Agent Skills Examples Library: 15 Paste-Ready Job Patterns

TL;DR: Bloating global system prompts with every possible workflow instruction degrades model reasoning and burns context tokens on tasks where those instructions are irrelevant. The Agent Skills open standard solves this through progressive disclosure. Your coding agent reads only a lightweight name and trigger-dense description at startup, loading the complete instruction body and scripts solely when a relevant task begins. This library delivers fifteen copy-paste SKILL.md patterns spanning software development, security, API generation, and database refactoring compatible with Claude Code, Cursor, and Antigravity as of late 2026. PromptMake at https://promptmake.net/skills generates tailored skill definitions and trigger frontmatter configured for your project structure.

Anatomy of an agent skill across modern runtimes

The Agent Skills standard organizes procedural expertise into portable filesystem bundles. Whether executing in Claude Code under .claude/skills/, Cursor under .cursor/skills/, or Google Antigravity under .agents/skills/, every skill follows an identical directory architecture. Separating metadata from operational instructions allows agents to support hundreds of skills simultaneously without degrading attention. The subsections below explain the frontmatter trigger mechanism and the standard multi-file skill folder layout.

Frontmatter triggers and progressive disclosure

Every skill begins with YAML frontmatter containing two mandatory fields: name and description. The description field functions as the operational trigger. Language models review descriptions during tool planning to decide whether to load the skill into the active context window.

A weak description such as "database helper" rarely activates because it lacks contextual intent. An effective description specifies file extensions, framework names, and user actions: "Generates PostgreSQL migration files, updates Prisma schemas, and writes rollback queries. Use when altering database models or running schema migrations." This precision ensures the skill activates only when you touch relevant files.

Bundled directory layout: scripts, references, and assets

While a single SKILL.md file suffices for simple rules, enterprise skills incorporate supporting folders. An assets/ directory houses boilerplate code templates and standard configuration files.

A references/ directory contains complete API specifications or style guides, which the agent reads on demand rather than storing in memory. A scripts/ directory provides deterministic Python or bash utilities that the agent executes directly, eliminating the need to write repetitive validation logic on the fly.

Fifteen paste-ready skill patterns by domain

The patterns below are organized into three functional categories: core backend engineering, frontend and design systems, and operations and quality assurance. Each pattern provides tested frontmatter and core instruction bodies ready to drop into your agent skill directory.

Domain 1: Backend and API engineering patterns

Pattern 1: REST API Endpoint Scaffolder Frontmatter: ``yaml --- name: rest-endpoint-scaffolder description: Scaffolds TypeScript REST API routes with Zod validation, OpenAPI decorators, and error handling. Use when adding new backend API endpoints or controllers. --- Core instructions: Enforce Zod schemas for all request payloads. Return standard { success, data, error } envelopes. Wrap asynchronous route handlers in catchAsync middleware. Automatically generate a matching integration test in tests/integration/. Pattern 2: Database Migration Reviewer Frontmatter: yaml --- name: db-migration-reviewer description: Audits SQL migration files for destructive operations, missing indices, and lock risks. Use when reviewing schema changes or writing database migrations. --- Core instructions: Flag any DROP TABLE, DROP COLUMN, or destructive ALTER operations. Require concurrent index creation on PostgreSQL. Verify that every foreign key includes an appropriate ON DELETE constraint. Require a reversible down-migration script. Pattern 3: GraphQL Resolver Generator Frontmatter: yaml --- name: graphql-resolver-generator description: Generates typed GraphQL queries and mutations with DataLoader caching. Use when creating GraphQL schemas, resolvers, or DataLoader batches. --- Core instructions: Prevent N+1 database queries by batching relation fetches through DataLoader. Enforce strict authorization checks inside the resolve function before executing database lookups. Return explicit union types for domain errors. Pattern 4: Microservice RPC Client Builder Frontmatter: yaml --- name: rpc-client-builder description: Builds gRPC and tRPC client wrappers with retry policies, circuit breakers, and OpenTelemetry spans. Use when connecting distributed service endpoints. --- Core instructions: Wrap remote calls in exponential backoff retry handlers. Propagate trace headers across service boundaries. Inject mock fixtures into unit test suites for all external RPC stubs. Pattern 5: Background Queue Worker Worker Frontmatter: yaml --- name: queue-worker-scaffolder description: Scaffolds BullMQ and Redis job workers with idempotency locks, dead-letter queues, and telemetry. Use when implementing background queues or asynchronous tasks. --- `` Core instructions: Require unique job IDs for idempotent task handling. Configure maximum three retry attempts before routing to the dead-letter queue. Log job start, completion, and failure metrics with structured JSON payloads.

