← All posts
·4 min read

The Best Claude Code Agents, Skills, and Commands (2026)

I built AgentsCamp because I was tired of copy-pasting the same prompts into Claude Code every week. It's a curated, format-validated catalog of agents, skills, and slash commands you can install with one command. Here's the roundup I'd give a friend who just asked me what's actually worth using.

This isn't a top-ten list ranked by hype. It's grouped by what each thing is for, with the install command next to it.

Agents

Agents are specialized subagents — focused workers Claude can hand a job to without cluttering your main context. The good ones have a clear role and stay in their lane.

API Architect. Designs REST and GraphQL surfaces before you write a line of code. It thinks about resource modeling, versioning, pagination, and error shapes — the stuff you regret skipping later. Reach for it when you're starting a new service and want the contract right the first time.

npx agentscamp add agents/api-architect

Code Reviewer. Reads a diff the way a senior engineer would: naming, edge cases, missing tests, security smells. It's not a linter. It catches the logic mistakes a linter can't, and it's blunt about them.

npx agentscamp add agents/code-reviewer

Browser Agent Engineer. Drives a real browser for end-to-end testing and scraping. When you need Claude to actually click through a flow, fill a form, and verify what renders — not just reason about it — this is the one. It solves the "works in theory, broken in the browser" gap.

npx agentscamp add agents/browser-agent-engineer

ML Engineer. Handles the data-and-model work: feature pipelines, training loops, evaluation harnesses. Useful when you want help that understands the difference between a notebook experiment and something you'd ship.

npx agentscamp add agents/ml-engineer

The pattern across all of them: hand off a bounded task, get back a focused result, keep your main thread clean.

Skills

Skills are capabilities Claude loads on demand. Each one has a short description, and Claude only pulls in the full instructions when your request matches. That means you can install a pile of them and pay the context cost only when you actually use one.

Web Research. Fans out searches, fetches sources, and synthesizes an answer with citations instead of a confident guess. I use it when I need current information and want to see where each claim came from. It's the difference between "I think" and "here's the source."

npx agentscamp add skills/web-research

Changelog Generator. Turns a range of commits into a clean, human-readable changelog. It groups changes by type and writes them for people, not for git log. Saves the half hour I used to spend before every release.

npx agentscamp add skills/changelog-generator

Test Writer. Looks at a function and writes the tests you'd write if you had the patience — happy path, edge cases, failure modes. It's loaded on demand, so it stays out of the way until you're actually writing tests.

npx agentscamp add skills/test-writer

The honest pitch for skills: you teach Claude something once, package it, and never re-explain it again.

Slash commands

Slash commands are the ones you trigger by name. They're for the repetitive moves you make every day, where you want a verb, not a paragraph of instructions.

Commit message. Reads your staged diff and writes a tight conventional-commit message. No more git commit -m "fix stuff". Type the command, get a message that actually says what changed.

npx agentscamp add commands/commit-message

PR review. Pulls a pull request, walks the diff, and leaves structured feedback — what's risky, what's missing, what's fine to ship. Great for the first pass before a human reviewer spends their time.

npx agentscamp add commands/pr-review

Explain code. Point it at a file or function and get a plain-language walkthrough. I lean on it when I land in an unfamiliar codebase and need the map before I touch anything.

npx agentscamp add commands/explain-code

Commands shine for the small, frequent things. They turn a five-line prompt you'd retype constantly into a single word.

How to install

Every install follows the same shape: npx agentscamp add /. The categories are agents, skills, and commands, so:

npx agentscamp add agents/code-reviewer
npx agentscamp add skills/web-research
npx agentscamp add commands/commit-message

The CLI drops each resource into your project in the right place. Because the catalog is format-validated, what you install actually loads — no fixing broken frontmatter or malformed config before you can use it. That validation is the whole reason I built a directory instead of a list of gists.

Start small. Add one agent and one command that match something you do every day, use them for a week, and add more when you hit the next bit of friction. Installing thirty things you never trigger just clutters your setup.


That's the short list I'd actually recommend. There's a lot more in the catalog — hundreds of agents, skills, and commands across categories, plus guides and a glossary if you want to go deeper. Browse the full directory at AgentsCamp, or stay in your terminal and run npx agentscamp list to see everything, or npx agentscamp search to find the one that fits the problem in front of you. Then add it and get back to work.