DEV Community: Gearscope

How to Set Up AI Agent Skills: A Practical Guide

Gearscope — Sat, 23 May 2026 17:27:15 +0000

You have been using an AI coding agent for a while now. Maybe Claude Code, maybe Cursor, maybe Codex CLI. Things work fine, but you keep seeing people talk about "agent skills" and you are not sure what that means or where to start. This post is for you.

What Are Agent Skills?

An agent skill is a markdown file that teaches your AI coding assistant how to handle a specific kind of task. Think of it like a detailed instruction manual you hand to a new hire. Instead of explaining your deployment process every single time, you write it down once and the agent reads it whenever it needs to.

Skills live in a file called SKILL.md. That file has YAML frontmatter at the top (name, description, a few metadata fields) and then structured instructions in plain markdown below. The format is defined by the Agent Skills Specification, an open standard originally developed by Anthropic and now community-governed.

When you install a skill, you are really just dropping a folder with a SKILL.md file into a location your agent knows to look at. Some skills are a single file. Others ship with supporting assets like templates, config files, or scripts.

How to Install Them

Installation is straightforward. Here is the basic pattern:

Find the skill repo. Most skills live on GitHub. Clone or download the repo.
Copy the skill folder into your agent's skills directory. For Claude Code that is typically .claude/skills/ in your project root. For Cursor it is .cursor/skills/. Each agent has its own convention, so check your agent's docs.
Restart your agent session or reload skills. Some agents pick up new skills automatically, others need a restart.

Some skill packs include an installer. Antigravity Awesome Skills, a community collection of 1,464 SKILL.md playbooks, ships an npm installer that drops skills into the right place for you.

If you want to go beyond manual installs, SkillHub from iFlytek is basically an npm registry for agent skills. It has a CLI for publishing and installing, plus a web UI and RBAC if you are managing skills for a team.

The SKILL.md Format

Every skill follows the same basic structure. Here is a minimal example:

---
name: my-deploy-skill
description: "Deploys the current project to staging"
allowed-tools:
  - Bash
  - Write
---

# Deploy to Staging

## When to use
When the user asks to deploy to staging.

## Steps
1. Run the test suite and confirm all pass.
2. Build the production bundle.
3. Push to the staging branch.
4. Monitor the deploy logs for errors.

## Constraints
- Never deploy if tests fail.
- Always ask before pushing to production.

The frontmatter handles the metadata. The body handles the instructions. The spec supports optional fields like compatibility (which agents the skill targets), license, and metadata for custom key-value pairs.

One important detail: the allowed-tools field. This tells the agent which tools it is allowed to use when running this skill. It is both a permission boundary and a hint. If your skill should never touch the network, do not list tools that make HTTP requests.

Which Skills to Start With

This is where most people get decision paralysis. There are hundreds of skill packs out there. Here are the ones worth your time based on what we have actually tested.

For general software engineering

Addy Osmani's Agent Skills is a 23-skill framework that covers the full software development lifecycle. Spec writing, planning, implementation, testing, shipping. It scored 5 out of 5 for quality and documentation in our review. If you only install one pack, make it this one.

Superpowers by Jesse Vincent takes a different approach. It enforces a disciplined TDD-first methodology across 14 interlocking skills. Your agent will brainstorm before building, write specs, create implementation plans, and test as it goes. We rated it 4 out of 5.

For official vendor skills

Anthropic's official skills repo has 17 production and example skills including the same DOCX, PDF, PPTX, and XLSX generators that power Claude.ai's document features.

Amazon Web Services ships 43 skills covering CDK, CloudFormation, Bedrock, EC2, VPC, and more. If you work on AWS, this pack is essential. Documentation scored a perfect 5.

Vercel's agent skills cover deploying to Vercel, cost optimization, and Next.js best practices. Eight focused skills from the team that built the platform.

For research and memory

last30days is a research skill that searches Reddit, X, YouTube, TikTok, Hacker News, and a dozen other platforms, then synthesizes what it finds into a structured brief. We gave it a 5 out of 5.

Engram gives your agent persistent memory across sessions. It is a single Go binary backed by SQLite that exposes an MCP server, HTTP API, CLI, and TUI. Also scored 5 out of 5 across every dimension.

For breadth

If you just want a huge library to browse, Antigravity Awesome Skills has 1,464 skills. Quality varies since it is community-maintained, but breadth is the whole point. We rated it 4 out of 5 with a "TRY IT" verdict.

Common Pitfalls

Installing too many skills at once

This is the number one mistake. You find a pack with a thousand skills, install the whole thing, and suddenly your agent is confused about which instructions to follow. Skills are prompts. More prompts means more context, which means more potential for conflicting instructions.

Start with two or three focused packs. Add more only when you have a specific need.

