Last Verified: October 11, 2026 — installed on Claude Code 2.1.296, 43 agents loaded, 2 run on a real project
VoltAgent/awesome-claude-code-subagents is a free, MIT-licensed collection of ready-made Claude Code subagents: 163 of them on October 11, in 10 departments.
Each department installs as one plugin. Below: the three to install first, what you’ll see when it worked, a real run of two agents working at the same time, and fixes for the usual errors.
Quick Start (paste into your terminal)
Adds the collection, then installs developers, reviewers/testers, and the organizer. Restart Claude Code afterwards.
claude plugin marketplace add VoltAgent/awesome-claude-code-subagents
claude plugin install voltagent-core-dev
claude plugin install voltagent-qa-sec
claude plugin install voltagent-metaType these in a terminal, not in the Claude Code chat
These are shell commands: run them in PowerShell, Terminal or your editor’s terminal, outside a Claude Code session. Pasted into the chat box, Claude just reads them as a message.
Already inside Claude Code? Type /plugin to open the plugin manager instead.
On this page (10 sections)
What a Subagent Actually Is
A subagent is a Markdown file with a name, a one-line job description, a tool list and (usually) a model.
When Claude hands a task to one, that agent works in a fresh context window with its own instructions, then reports back a summary. Your main conversation only sees the result, not every file it read along the way.
Claude picks one on its own when a task matches the description, or you name it. These 163 files are well-written job descriptions, so you don’t have to write your own.
Install It, Step by Step
Step 1. Add the VoltAgent marketplace
~20 sec
This downloads the list of plugins. Nothing is installed yet. Needs Git and an internet connection.
claude plugin marketplace add VoltAgent/awesome-claude-code-subagentsYou should see:
Clone complete, validating marketplace…
✔ Successfully added marketplace: voltagent-subagents
Step 2. Install three departments
~30 sec
core-dev = the developers, qa-sec = reviewers, testers and debuggers, meta = the organizer agents that plan who does what. 43 agents in total.
claude plugin install voltagent-core-dev
claude plugin install voltagent-qa-sec
claude plugin install voltagent-metaEach line should end with:
✔ Successfully installed plugin: voltagent-qa-sec@voltagent-subagents (scope: user)
✔ Successfully installed plugin: voltagent-meta@voltagent-subagents (scope: user)
scope: user means the agents work in every project, not just this folder.
Step 3. Check they’re switched on
~10 sec
claude plugin listAll three should show Status: ✔ enabled:
Status: ✔ enabled
❯ voltagent-meta@voltagent-subagents
Status: ✔ enabled
❯ voltagent-qa-sec@voltagent-subagents
Status: ✔ enabled
Step 4. Restart Claude Code and ask
~1 min
Close any open Claude Code session, start a new one with claude, and ask:
List the voltagent agents you can use, grouped by plugin.In our test it listed all 43, named like voltagent-qa-sec:code-reviewer. That plugin:name form is the agent’s full name.
Your First Run: Two Agents at Once
Open Claude Code in a project folder and paste this (swap in your own file names). Both agents are in the departments you just installed:
In parallel: have the code-reviewer agent review src/server.js (real bugs only), and have the test-automator agent write a test for the /health endpoint in test/health.test.js. Then summarize what each agent did.Sample run · Claude Code 2.1.296, Sonnet, October 2026
Input used: the prompt above, on a small Node project with one server file and no tests.
code-reviewer read src/server.js, changed nothing, and found three bugs: the /health text didn’t match what the project notes promised, /health?x=1 returned 404 because the route compared the raw URL, and any HTTP method got the health response.
test-automator created test/health.test.js with two tests. The 404 test passed; the /health test failed on exactly the mismatch the reviewer had flagged.
The two agents agreed without talking to each other: the failing test pointed at the same line the review did. Each worked in its own context, so the main chat only got the two summaries.
More Prompts to Try
Name one agent
Use the code-reviewer agent to review my uncommitted changes. List real bugs only, no style nits.Use the debugger agent: npm test fails with the error below. Find the cause before changing anything.
[paste the error]Let the organizer pick the team
Use agent-organizer to plan which agents should build and test a /orders endpoint, and in what order. Then run that plan.What the organizer agents really do
agent-organizer and multi-agent-coordinator write a plan (who does what, in what order). Your main Claude session is the one that launches the agents.
