Best Claude Code Subagents in 2026: Ranked and Explained

The best Claude Code subagent collection in 2026 is wshobson/agents, a plugin marketplace with 194 agents across 88 plugins and 37,400+ GitHub stars as of July 2026. Subagents are specialized AI assistants defined as Markdown files with YAML frontmatter in .claude/agents/; each one runs in its own context window with its own system prompt, tool access, and model. This guide explains how subagents work, ranks the collections and individual subagents worth installing, and shows exactly how to add them to your setup.

What Claude Code Subagents Are

A Claude Code subagent is a task-specific assistant that the main session delegates work to, defined in a Markdown file with YAML frontmatter. Per the official subagents documentation, project subagents live in .claude/agents/ (check them into version control for your team) and personal subagents live in ~/.claude/agents/ (available in every project). Only two frontmatter fields are required:

---
name: code-reviewer
description: Reviews code for quality and best practices
tools: Read, Glob, Grep
model: sonnet
---

You are a senior code reviewer. Analyze code for readability,
performance, and security issues, and report concrete fixes.

The body below the frontmatter becomes the subagent's system prompt. The payoff is context isolation: each subagent runs in its own context window, so a test-runner subagent can chew through thousands of lines of logs and return a three-line summary to your main conversation. Claude decides when to delegate based on each subagent's description, which is why vague descriptions produce subagents that never fire.

Claude Code also ships built-in subagents: Explore for read-only codebase research, Plan for research during plan mode, and general-purpose for multi-step tasks that mix exploration and modification. As of v2.1.198 the /agents command no longer opens a creation wizard; you ask Claude to write the file or edit .claude/agents/ directly, and running sessions pick up file changes within seconds.

The Best Claude Code Subagents, Ranked

This ranking covers the two dominant community collections first, then the individual subagent archetypes that deliver the most value, verified against each repository on July 3, 2026.

1. wshobson/agents: the plugin marketplace (best overall)

wshobson/agents has 37,400+ stars and packages 194 domain-expert agents into 88 granular plugins, alongside 158 skills and 106 commands. It is #1 for two reasons: it installs through Claude Code's native plugin system rather than file copying, and it assigns models by task tier, running architecture and security agents on stronger models while routing operational tasks to Haiku, so you are not paying Opus prices for boilerplate. Coverage spans architecture, languages, infrastructure, security, data, and 16 multi-agent orchestrators.

/plugin marketplace add wshobson/agents
/plugin install python-development

2. VoltAgent/awesome-claude-code-subagents: the curated catalog

VoltAgent/awesome-claude-code-subagents (22,700+ stars) collects 154+ production-focused subagents in 10 categories, from Core Development (11 agents) and Language Specialists (31) to Quality and Security (17). It ranks second because it is the easiest place to read well-written subagent definitions before committing: every file is a plain Markdown definition you can copy into .claude/agents/ and edit. An interactive install-agents.sh script installs selected agents at project or user scope.

3. code-reviewer: the highest-ROI individual subagent

A code-reviewer subagent (both collections ship one) reviews diffs against quality, security, and convention rules in an isolated context. It earns third place because it is the subagent most teams actually use daily: restricted to read-only tools like Read, Grep, and Glob, it cannot modify code, which makes it safe to run on every change before commit.

4. security-auditor: pre-merge vulnerability review

A security-auditor subagent scans for injection risks, secret leaks, auth flaws, and dependency issues, and VoltAgent's Quality and Security category pairs it with a penetration-tester agent. Running it as a separate subagent rather than a prompt matters: a dedicated system prompt keeps the review adversarial instead of politely agreeing with code the main session just wrote.

5. Language specialists: python-pro and friends

Language specialist subagents like VoltAgent's python-pro and backend-developer encode idiomatic patterns, testing conventions, and framework knowledge for one stack. With 31 language specialists in VoltAgent alone, the right move is installing the two or three that match your stack, not the whole category.

6. Infrastructure agents: kubernetes-specialist and devops-engineer

Infrastructure subagents such as kubernetes-specialist and devops-engineer handle manifests, CI/CD pipelines, and deployment reviews. They pair naturally with restricted tool lists (read-only plus Bash) so an agent can inspect a cluster without being able to edit application code.

7. Orchestrators: multi-agent coordination

Orchestrator subagents coordinate other subagents for larger workflows, and wshobson/agents ships 16 of them. Claude Code supports this natively: a subagent's tools field can include Agent(worker, researcher) to let it spawn specific sub-subagents. Treat these as an advanced pick; for parallel sessions that talk to each other, agent teams are usually the better primitive.

How to Install Subagents

There are three install paths, and all of them end with Markdown files Claude Code discovers automatically. First, the manual route: create a file in .claude/agents/ (project) or ~/.claude/agents/ (user), or simply ask Claude Code to "create a code-reviewer subagent" and review what it writes. Both directories are scanned recursively, so you can organize files into subfolders.

Second, the plugin route, which is best for wshobson/agents:

/plugin marketplace add wshobson/agents
/plugin install security-scanning

Third, the installer route for VoltAgent's catalog:

git clone https://github.com/VoltAgent/awesome-claude-code-subagents.git
cd awesome-claude-code-subagents
./install-agents.sh

Two verification notes from the official docs: if Claude cannot find a brand-new subagent, restart the session (this only happens when the agents directory did not exist at startup), and when several scopes define the same name, the more specific scope wins, with project definitions overriding user ones.

Subagents vs Skills vs Hooks

Subagents, skills, and hooks solve different problems, and most mature setups use all three. Subagents isolate work in a separate context; skills inject reusable instructions into the current context; hooks run deterministic shell commands on lifecycle events, as covered in our Claude Code hooks guide and Claude Code rules guide.

The practical heuristic: if the behavior must always happen, use a hook. If it is knowledge Claude should apply inline, use a skill or CLAUDE.md. If it is work you want done off to the side without polluting your conversation, use a subagent.

Limitations and Tradeoffs

Subagents have real costs. Each delegation starts with a clean context, so the subagent re-reads files your main session already knows, which adds latency and tokens; do not delegate trivial lookups. Subagents also cannot spawn unrestricted nested subagents, and plugin-distributed subagents ignore hooks, mcpServers, and permissionMode frontmatter for security reasons, so copy a plugin agent into .claude/agents/ if you need those fields. Installing a 154-agent catalog wholesale is an anti-pattern: every description competes for Claude's delegation decision, and overlapping descriptions cause wrong or missed delegations. Start with three to five agents that match your daily work. Finally, quality varies inside every community collection; read a definition before trusting it with Edit or Bash access.

Related Guides

• Claude Code Hooks: The Complete Guide
• Claude Code Rules: CLAUDE.md Best Practices
• Claude Code Agent Teams Explained
• Best Claude Code Plugins in 2026