Use skills and agent roles for game development
Read bundled guides, inspect project trust and eligibility, and assign focused responsibilities to agent work.
On this page
Skills supply instructions and references for a task. Roles describe the responsibility an agent carries and the limits within which it works. Neither one creates a missing engine installation or bypasses tool access rules.
Browse the bundled library
Open Skills & Roles and select a project. The current source documents thirty bundled game-development skills covering planning, implementation, art, audio, testing, accessibility, research, and release work.
Inspect the origin, scope, and enabled state. Use Read guide for the skill's entrypoint and the document selector for supporting references. Check engine tags and role or work-type restrictions before assuming a skill is eligible for a particular task.
The catalog is shown in a disposable project, using application-provided guides.
Choose a focused role
Roles such as planner, gameplay engineer, asset producer, reviewer, and validator have different responsibilities. Use them to divide an implementation from its review and acceptance evidence. Model pools and tool ceilings still apply.
Ask an asset role for editable source and export checks; ask a validator for the relevant engine result. Reading instructions alone is not evidence that either operation was executed.
Add project-specific instructions
Project skills use a directory containing SKILL.md, frontmatter, task instructions, and optional references or scripts under .agents/skills/. Keep the main guide focused and place lengthy reference material in separate files.
Project-local skills require project trust and can override a matching name. Platform or plugin overrides also affect which version becomes visible; review existing pins when replacing one. Bundled copies are application assets and are not uninstalled through the same control as an installed override.
See extension installation for package workflows and approvals for the separate execution boundary.