Domain 2: Frontend, design, and UI architecture patterns

Pattern 6: Accessible Component Architect Frontmatter: ``yaml --- name: accessible-component-architect description: Creates WCAG 2.1 AA accessible React and Tailwind components with Radix UI primitives. Use when building UI components, modal dialogs, forms, or menus. --- Core instructions: Enforce keyboard navigation for all interactive elements. Add visible focus indicators with minimum 3:1 contrast ratios. Include aria-labels on icon-only buttons. Test color pairings against 4.5:1 text contrast standards. Pattern 7: Design Token Extractor Frontmatter: yaml --- name: design-token-extractor description: Converts CSS variables and Tailwind theme configurations into typed design tokens. Use when updating design systems, spacing scales, or color palettes. --- Core instructions: Map primitive values to semantic tokens before assigning component variables. Never allow raw hex codes inside component files. Export type-safe theme definitions for dark mode and light mode. Pattern 8: Responsive Layout Auditor Frontmatter: yaml --- name: responsive-layout-auditor description: Identifies fixed widths, horizontal overflow risks, and touch target violations. Use when reviewing mobile CSS, viewport layouts, or responsive bugs. --- Core instructions: Flag any fixed pixel widths exceeding 320px. Ensure interactive touch targets measure at least 44x44 pixels. Check that mobile navigation drawers trap focus properly when open. Pattern 9: State Machine Form Builder Frontmatter: yaml --- name: state-machine-form-builder description: Implements multi-step form workflows with XState and React Hook Form. Use when building complex wizards, checkout funnels, or stateful forms. --- Core instructions: Model form progression as finite states (idle, validating, submitting, error, success). Persist partial form drafts to sessionStorage. Forbid advancing steps when intermediate validation fails. Pattern 10: Performance Budget Enforcer Frontmatter: yaml --- name: performance-budget-enforcer description: Audits bundle sizes, dynamic imports, and image optimization flags in Next.js applications. Use when optimizing page load speed or Core Web Vitals. --- `` Core instructions: Require next/image with explicit width and height dimensions to prevent Cumulative Layout Shift. Lazy-load heavy third-party libraries using dynamic imports with SSR disabled. Ban synchronous script tags in page layouts.

Domain 3: Testing, security, and operational patterns

Pattern 11: Unit Test Coverage Generator Frontmatter: ``yaml --- name: unit-test-coverage-generator description: Writes Vitest and Jest unit test suites with boundary tests, mock stubs, and error assertions. Use when adding tests for utility functions or services. --- Core instructions: Follow the Arrange-Act-Assert structure. Test happy paths, boundary inputs (null, undefined, empty strings, max integer), and expected error throws. Never mock the module under test. Pattern 12: Secret and Credential Scanner Frontmatter: yaml --- name: secret-credential-scanner description: Scans git diffs and code proposals for hardcoded API keys, private tokens, and passwords. Use before committing changes or during pre-merge reviews. --- Core instructions: Match against regex signatures for AWS keys, GitHub personal tokens, Stripe secrets, and private keys. Stop the agent immediately if a secret is found and output a remediation warning. Pattern 13: Semantic Release Notes Drafter Frontmatter: yaml --- name: release-notes-drafter description: Generates user-facing changelog entries from conventional git commits. Use when cutting a new release, tagging versions, or updating CHANGELOG.md. --- Core instructions: Group updates into Features, Bug Fixes, and Breaking Changes. Translate technical commit messages into plain, customer-facing benefit statements. Link commit SHAs and pull request IDs. Pattern 14: Docker Container Optimizer Frontmatter: yaml --- name: docker-container-optimizer description: Optimizes Dockerfiles using multi-stage builds, non-root users, and layer caching. Use when writing container configurations or production builds. --- Core instructions: Enforce multi-stage build architectures to keep production images minimal. Run application processes under an unprivileged non-root user. Order instructions from least-frequently changed to most-frequently changed. Pattern 15: Prompt Security Hardener Frontmatter: yaml --- name: prompt-security-hardener description: Hardens LLM system prompts against direct and indirect prompt injection attacks. Use when building AI chatbots, agent interfaces, or prompt templates. --- `` Core instructions: Separate system instructions from untrusted user data using explicit delimiter tags. Forbid the model from revealing system instructions or executing unauthorized tool calls. Implement strict refusal boundaries.

