provider

kind: skill

Defines a reusable procedure that agents invoke on-demand. Compiled to skills/<id>/SKILL.md with YAML frontmatter for all registered providers.

Required: kind, version, name

Source Directory

xcaf/skills/<name>/skill.xcaf

Per-provider overrides use a sibling file with a provider suffix:

xcaf/skills/<name>/skill.claude.xcaf
xcaf/skills/<name>/skill.cursor.xcaf

Example Usage

Minimal skill

---
kind: skill
version: "1.0"
name: conventional-commits
description: >-
  Structured Git commit workflow using Conventional Commits with pre-commit
  checks, logical grouping, and anti-jargon enforcement.
---
# Conventional Commits

Use `feat`, `fix`, `chore`, `docs`, `refactor`, `test` as commit type.
Format: `<type>(<scope>): <description>`

Never include `Co-Authored-By` trailers.
Run `git diff --staged` before composing the message.

Simple skill — instructions and rules

---
kind: skill
version: "1.0"
name: component-patterns
description: >-
  Step-by-step procedure for implementing a new React component following
  frontend-app conventions — file layout, prop types, Tailwind styling,
  accessibility attributes, and co-located test scaffolding.
---
# Component Patterns

## When to Use
Invoke this skill whenever you are asked to create a new React component
or refactor an existing one to meet project standards.

## Step 1 — Confirm the component location
All new components live under `src/components/<Category>/<ComponentName>/`.
If the category does not exist, create it.

## Step 2 — Create the file structure
Each component directory contains four files:
- `index.ts` — re-export barrel
- `ComponentName.tsx` — implementation
- `ComponentName.test.tsx` — co-located tests
- `ComponentName.stories.tsx` — Storybook stories (optional)

## Step 3 — Implement with forwardRef
Use `forwardRef` for all interactive components. Accept a `className`
prop and merge it with `cn()` from `@/lib/utils`.

## Step 4 — Write co-located tests
Every component ships with at least two tests: renders correctly
and responds to its primary interaction.

## Step 5 — Export from the barrel
Append the public export to `src/components/index.ts`.

## Rules
- One component per directory. No multi-component files.
- All props must have TypeScript types. No `any`.
- Accessibility: every interactive element needs an aria label or role.

Skill with artifacts — single folder

The artifacts field declares subdirectories that are composed alongside the compiled SKILL.md. Files within each subdirectory are discovered automatically from the filesystem.

Source directory layout:

xcaf/skills/component-patterns/
  skill.xcaf
  references/
    design-tokens.md
    shadcn-primitives.md

Manifest:

---
kind: skill
version: "1.0"
name: component-patterns
description: >-
  Step-by-step procedure for implementing a new React component following
  frontend-app conventions.
artifacts:
  - references
---
# Component Patterns

## When to Use
Invoke this skill when creating or refactoring React components.

## Step 1 — Confirm the component location
…

## Reference Material
Consult `references/design-tokens.md` for spacing and color values.
Consult `references/shadcn-primitives.md` for base component APIs.

The references/ directory and all its files are placed alongside SKILL.md in the compiled output (e.g., .claude/skills/component-patterns/references/design-tokens.md).

Skill with artifacts — multiple folders

A skill can declare more than one artifact subdirectory. Each name maps to a sibling directory on disk.

Source directory layout:

xcaf/skills/deploy-checklist/
  skill.xcaf
  references/
    pre-deploy-gates.md
    rollback-playbook.md
  scripts/
    smoke-test.sh
    health-check.sh
  examples/
    staging.env.example
    production.env.example

Manifest:

---
kind: skill
version: "1.0"
name: deploy-checklist
description: >-
  Production deployment checklist with automated smoke tests
  and rollback procedures.
artifacts:
  - references
  - scripts
  - examples
---
# Deploy Checklist

## Step 1 — Run pre-deploy gates
Review the checklist in `references/pre-deploy-gates.md`.

## Step 2 — Execute smoke tests
Run `scripts/smoke-test.sh` against the staging environment.

## Step 3 — Verify health
Run `scripts/health-check.sh` and confirm all endpoints return 200.

## Step 4 — Rollback plan
If any check fails, follow `references/rollback-playbook.md`.

Valid artifact names:

Name	Typical contents
references	Lookup tables, checklists, design tokens, specification excerpts
scripts	Shell scripts, automation helpers
assets	Static files, images, templates
examples	Sample configurations, fixture files
(custom)	Any kebab-case directory name is accepted

Field Reference

Required Fields

Field	Type	Description
kind	string	Resource type. Must be skill.
version	string	File format version. Must be "1.0".
name	string	Unique skill identifier. Must match [a-z0-9-]+.

