognitive3d-sdk-implementation plugin replaces the retired cognitive3d-unity-implementation plugin. All of the Unity guidance carries over. If you installed the Unity plugin, run these two commands in a Claude Code session:
/plugin marketplace update c3d-marketplace-public
/plugin install cognitive3d-sdk-implementation@c3d-marketplace-public
After the first command, Claude Code moves your settings to the new name and shows a one-time "Renamed to cognitive3d-sdk-implementation" notice. Until you run the second command, sessions report Plugin "cognitive3d-sdk-implementation" not cached. You do not need to uninstall the old plugin.
See also
- Agentic tools overview — compare every agentic tool Cognitive3D offers.
- MCP Server — connect an AI assistant to your analytics data.
- CLI for agents — drive the command line tool from a coding agent or a script.
- Get started with ROS 2 — install and configure the Cognitive3D SDK for ROS 2.
If you have a question or any feedback about our documentation please use the Intercom button (purple circle) in the lower right corner of any web page or join our Discord.