The Best Claude Code Agents, Skills & Commands to Install First

Start with five: the code-reviewer and debugger agents (delegate review and diagnosis), the conventional-commits skill and create-pr command (polish the git loop), and one domain specialist for your stack. Add the test-runner pattern, security-auditor, and workflow skills as friction appears. Everything here is a Markdown file — copy it in, restart, done.

The fastest upgrade to a stock Claude Code setup isn't a prompt trick — it's installing a few well-built extensions. Everything below comes from this site's library, is a plain Markdown file, and installs by copy-paste. Here's the starter kit we'd give a new teammate, by workflow.

The first five

code-reviewer (agent) — fresh-eyes review of every diff for correctness, security, and maintainability, in its own context. The single highest-value delegate: it reads what you've stopped seeing.
debugger (agent) — hand it the failing test or stack trace; it reproduces, traces the root cause, and reports — noise stays in its window.
conventional-commits (skill) — staged changes become clean Conventional Commits messages, every time.
create-pr (command) — /create-pr pushes the branch and opens a PR with a real title and description drafted from the diff.
One stack specialist — match your daily language: react-specialist, python-pro, typescript-pro, sql-pro, or terraform-specialist.

The second wave, by friction

Tests feel like a chore → write-tests command + test-scaffolder skill, and fix-failing-test for the red ones.
Security review keeps slipping → security-auditor agent + security-scan on diffs; secret-scanner before pushes.
Errors eat your mornings → explain-error — paste the trace, get the diagnosis and fix.
Branches drift → sync-branch and the branch-rebaser skill for conflict-walking rebases.
Docs rot → readme-generator and update-readme, grounded in the actual repo.

The Claude Code power tools

The library also extends Claude Code itself: hook-writer turns "always run prettier after edits" into an enforced hook; claude-settings-auditor reviews a repo's checked-in config before you trust it; plugin-scaffolder packages your setup for the team; setup-claude-ci wires the GitHub Action.

Install once, then make them yours

Mechanics in one breath: agents are files in ~/.claude/agents/, skills are ~/.claude/skills/<name>/SKILL.md, commands are ~/.claude/commands/ — use a repo's .claude/ instead to share with the team, and start a new session to load. (Full walkthrough.) Then treat every install as a draft: tighten the description so delegation fires when you'd want it, restrict tools to the job, and rewrite house rules into the body. The library's real product isn't the files — it's working examples of the craft, pre-installed.

Continue exploring

system-architect — Use this agent for high-level system design — service boundaries, data flow, scaling, trade-offs.
CLI Tooling Engineer — Use this agent to design or build a command-line tool — subcommand and flag layout, --help and error UX, exit codes, --json/machine output, config precedence,…

Frequently asked questions

How do I install a Claude Code agent or skill from AgentsCamp?

Copy the file. Agents are single Markdown files for ~/.claude/agents/ (or .claude/agents/ in a repo to share with the team); skills are folders with a SKILL.md under ~/.claude/skills/<name>/; commands go in ~/.claude/commands/. Every item page has the install path and a copy button; start a new session and it's live.

How many should I install?

Fewer than you think. Subagent descriptions are read during delegation routing, so dozens of overlapping agents blur each other. A working set of five to eight focused items — review, debugging, your stack's specialist, your git loop — outperforms a hoard. Add when you feel friction, prune what never fires.

Can I customize them?

That's the point — they're plain Markdown with a frontmatter header and a system prompt. Edit the description to tune when Claude delegates, restrict tools, pin a model tier, and rewrite the body for house rules. The library entries are starting points engineered to be edited.