Not checking what a skill actually does

Skills are just markdown files. Read them before you install them. A skill that asks your agent to run arbitrary shell commands or make network requests could be risky if you do not trust the source.

This is also where Snyk Agent Scan comes in. It crawls your local machine for agent configs, MCP server definitions, and skill files, then checks them for prompt injection, tool poisoning, and supply-chain threats. Run it after installing skills you do not fully trust.

Ignoring compatibility

Not every skill works with every agent. The compatibility field in SKILL.md frontmatter tells you which agents the author tested against. A skill written for Claude Code might reference tools that Codex CLI does not have. Check before you install.

Skipping the spec

If you are going to write your own skills, read the Agent Skills Specification first. It is short, practical, and will save you from inventing your own conventions.

Forgetting to update

Skills are software. They get bug fixes, new instructions, better prompts. If you cloned a repo six months ago, you might be running an outdated version. Add a reminder to pull the latest from upstream every few weeks.

Wrapping Up

Agent skills are not complicated. They are markdown files that teach your AI assistant how to handle specific tasks. The SKILL.md format is open, well-specified, and supported by every major agent platform. Start with Addy Osmani's pack or Anthropic's official repo. Add vendor-specific packs if you work on AWS or Vercel. Install last30days and Engram if you want research and memory. Do not over-install. Read what you install. Run Snyk Agent Scan if you are cautious.

That is really the whole thing. The skill ecosystem is moving fast but the fundamentals are stable. Get the basics right and everything else follows.

Guide sourced from Gearscope, where we test agent tooling so you do not have to guess.

MCP Servers for AI Agents: Which Ones Are Worth Installing

Gearscope — Sat, 23 May 2026 17:25:49 +0000

If you are building anything with AI agents this year, you have probably bumped into MCP servers by now. The Model Context Protocol went from a curious spec to something your agents actually depend on, and the server ecosystem has exploded. Last count there were hundreds of MCP servers floating around GitHub, npm, and PyPI. Most of them are half-finished weekend projects.

I pulled review data from the Gearscope catalog to figure out which MCP servers are genuinely worth installing. Every pick below was tested hands-on or in a sandbox, rated across quality, documentation, setup, value, and ecosystem fit. No hand-waving.

Here are six MCP servers that earned their spot.

1. Engram

Rating: 5/5 | Verdict: KEEP IT
Author: Gentleman Programming (Alan Buscaglia)
License: MIT

GitHub

Engram is a persistent memory system for AI coding agents. It ships as a single Go binary backed by SQLite with full-text search, and it exposes four interfaces: MCP server (stdio transport), HTTP API, CLI, and an interactive TUI.

When your agent finishes a session, Engram captures what happened and stores it so the next session can pick up where things left off. That sounds simple, but it solves one of the most annoying problems with agent workflows: context amnesia between runs.

Strengths:

Single binary, zero config drama. Download and run.
Perfect 5/5 scores across every review dimension: quality, docs, setup, value, ecosystem fit.
SQLite with FTS5 means search is fast and reliable.
Four interfaces mean you can use it however you want.

Weaknesses:

Partial functional verification in testing. The core works, but edge cases around concurrent access patterns need more real-world stress testing.
Stdio-only MCP transport. If you need SSE transport for a remote setup, you will need to wrap it yourself.

If you run agents locally and are tired of them forgetting everything, Engram is the first thing to install.

2. Laravel Boost

Rating: 5/5 | Verdict: KEEP IT
Author: Laravel (Taylor Otwell)
License: MIT

GitHub

Laravel Boost is the official Laravel MCP server. It lives inside your Laravel application and gives agents direct access to your database schema, query logs, application info, and documentation search. It ships 10 MCP tools and 4 prompts, with auto-config for 11 different AI agents.

This is what vendor-backed MCP tooling should look like. Instead of a generic code assistant guessing at your Eloquent models, Boost hands the agent your actual schema, your actual routes, and your actual error logs.

Strengths:

Dead simple setup. Composer install, one command, done. 5/5 setup score.
5/5 quality, value, and ecosystem fit. It does exactly what it promises.
Auto-config means it works with Claude Code, Codex, Cursor, and eight more agents out of the box.
Real production tooling from the Laravel team, not a community prototype.

Weaknesses:

Laravel only. If you are on Django, Rails, or anything else, this does nothing for you.
Documentation is good but not exhaustive at 4/5. Some advanced tool configurations could use more examples.

If you build with Laravel, stop reading and go install this. It is not optional anymore.

3. bb-browser

Rating: 4/5 | Verdict: TRY IT
Author: epiral
License: MIT

GitHub

bb-browser is an MCP server and CLI that connects AI agents to your actual Chrome browser via the Chrome DevTools Protocol. Instead of spinning up headless browsers or fighting anti-bot captchas, it runs JavaScript inside browser tabs where you are already logged in. It ships 143 community platform presets.

