Agent Skills for TypeScript & Next.js Developers: Install, Use, and Create Custom Skills (2026)
Practical guide to Agent Skills — the open standard that packages engineering workflows for AI coding agents. Install Matt Pocock skills, understand progressive disclosure, and build production TypeScript/Next.js skills that work across Claude Code, Cursor, and Codex.
By Mussawar Hayat
Why Agent Skills Matter More Than Another Coding Agent
In August 2026 the conversation on X and GitHub shifted from which coding agent is best to how do we give any agent reliable engineering discipline. Agent Skills — folders that contain a SKILL.md file plus optional scripts and references — are the portable answer.
What You Will Learn
- What Agent Skills actually are and how progressive disclosure keeps context cheap
- How to install Matt Pocock production TypeScript skills
- The exact SKILL.md format required by the open standard
- How to write a custom skill for Next.js Server Actions and Prisma
- Security, versioning, and common mistakes
1. What an Agent Skill Is (and Is Not)
An Agent Skill is a directory whose only required file is SKILL.md. That file starts with YAML frontmatter (name + description) and continues with Markdown instructions the agent follows when the skill is active.
The open specification lives at agentskills.io. Anthropic originated the format; it is now adopted by Claude Code, Cursor, Codex CLI, Gemini CLI, and other tools.
Key properties:
- Progressive disclosure — only name and description load at session start (about 30–50 tokens each). Full instructions load when the task matches.
- Composable — multiple skills can activate in the same session.
- Portable — the same SKILL.md works across agents.
- Version-controllable — skills live in the repo and travel with the project.
2. Install Matt Pocock Skills
Matt Pocock's mattpocock/skills collection is the highest-signal set for TypeScript engineers.
Quick install
npx skills@latest add mattpocock/skillsOr inside Claude Code:
claude plugins install mattpocock-skillsThen run the one-time setup:
/setup-matt-pocock-skillsHigh-value skills for Next.js / TypeScript:
/grill-meor/grill-with-docs— adversarial questioning before code/to-prdthen/to-issues— conversation to PRD to vertical-slice issues/tdd— red-green-refactor loop/codebase-design— deep modules and clear seams/diagnose— reproduce, minimise, instrument, fix
3. Anatomy of a Valid SKILL.md
Every skill must follow the open format:
---
name: nextjs-server-actions
description: Enforce thin Server Actions, Zod validation, and a server-only DAL for Next.js App Router. Use when creating or editing Server Actions or mutations.
---
# Next.js Server Actions Skill
## Rules
1. Every Server Action file starts with use server.
2. Keep the action thin: validate with Zod, check session, call a server-only DAL, return a DTO.
3. Never import Prisma directly inside the action file.
4. Always scope queries by the authenticated user id.Frontmatter rules: name is required (lowercase, hyphens), description is the primary trigger — write it so the agent knows when to load the skill.
4. Create a Custom Skill for Your Stack
Highest ROI skills encode patterns your team already fights about.
Step 1 — Create the directory
mkdir -p .agents/skills/nextjs-prisma-dalStep 2 — Write SKILL.md
---
name: nextjs-prisma-dal
description: Enforce the Data Access Layer pattern for Next.js + Prisma. Use when writing Server Actions, API routes, Prisma queries, or ownership checks.
---
# Next.js + Prisma Data Access Layer
## Non-negotiable architecture
- Prisma client lives only in lib/prisma.ts (singleton).
- All database access goes through data/ modules marked server-only.
- Server Actions stay thin: validate, auth, DAL, map to DTO.
## Ownership
Every query that returns user data must include the authenticated user id in the WHERE clause.
## Example
// data/projects.ts
import "server-only"
import { prisma } from "@/lib/prisma"
export async function getProjectForUser(projectId: string, userId: string) {
return prisma.project.findFirst({
where: { id: projectId, ownerId: userId },
select: { id: true, name: true, updatedAt: true },
})
}Step 3 — Verify triggering
Restart the agent and ask it to add a create-project Server Action. A well-written description should cause the skill to load automatically.
5. Production Best Practices
- Keep skills small and focused — one clear workflow per skill.
- Put always-on context in AGENTS.md / CLAUDE.md.
- Version skills with the repo.
- Test the description — under-triggering is more common than over-triggering.