That’s why the prompt above ends with Then run that plan.
All 10 Departments
| Plugin | Agents | Examples |
|---|---|---|
voltagent-core-dev | 13 | backend-developer, frontend-developer, api-designer, mobile-developer |
voltagent-lang | 30 | python-pro, typescript-pro, nextjs-developer, rust-engineer, sql-pro |
voltagent-infra | 16 | devops-engineer, kubernetes-specialist, terraform-engineer, sre-engineer |
voltagent-qa-sec | 18 | code-reviewer, debugger, penetration-tester, test-automator |
voltagent-data-ai | 13 | data-engineer, llm-architect, prompt-engineer, postgres-pro |
voltagent-dev-exp | 16 | refactoring-specialist, documentation-engineer, git-workflow-manager |
voltagent-domains | 16 | payment-integration, fintech-engineer, game-developer, seo-specialist |
voltagent-biz | 18 | product-manager, sales-engineer, legal-advisor, landing-page-copywriter |
voltagent-meta | 12 | agent-organizer, multi-agent-coordinator, context-manager |
voltagent-research | 11 | market-researcher, competitive-analyst, trend-analyst |
Add one later the same way: claude plugin install voltagent-lang, then restart Claude Code.
The department most people skip
voltagent-biz: product manager, sales engineer, legal advisor, UX researcher and landing-page copywriter. Useful even if you never write code, e.g. a PM agent turning meeting notes into a spec.
Which to Add Next
| If you mostly… | Add |
|---|---|
| Work in one language or framework | voltagent-lang (30 agents, Python to Swift) |
| Run servers and deploys | voltagent-infra |
| Do product, marketing or research work | voltagent-biz + voltagent-research |
| Work with data or LLM apps | voltagent-data-ai |
Don’t install all ten “just in case”
Every installed agent’s description is shown to Claude so it can choose between them, and that costs context in every session. Our token-usage measurements show how these small always-on costs add up.
Remove a department any time: claude plugin uninstall voltagent-lang.
If Something Goes Wrong
| You see | Why | Fix |
|---|---|---|
claude: command not found | Claude Code isn’t installed, or the terminal was open before you installed it | Install Claude Code, then open a new terminal window |
| Clone or marketplace error in Step 1 | Git is missing, or no network access | Install Git, check your connection, run Step 1 again |
| Claude says it has no such agent | The session started before you installed | Exit Claude Code and start a new session |
| Status shows disabled in Step 3 | The plugin was switched off | claude plugin enable voltagent-qa-sec (use the right name) |
| Claude does the job itself instead | Your request didn’t clearly match an agent | Name it: use the code-reviewer agent, or the full voltagent-qa-sec:code-reviewer |
| A test run never finishes | The test started a server that never closes | Ask the agent to close the server after the test (ours had to) |
Which Model Each One Uses
Most agent files pin a model, so cheap jobs don’t burn your best model. On October 11: 109 use Sonnet, 19 use Haiku (docs, git workflow, project management and similar lighter work) and 26 inherit whatever your main session uses.
To change one, copy its file into ~/.claude/agents/ and edit the model: line in your copy.
Prefer Single Files Over Plugins?
Copy just the agents you want into ~/.claude/agents/ (all projects) or .claude/agents/ (one project). They then appear without the plugin prefix. The repo also ships an interactive install-agents.sh for macOS and Linux.
FAQ
Is it free?
Yes. The collection is MIT-licensed and free. Agents run on your normal Claude Code plan, so their work counts toward your usage like any other request.
Is it 161 or 163 agents?
The count keeps growing. Our Short said 161; the repo had 163 agent files on October 11. The marketplace description still says 158.
Do subagents see my whole conversation?
No. A subagent starts fresh with only the task Claude gives it, plus its own instructions. That keeps your main chat clean, and it’s also why a vague hand-off gets a vague result.
Are they safe?
They are plain Markdown instructions, not code, and each one lists the tools it may use. Read a file before you rely on it, the same as any prompt you’d copy from the internet.
Can I write my own?
Yes: start from any VoltAgent file and save your version in ~/.claude/agents/. Related: Skills and hooks and other settings are the other ways to teach Claude Code a repeatable job.