PlayCanvas Skills

| Skills.sh | Developer Site | Engine Manual | API Reference | Examples | Forum |

Portable skills that help AI coding agents build and verify polished applications with playcanvas, @playcanvas/react, and @playcanvas/web-components.

Install

Add the skills to your project with the Agent Skills installer:

npx skills add playcanvas/skills

Use --all to install every skill into all detected agents or -g for a user-level installation. Start a new conversation after installing or updating so the agent discovers the new skill metadata.

Choose one installation route for each agent. Installing both Agent Skills and a native plugin for the same agent exposes duplicate skills.

Native plugins

Why PlayCanvas Skills

• Resolve the active PlayCanvas authoring surface from the project instead of guessing from installed dependencies.
• Check installed package declarations, exports, examples, and production scripts before using an API.
• Inspect GLB geometry, transforms, clips, joints, morphs, and hierarchy before choosing placement or animation values.
• Keep scene ownership, model calibration, physics, effects, UI, and app state predictable.
• Reuse Engine features such as CameraControls, Water, ProceduralSky, and CameraFrame with their required integrations intact.
• Finish with runtime and screenshot evidence instead of treating a successful build as visual proof.

Usage

Ask your coding agent for the outcome you want. The matching skills supply the PlayCanvas-specific workflow and verification steps. For example:

Build a Direct Engine product viewer for these GLBs. Inspect and calibrate every model, reuse the
shipped camera controls, and verify grounding and framing in the browser.

Polish this @playcanvas/react prototype with compressed assets, reused Engine scripts, pooled
effects, deliberate lighting and tone mapping, and screenshot checks.

Supported surfaces

The skills resolve the active surface from imports and markup, then read only its matching reference:

Skills prefer the packages installed in the target project. CI compiles representative usage against pinned supported package versions, and Renovate proposes dependency updates.

Skills

Every integration loads the same canonical files from skills/. Host manifests contain distribution metadata only.

PlayCanvas Skills targets Engine application workflows. It does not automate the PlayCanvas Editor, create or publish Editor projects, manage cloud services, or replace project-specific art direction and gameplay design.

Context cost

Agents load only each skill’s name and description at session start, about 700 tokens for all 16 skills. A skill’s full instructions load when a task triggers it, adding 350–550 tokens, plus at most one surface reference of up to 500 tokens. Estimates assume 4 characters per token. Each skill stays within the word budget in AGENTS.md.

Data and network use

The skills collect, store, and send no user data, and the plugin has no hooks or MCP servers. A few skills tell the agent to run commands on the user’s machine:

• inspect-glb runs its bundled scripts/inspect.mjs on local GLB files. For compressed files it runs npx --yes @gltf-transform/[email protected], which downloads that pinned package from npm.
• inspect-runtime injects its bundled scripts/devtools-hook.js into the app’s own page.
• find-examples reads official examples from the public playcanvas/engine, playcanvas/web-components, and playcanvas/developer-site repositories with gh api when they are not installed locally.

Contributing

See CONTRIBUTING.md for pull-request expectations. For repeatable skill failures, use the skill feedback template and include the prompt, surface, package versions, output, and smallest incorrect behavior. Report vulnerabilities through GitHub Security Advisories.

License

PlayCanvas Skills is released under the MIT License.