The idea is straightforward: your browser becomes the API. No API keys, no headless Chrome, no anti-bot detection. Your agent just uses your existing session.

Strengths:

5/5 value. Using your real login session removes an entire class of integration headaches.
143 platform presets means it already knows how to interact with most sites you care about.
No API keys to manage. Your cookies and sessions are the auth.

Weaknesses:

Setup is 3/5. Chrome DevTools Protocol configuration is not always smooth.
Security model requires thought. You are giving an agent access to your logged-in browser.

bb-browser is worth trying if you need agents to interact with web apps that do not have APIs.

4. Agent Toolkit for AWS

Rating: 4/5 | Verdict: KEEP IT
Author: Amazon Web Services
License: Apache-2.0

GitHub

Amazon's official agent toolkit ships 43 skills (13 core, 30 specialized) covering CDK, CloudFormation, Bedrock, EC2, VPC networking, and data analytics. It also includes a managed MCP server that works with Claude Code, Codex, and Kiro.

This is the real deal for AWS shops. The depth of reference material outclasses every third-party AWS skill pack I have seen.

Strengths:

5/5 documentation. AWS clearly invested in making the reference material thorough and accurate.
43 skills covering everything from infrastructure-as-code to data analytics.
Managed MCP server means no manual server process to babysit.

Weaknesses:

AWS-only. If you are on GCP or Azure, this is not for you.
You need valid AWS credentials configured before anything works.

If AWS is your cloud, this toolkit is a no-brainer install.

5. Snyk Agent Scan

Rating: 4/5 | Verdict: KEEP IT
Author: Snyk (formerly Invariant Labs)
License: Apache-2.0

GitHub

Snyk Agent Scan is not an MCP server that gives your agent new capabilities. It is a security scanner built for the agent ecosystem. It crawls your machine for MCP server definitions, SKILL.md files, and tool descriptions, then checks everything against a catalog of 15+ issue types: prompt injection, tool poisoning, supply-chain threats, and more.

This is the kind of tool you did not know you needed until you realize you are installing random MCP servers from GitHub without auditing them first.

Strengths:

5/5 setup. Install, run, get results. As simple as security tooling gets.
Actually connects to your MCP servers, retrieves their tool manifests, and inspects what they expose.
Addresses a real and growing problem as MCP adoption increases.

Weaknesses:

It tells you what is wrong but does not fix anything. You still need to understand the findings.
New issue types are being added regularly, so coverage is a moving target.

Install Snyk Agent Scan alongside whatever MCP servers you pick up. Think of it as the antivirus for your agent stack.

6. Browserbase Skills

Rating: 4/5 | Verdict: KEEP IT
Author: Browserbase
License: MIT

GitHub

Browserbase Skills is the official skill pack from Browserbase, a cloud browser infrastructure provider. It contains 13 skills for browser automation, web scraping, serverless browser functions, UI testing, trace capture, cookie sync, and company research.

Think of this as the production-grade sibling to bb-browser. Where bb-browser uses your local Chrome session, Browserbase gives you cloud browsers that scale.

Strengths:

5/5 documentation and 5/5 value. The skills are well-written and cover real workflows.
Covers the full browser automation pipeline from basic fetch to adversarial UI testing.

Weaknesses:

Requires a Browserbase account and API key. This is a cloud service, not a self-hosted tool.
4/5 quality. Some skills are tighter than others.

If you need production browser automation at scale, this is the better pick over bb-browser.

Quick Comparison

Server	Rating	Best For	Setup	Cost
Engram	5/5	Persistent agent memory	Dead simple	Free, self-hosted
Laravel Boost	5/5	Laravel project context	Dead simple	Free, self-hosted
bb-browser	4/5	Local browser control	Moderate	Free, self-hosted
AWS Agent Toolkit	4/5	AWS infrastructure	Moderate	Free, needs AWS creds
Snyk Agent Scan	4/5	Security auditing	Dead simple	Free, open source
Browserbase Skills	4/5	Cloud browser automation	Easy	Needs Browserbase account

The MCP server space is moving fast. New servers show up weekly, and the quality range is wild. The six above are the ones that held up under actual testing. Install what fits your stack, skip what does not, and always scan before you trust.

Reviews sourced from Gearscope, where we test agent tooling so you do not have to guess.

7 Agent Skill Packs That Actually Make AI Coders Better

Gearscope — Sat, 23 May 2026 17:20:30 +0000

Agent skill packs are everywhere now. Open GitHub, search for "agent skills," and you will drown in repos promising to turn your AI coder into a 10x engineer. Most of them are just rehashed prompt templates that add noise, not signal.