- Prefer progressive disclosure for long reference material.
Security considerations
Skills can contain executable scripts. Treat third-party skill repositories like npm packages: review SKILL.md and any scripts before installing. Prefer known authors (Anthropic, Matt Pocock, Addy Osmani, your own team).
6. Common Mistakes
- Vague descriptions that never trigger.
- Putting everything in one mega-skill.
- Duplicating CLAUDE.md content inside skills.
- Forgetting ownership checks in examples.
- Installing both the plugin and the npx copy.
- Never running the setup skill.
7. Real Use Cases on a Next.js Codebase
- New feature — grill → to-prd → to-issues → tdd with the nextjs-prisma-dal skill keeping mutations safe.
- Bug investigation — diagnosing skill forces reproduce → minimise → instrument before a fix.
- Architecture review — improve-codebase-architecture report, then grill the deepenings.
- Onboarding — new teammates inherit Server Action and Prisma conventions from the shared skill set.
8. FAQ
Do Agent Skills work only with Claude?
No. The format is an open standard. The same SKILL.md works with Claude Code, Cursor, Codex CLI, Gemini CLI, and other tools that implement the specification.
Where should I put skills — personal or project?
Project skills under .agents/skills/ or .claude/skills/ travel with the codebase. Personal skills under ~/.claude/skills/ suit cross-project workflows.
How is a skill different from CLAUDE.md?
CLAUDE.md is always loaded and should stay short. Skills use progressive disclosure and load only when the task matches the description.
What if the agent ignores my skill?
Almost always a description problem. Expand the description with phrases users actually type, then restart the agent.
9. Summary
Agent Skills are the practical layer between raw model capability and reliable engineering output. Install a proven collection such as Matt Pocock's for TypeScript discipline, then write thin, well-described skills that encode your Next.js, Prisma, and auth patterns.
Key Takeaway
A good skill is a portable onboarding guide for the agent: short metadata that always loads, full instructions that load only when needed, and concrete rules that match how your team ships production TypeScript.
Need help wiring AI coding agents into a production Next.js stack?
I help teams design secure Server Actions, Prisma data layers, and agent-friendly project conventions. Get in touch or explore full-stack and AI-assisted development services.
Related reading: Meta Muse Code for TypeScript & Next.js and Secure Server Actions in Next.js 16.
Frequently Asked Questions
Do Agent Skills work only with Claude?
No. The format is an open standard published at agentskills.io. The same SKILL.md is discovered by Claude Code, Cursor, Codex CLI, Gemini CLI, and other tools that implement the specification.
Where should I put skills — personal or project?
Project skills under .agents/skills/ or .claude/skills/ inside the repo travel with the codebase and are ideal for team conventions. Personal skills under ~/.claude/skills/ (or equivalent) suit your own cross-project workflows.
How is a skill different from CLAUDE.md or AGENTS.md?
CLAUDE.md / AGENTS.md is always loaded and should stay short. Skills use progressive disclosure: only name and description load at session start; full instructions load only when the task matches the description.
What if the agent ignores my skill?
Almost always a description problem. Expand the description with the exact phrases users type when the skill should fire, then restart the agent.
Are third-party skills safe to install?
Treat them like npm packages. Review SKILL.md and any scripts before install. Prefer known authors (Anthropic, Matt Pocock, Addy Osmani, your own team) and avoid untrusted marketplaces on machines that hold secrets.
Related guides
Grok Bot gives AI teammates a persistent cloud computer with a browser, filesystem, and terminal. Here is what it is, how it differs from Cursor Cloud Agents and coding agents, and the production rules that keep always-on bots from becoming a liability.
SEO for Google AI Overviews: What Actually Changed in 2026 (And What Still Works)Google AI Overviews and generative search changed how users find answers. SEO is not dead. Here is what Google officially recommends, what GEO hacks to ignore, and how to structure content so it remains visible in both classic results and AI answers.
Building Production Multi-Agent Workflows with the OpenAI Agents SDK in TypeScript (2026 Guide)A practical, production-oriented guide to the OpenAI Agents SDK for TypeScript. Learn agents, tools, handoffs, agents-as-tools, guardrails, and how to orchestrate reliable multi-agent systems for Next.js and Node.js applications.
