Awesome Claude Code Subagents
The 130+ subagent collection plus 8 specialists picked from it — K8s, SEO, Security, Smart Contract, GraphQL, API Architect, Search, Prompt Engineer.
What's in this pack
Claude Code subagents are markdown files in .claude/agents/ that define a focused expert. The host invokes one when the task matches its description — letting you compose a long workflow out of small, single-purpose agents instead of one giant CLAUDE.md. The subagent ecosystem exploded in 2025; this pack gives you a curated entry point.
| # | Asset | What it does |
|---|---|---|
| 1 | awesome-claude-code-subagents | The 130-agent reference collection (VoltAgent) |
| 2 | k8s-specialist | YAML-aware Kubernetes manifests + kubectl-fluent diffs |
| 3 | seo-specialist | On-page SEO audits, schema, and meta-description writing |
| 4 | security-auditor | OWASP-aware code review with severity-classified findings |
| 5 | smart-contract-auditor | Solidity / Move review for re-entrancy, access control, gas |
| 6 | graphql-architect | Schema design, federation, N+1 detection |
| 7 | api-architect | REST + OpenAPI design with versioning patterns |
| 8 | search-specialist | Anthropic-style multi-tool search composition |
| 9 | prompt-engineer | Prompt revision and self-critique |
Why curate from 130
Most engineers cloning the full awesome-list get overwhelmed: a 130-agent dropdown is worse than no agents. Claude Code's selection algorithm picks the best-matching agent by description, so a directory full of overlapping specialists creates noise. We picked 8 that map to common production tasks where the focused agent demonstrably beats vanilla Claude Code.
The full collection still has its place — when you need a niche (Cobol modernization, Stripe integration, etc), grep the 130 first before writing your own. But for the day-to-day, the 8 specialists are the workbench.
Install in one command
tokrepo install pack/claude-code-subagents
# Or pick individuals
tokrepo install seo-specialist
tokrepo install security-auditor
tokrepo install graphql-architect
The TokRepo CLI writes each subagent to .claude/agents/<name>.md in your current project. They're loaded automatically the next time you launch Claude Code.
How to invoke a subagent
You don't have to call them by name. Claude Code's subagent dispatch reads the description frontmatter and picks the best match for your task. To force a specific one, prefix:
> use the security-auditor subagent and review the changes since main
> use the graphql-architect subagent to design a federated schema for our 3 services
When this pack alone isn't enough
Pair with Boris Cherny Files for the workflow-glue subagents (Ralph Wiggum loop, code-simplifier, verify-app), or Anthropic Builders for the 17 official skills.
9 assets in this pack
Frequently asked questions
What's the difference between a Claude Code subagent and a slash command?
A subagent is a focused expert with its own system prompt; Claude Code dispatches to it automatically based on the task description. A slash command is a parameterized prompt you invoke explicitly with /name. Subagents are for capability composition; slash commands are for shortcuts. Both live in .claude/ and can be combined.
Will these 8 subagents work outside Claude Code?
Subagents are Claude Code's specific abstraction. The TokRepo CLI maps them to the closest equivalent in other tools — Cursor .mdc rules, Codex CLI AGENTS.md, etc — but the auto-dispatch behavior is Claude-Code-specific. For a true multi-host setup, expect to invoke them by name in non-Claude-Code tools.
Is the awesome-claude-code-subagents collection officially endorsed by Anthropic?
It's a community project (VoltAgent), not Anthropic-owned. Anthropic has published a few official subagents (code-simplifier, search-specialist) — those are also in TokRepo, separately. The 130-agent collection is the de-facto starting point most engineers use.
How does Claude Code pick which subagent to use?
Claude Code reads the description field in each subagent's frontmatter and matches it against the user's task. If multiple match, it picks the most specific. You can force selection with 'use the
Will adding 8 subagents bloat my context?
No. Subagent definitions only enter context when Claude Code dispatches to one. Sitting in .claude/agents/ they cost nothing. The 130-collection has the same property — even 130 agents cost zero tokens until invoked.
12 packs · 80+ hand-picked assets
Browse every curated bundle on the home page
Back to all packs