All Formats
StarPillar Specification & Hub Directory

Claude Code Skills (SKILL.md) Directory

Modular, on-demand capabilities for Anthropic's Claude Code agent CLI. Loaded dynamically based on natural language task descriptions.

Open Blank Studio Editor
Target: .claude/skills/<skill-name>/SKILL.md
01 / SPECIFICATION & RUNTIME BEHAVIOR

How Claude Code Skills (SKILL.md) Operates in Production

Claude Code Skills are modular capabilities packaged into a SKILL.md file with YAML frontmatter. Claude Code reads the `description` field and intelligently decides when to load the full skill instructions into its context window, or allows manual execution via slash commands (e.g. `/audit` or `/migrate`). Skills allow you to teach Claude custom refactoring workflows, complex auditing protocols, and deployment procedures.

YAML Frontmatter DescriptionSpec Syntax

Claude Code uses the description to semantically match user prompts with available skills.

---
name: codebase-auditor
description: Audit codebases for security vulnerabilities, dead code, and architectural anti-patterns. Trigger when asked to review or evaluate code.
---
Sequential Execution ProceduresSpec Syntax

Step-by-step procedures tell Claude the exact sequence of commands and inspection steps.

## Step-by-Step Procedure
1. Inspect package manifest (package.json, Cargo.toml) to map dependencies.
2. Search for sensitive tokens or unparameterized queries.
3. Categorize findings into CRITICAL, HIGH, and MEDIUM.
02 / REPOSITORY PLACEMENT
.claude/skills/codebase-auditor/SKILL.mdProject Level

Audits codebase for dead code, unhandled async errors, and security vulnerabilities.

~/.claude/skills/personal-style/SKILL.mdGlobal User Level

Applies your personal coding conventions and git commit style across all repositories on your machine.

03 / PRESET SPOKE MATRIX

Available Tech Stack Presets (18)

Click any tech stack below to open the client-side editor preloaded with this verified configuration.

FrontendLaunch

Next.js 15 App Router

Generate a Next.js 15 Claude Code skill to enforce RSCs, Server Actions, and Tailwind CSS v4.

Enforced Invariant:

"Enforce React Server Components (RSC) for data fetching."

FrontendLaunch

React 19 & Vite

Generate a specialized Claude Code skill (SKILL.md) for React 19 & Vite.

Enforced Invariant:

"Use React 19 action hooks (useActionState, useOptimistic, useFormStatus) for async mutations."

BackendLaunch

Python FastAPI & AI

Generate a strict Python FastAPI Claude Code skill prioritizing asynchronous routing and Pydantic validation.

Enforced Invariant:

"Enforce Pydantic v2 BaseModels with Field constraints."

Fullstack & DBLaunch

Supabase & PostgreSQL

Generate a specialized Claude Code skill (SKILL.md) for Supabase & PostgreSQL.

Enforced Invariant:

"Always include `ENABLE ROW LEVEL SECURITY;` on every created table."

Fullstack & DBLaunch

Prisma ORM

Generate a specialized Claude Code skill (SKILL.md) for Prisma ORM.

Enforced Invariant:

"Always use `select` to specify required fields; avoid raw `findMany()` fetching unused relations."

Fullstack & DBLaunch

Drizzle ORM

Generate a specialized Claude Code skill (SKILL.md) for Drizzle ORM.

Enforced Invariant:

"Define schemas using pgTable/sqliteTable with explicit foreign keys and constraints."

BackendLaunch

Go (Golang) & Fiber

Generate a specialized Claude Code skill (SKILL.md) for Go (Golang) & Fiber.

Enforced Invariant:

"Check every `if err != nil` explicitly; never swallow errors or ignore returned values."

BackendLaunch

Rust & Axum

Generate a specialized Claude Code skill (SKILL.md) for Rust & Axum.

Enforced Invariant:

"Never call `.unwrap()` or `.expect()` in production route handlers; propagate errors via `Result<T, AppError>`."

FrontendLaunch

Tailwind CSS v4

Generate a specialized Claude Code skill (SKILL.md) for Tailwind CSS v4.

Enforced Invariant:

"Configure themes directly in globals.css using `@theme` and `@import 'tailwindcss';`."

FrontendLaunch

Vue 3 & Nuxt 3

Generate a specialized Claude Code skill (SKILL.md) for Vue 3 & Nuxt 3.

Enforced Invariant:

"Always use `<script setup lang='ts'>` and the Composition API; ban legacy Options API."

FrontendLaunch

SvelteKit 5 & Svelte 5

Generate a specialized Claude Code skill (SKILL.md) for SvelteKit 5 & Svelte 5.

Enforced Invariant:

"Use Svelte 5 runes: `$state()`, `$derived()`, and `$props()`."

Specialist PersonaLaunch

Universal Cursor Pro

Generate a specialized Claude Code skill (SKILL.md) for Universal Cursor Pro.

Enforced Invariant:

"Deliver surgical, targeted diffs; never replace entire files unnecessarily."

Specialist PersonaLaunch

Pragmatic Rapid Prototyping

Generate a specialized Claude Code skill (SKILL.md) for Pragmatic Rapid Prototyping.

Enforced Invariant:

"Prioritize working software over complex design patterns or deep inheritance hierarchies."

Specialist PersonaLaunch

Security Vulnerability Auditor

Generate a security auditing Claude Code skill focused on identifying vulnerabilities and API key leaks.

Enforced Invariant:

"Inspect source files for leaked API keys, tokens, and private certificates."

DevOps & ToolingLaunch

Docker & CI/CD

Generate a specialized Claude Code skill (SKILL.md) for Docker & CI/CD.

Enforced Invariant:

"Always use multi-stage Docker builds to separate builder tools from the lean production image."

Specialist PersonaLaunch

Vitest & Playwright TDD

Generate a specialized Claude Code skill (SKILL.md) for Vitest & Playwright TDD.

Enforced Invariant:

"Write unit tests covering happy paths and at least two edge/error cases for every new function."

Specialist PersonaLaunch

Automated Code Reviewer

Generate a Claude Code codebase auditing skill to identify structural health, security flaws, and performance anti-patterns.

Enforced Invariant:

"Scan repository manifests to map stack idioms before evaluating logic."

Specialist PersonaLaunch

Multi-Agent System Orchestration

Generate a specialized Claude Code skill (SKILL.md) for Multi-Agent System Orchestration.

Enforced Invariant:

"Define subagent roles: Lead Architect, Frontend Engineer, Backend Specialist, Security Reviewer."

04 / BEST PRACTICES
Write clear, trigger-rich descriptions in the frontmatter so Claude knows exactly when to auto-activate the skill.
Structure skills into discrete phases (Discovery, Analysis, Remediation, Verification).
Include negative constraints (e.g., 'Never execute destructive migrations without user approval').
Add slash command alias support for frequent manual invocations.
Other Format DirectoriesHub Network
05 / FREQUENTLY ASKED QUESTIONS

How does Claude Code discover SKILL.md files?

Claude Code scans the `.claude/skills/<name>/SKILL.md` directory in your current workspace, as well as the global `~/.claude/skills/` directory in your user home folder.

Does Claude load all skills into context at once?

No. Claude only keeps skill metadata (name and description) in its index. The full instruction body is loaded on demand when the task warrants it, keeping context windows lightweight.