Optional Fields

Field	Type	Description
description	string	Human-readable purpose of this skill. What it does and when to invoke it.
when-to-use	string	Guidance for the model on when to invoke this skill. Emitted only for Claude (as when_to_use in SKILL.md frontmatter); ignored by other providers.
license	string	SPDX license identifier (e.g. "MIT", "Apache-2.0"). Emitted for Claude and Copilot; ignored by other providers.
allowed-tools	[]string	Tools the skill is permitted to use. Claude emits as a space-separated string. Copilot emits as a YAML list. Gemini drops this field with a fidelity warning. Cursor and Antigravity do not emit this field.
disable-model-invocation	bool	When true, prevents the skill from spawning a sub-agent. Supported by Claude and Cursor.
user-invocable	bool	When true, exposes the skill as a slash command the user can invoke directly. Claude-only.
argument-hint	string	Hint text shown during slash-command invocation. Has effect only when user-invocable: true. Claude-only.
artifacts	[]string	Named subdirectories to compose with the skill. Valid canonical names: references, scripts, assets, examples. Custom names also accepted. Files within each subdirectory are discovered automatically from the filesystem.
targets	map[string]TargetOverride	Per-provider overrides keyed by provider name.

Filesystem-as-Schema

When a skill .xcaf file lives at xcaf/skills/<name>/skill.xcaf, the kind: and name: fields can be omitted from the YAML. The parser infers:

• kind: skill from the parent directory name (skills/)
• name: from the grandparent directory name (e.g., component-patterns from skills/component-patterns/skill.xcaf)

When kind: or name: are present in the YAML, they must match the inferred values.

Compiled Output

Claude

Output path: .claude/skills/component-patterns/SKILL.md

---
name: component-patterns
description: >-
  Step-by-step procedure for implementing a new React component following
  frontend-app conventions — file layout, prop types, Tailwind styling,
  accessibility attributes, and co-located test scaffolding.
---
# Component Patterns

## When to Use
Invoke this skill whenever you are asked to create a new React component
or refactor an existing one to meet project standards.
...

When artifacts are declared, files discovered in each artifact subdirectory are placed alongside SKILL.md. For example, a skill declaring artifacts: [references] with references/design-tokens.md compiles to .claude/skills/component-patterns/references/design-tokens.md.

Cursor

Output path: .cursor/skills/component-patterns/SKILL.md

Frontmatter emits name, description, disable-model-invocation, paths, and metadata. Provider-specific fields (when-to-use, license, user-invocable, argument-hint, allowed-tools) are omitted with fidelity warnings. Artifact subdirectories are seeded at .cursor/skills/component-patterns/<artifact>/. Example files are collapsed into references/ rather than a separate examples/ subdirectory.

Copilot

Output path: .github/skills/component-patterns/SKILL.md

Emits name, description, and license in the frontmatter. Artifact subdirectories are seeded at .github/skills/component-patterns/<artifact>/. Example files are seeded in an examples/ subdirectory.

Gemini

Output path: .gemini/skills/component-patterns/SKILL.md

Frontmatter is limited to name and description. Provider-specific fields are omitted. Artifact subdirectories are seeded at .gemini/skills/component-patterns/<artifact>/. Example files are collapsed into references/ rather than a separate examples/ subdirectory.

Antigravity

Output path: .agents/skills/<id>/SKILL.md

Antigravity follows the agentskills.io format and emits name, description, when-to-use, argument-hint, allowed-tools, user-invocable, and disable-model-invocation in the frontmatter. The markdown body is preserved.

When artifacts are declared, files discovered in each artifact subdirectory are placed alongside SKILL.md with artifact directory mapping applied (references -> examples, assets -> resources).

Codex (Preview)

Output path: .agents/skills/component-patterns/SKILL.md

Codex shares the .agents/skills/ output path with the Antigravity family. Xcaffold emits name and description in the frontmatter; all other frontmatter fields are stripped. The markdown body is preserved. Artifact subdirectories are placed as-is alongside SKILL.md.

Artifact path remapping

Some providers remap artifact directory names during compilation:

Source directory	Claude	Cursor	Copilot	Gemini	Antigravity	Codex
references/	references/	references/	references/	references/	examples/	references/
scripts/	scripts/	scripts/	scripts/	scripts/	scripts/	scripts/
assets/	assets/	assets/	assets/	assets/	resources/	assets/
examples/	examples/	references/	examples/	references/	examples/	examples/

Provider Fidelity

Skill output is not uniform across providers. The table below summarises what each provider preserves.

Field	Claude	Cursor	Copilot	Gemini	Antigravity	Codex (Preview)
name	yes	yes	yes	yes	yes	yes
description	yes	yes	yes	yes	yes	yes
when-to-use	yes (when_to_use)	no	no	no	yes (when-to-use)	no
license	yes	no	yes	no	no	no
allowed-tools	yes (space-separated)	no	yes (YAML list)	no (warning)	yes (bracket list)	no
argument-hint	yes	no	no	no	yes	no
disable-model-invocation	yes	yes	no	no	no	no
paths	yes	yes	no	no	no	no
metadata	no	yes	no	no	no	no
user-invocable	yes	no	no	no	yes	no
Markdown body	yes	yes	yes	yes	yes	yes
references/ subdirectory	yes	yes	yes	yes	compiled to examples/	yes
examples/ placement	skill root	references/	examples/	references/	examples/	examples/
scripts/ subdirectory	yes	yes	yes	yes	yes	yes
assets/ subdirectory	yes	yes	yes	yes	compiled to resources/	yes