Best practices for structuring `SKILL.md` files

A skill is only as reliable as its internal constraints. High-performing skills establish clear boundaries that prevent agents from overstepping their mandate.

Defining strict non-goals and refusal lines

Every skill should feature a dedicated Non-Goals section. If an agent specialized in API scaffolding encounters an architectural request to rewrite the database ORM, it must decline and notify the developer.

Adding explicit refusal statements such as "Do not alter database models, run migrations, or install external npm packages without user consent" prevents scope creep and keeps diffs small and reviewable.

Linking reference docs instead of pasting manuals

Keep your primary SKILL.md under two hundred lines. When a workflow requires extensive documentation—such as complete API schemas or internal coding style manuals—place those files inside the references/ subfolder.

Reference the file within the main skill: "When configuring AWS S3 clients, consult references/s3-policy-guidelines.md for bucket permission requirements." The agent reads the reference file only when that specific sub-task arises, preserving context tokens.

Common mistakes when building agent skills

Writing vague description frontmatter tops the list of failure modes. If your description does not clearly describe the triggering scenario, the model will overlook the skill and rely on generic training defaults.

Creating monolithic skills that attempt to manage entire applications is another frequent error. Break complex workflows into specialized, composable skills. A team is better served by separate skills for database migrations, API routing, and frontend forms than by a single mega-skill.

Failing to update skills as dependencies evolve leads to broken generations. If your team migrates from Prisma to Drizzle, update your database skill immediately to reflect new import paths and schema idioms.

Assuming PromptMake deploys or executes your agent skills is also incorrect. PromptMake is an optimization platform that helps you design, validate, and structure your SKILL.md definitions and frontmatter. You save the generated skill files directly into your repository's .claude/skills/, .cursor/skills/, or .agents/skills/ directory.

Step-by-step workflow: deploying a new skill to your repository

Follow this repeatable process to create and activate custom agent skills across your team.

1. Identify a repetitive engineering workflow: Choose a task your team executes frequently, such as creating CRUD endpoints or writing component tests.
2. Document the standard operating procedure: Write out the exact architectural steps, required libraries, and quality checks.
3. Craft precision trigger frontmatter: Define a descriptive name and write a trigger-rich description highlighting key file patterns.
4. Format the skill in PromptMake: Open https://promptmake.net/skills to validate the YAML frontmatter and format instruction sections according to progressive disclosure standards.
5. Save the bundle in your repository: Create .claude/skills/<skill-name>/SKILL.md or .agents/skills/<skill-name>/SKILL.md and commit it to git.
6. Test activation in your agent CLI: Prompt the agent with a task matching your description and verify that the skill loads into context.
7. Iterate based on agent output: Refine constraints, add non-goals, or supply code templates in assets/ if the agent drifts from team conventions.

FAQ

What is an agent skill in modern AI coding tools?

An agent skill is a modular folder containing a SKILL.md file that teaches AI coding assistants like Claude Code, Cursor, and Antigravity how to execute specific tasks. Skills use progressive disclosure, meaning the model only loads full instructions when a relevant task triggers them.

Where should I store agent skills in my repository?

Store skills in .claude/skills/<skill-name>/ for Claude Code, .cursor/skills/<skill-name>/ for Cursor, or .agents/skills/<skill-name>/ for Google Antigravity. For personal skills available across all projects, store them in your home directory, such as ~/.claude/skills/.

How does the agent know when to trigger a skill?

The agent evaluates the description field in the skill's YAML frontmatter during planning. If the user prompt or active file paths match the triggering intent described in the frontmatter, the runtime loads the full SKILL.md instructions into active memory.

Can agent skills include executable scripts?

Yes. Skills can include a scripts/ directory containing bash, Python, or Node.js scripts. The SKILL.md file can instruct the agent to execute these scripts directly to perform deterministic checks, format code, or validate schemas.

Does PromptMake host an agent skill marketplace?

No. PromptMake does not host a public skill marketplace or run background agents. PromptMake provides prompt engineering and skill generation tools that help developers build, validate, and format clean SKILL.md files for their own codebases.

What is the difference between an agent skill and a system prompt?

System prompts are static instructions loaded into every conversation turn, consuming context tokens regardless of relevance. Agent skills use progressive disclosure, remaining dormant in your filesystem until specifically activated by a matching user task.

How long should a typical `SKILL.md` file be?

Keep SKILL.md files focused and under one hundred to two hundred lines. For detailed technical manuals, comprehensive schemas, or large code templates, use supporting references/ and assets/ subfolders within the skill directory.