We spent the last month hands-on with dozens of these packs for Gearscope. The seven below are the ones we kept installed after testing. Each one solves a real problem, not a hypothetical one.

1. Addy Osmani Agent Skills

Verdict: The full software development lifecycle, packed into 23 portable skills.

Key strength: This is not a grab bag of tips. Addy Osmani (yes, the Google Chrome team guy) built a complete engineering methodology that walks your agent from idea refinement through spec writing, planning, implementation, and shipping. Each skill chains into the next one. The craft is obvious.

Who should use it: Teams who want their AI agent to follow a real process instead of just vomiting code. If you have ever watched an AI coder skip the thinking and jump straight to breaking things, this pack fixes that.

2. Engram

Verdict: The most complete agent memory layer you can install right now.

Key strength: Your AI coder forgets everything between sessions. Engram fixes that. It is a single Go binary backed by SQLite with full-text search, and it gives your agent persistent memory across conversations. It exposes an MCP server, HTTP API, CLI, and even a TUI. When your agent finishes a session, it can store what it learned and pick up where it left off next time.

Who should use it: Anyone doing multi-session work with an AI coder. If you are tired of re-explaining your project architecture every time you open a new chat, Engram is the answer. It scored a perfect 5 across every review dimension we track.

3. Antigravity Awesome Skills

Verdict: The biggest skill pack on GitHub. Breadth is the whole point.

Key strength: 1,464 SKILL.md playbooks covering everything from brainstorming to security auditing to React component scaffolding. It works with Claude Code, Cursor, Codex CLI, Gemini CLI, and more. The editorial bundles group related skills together so you are not lost in a sea of options. An npm installer gets you going fast.

Who should use it: Developers who want coverage over depth. If you work across a lot of different stacks and want a skill ready for whatever weird task comes up, this is your library. Quality varies across 1,400+ skills, but the hits outweigh the misses.

4. Agent Toolkit for AWS

Verdict: The AWS-built skill pack every cloud-bound agent needs.

Key strength: Amazon shipped 43 skills (13 core, 30 specialized) covering CDK, CloudFormation, Bedrock, EC2, VPC networking, and data analytics. This is not community cosplay. It is the real reference material from the people who build AWS, formatted as installable agent skills. Three plugin bundles and a managed MCP server round it out.

Who should use it: Anyone deploying on AWS. If your AI coder is touching CloudFormation templates or CDK stacks without this pack, it is flying blind. The documentation quality is top tier, which matters because cloud infrastructure is one area where guessing gets expensive.

5. last30days

Verdict: The research engine that searches where Google cannot.

Key strength: This skill searches Reddit, X, YouTube, TikTok, Hacker News, Polymarket, and 10+ other platforms within a configurable time window. It scores results by real engagement metrics, then synthesizes a structured brief. Most AI research tools just glue together a few API calls. This one actually understands what "recent and relevant" means.

Who should use it: Developers who need market research, competitive analysis, or trend tracking as part of their workflow. If you build products and need to know what people are actually saying about a topic in the last month, not what an old training data snapshot thinks, install this. It scored a perfect 5 in our review.

6. Superpowers

Verdict: A disciplined engineering workflow that forces AI coders to think before they type.

Key strength: 14 interlocking skills that enforce brainstorm-before-building, spec writing, implementation planning, and test-driven development across eight different agent harnesses. It works with Claude Code, Codex CLI, Cursor, Gemini CLI, GitHub Copilot CLI, and others. The methodology is cohesive: each skill picks up where the previous one left off.

Who should use it: Developers who treat AI coders as pair programmers, not autocomplete on steroids. If you want your agent to follow TDD, write specs, and plan before coding, Superpowers lays down that structure.

7. Browserbase Skills

Verdict: The most complete agent skill pack for production browser automation.

Key strength: 13 skills from Browserbase covering web scraping, UI testing, cookie sync, serverless browser functions, and company research. The skills range from basic page fetching all the way to adversarial UI testing. Everything works through the browse CLI, and the documentation is unusually good for an infrastructure tool.

Who should use it: Anyone doing browser automation at any scale. If your agent needs to interact with web pages that do not have a nice API, this pack gives it the tools to do the job properly.

Quick Comparison

Pack	Skills	Rating	Best For
Addy Osmani	23	4/5	Engineering methodology
Engram	1 (memory system)	5/5	Persistent agent memory
Antigravity	1,464	4/5	Broad coverage
AWS Toolkit	43	4/5	Cloud infrastructure
last30days	1 (research)	5/5	Multi-platform research
Superpowers	14	4/5	TDD-first workflow
Browserbase	13	4/5	Browser automation

All seven are free and open source. Install what fits your workflow, skip what does not.

Reviews sourced from Gearscope, where we test agent tooling so you do not have to guess.