awesome-gemini-cli-subagents
JosephHampton · JosephHampton/awesome-gemini-cli-subagents
Một bộ sưu tập được tuyển chọn gồm 81 sub-agent chuyên biệt, sẵn sàng sản xuất cho Gemini CLI, được thiết kế để cải thiện sự tập trung, cô lập ngữ cảnh và cho phép ủy quyền theo phạm vi công cụ cho các tác vụ lập trình.
A curated collection of 81 production-ready subagents for Gemini CLI. Drop them into .gemini/agents/ and let Gemini delegate the right specialist.
Xem trước
Bạn có thể làm gì với nó
- Ủy quyền các tác vụ lập trình cụ thể cho một sub-agent `code-reviewer` chuyên biệt trong Gemini CLI.
- Quản lý ngữ cảnh hiệu quả hơn bằng cách cho các sub-agent khác nhau làm việc trong các cửa sổ riêng biệt.
- Nhanh chóng cài đặt một bộ 81 sub-agent chuyên gia để tăng cường khả năng của Gemini CLI.
Kiểm tra an toàn
Dùng được, nhưng nên đọc ghi chú trước.
- ●Không có giấy phép rõ ràng: hỏi tác giả trước khi dùng cho sản phẩm thương mại.
Quét tự động các file trong repo. Luôn xem lại code trước khi chạy.
Trích từ README
🤖 Awesome Gemini CLI Subagents
A curated collection of production-ready subagents for Gemini CLI.
81 specialist subagents across 9 domains — drop them into .gemini/agents/ and let Gemini CLI delegate the right expert for the job.
What is this?
Gemini CLI supports subagents: focused, specialist AI assistants that the main agent can delegate to. Each subagent has its own system prompt, its own tools, and its own context window, so it stays sharp on one job instead of trying to be good at everything.
This repo is a curated, ready-to-use library of them. Every agent is a single Markdown file with the official frontmatter format, written to be genuinely useful rather than a one-line stub. Copy the ones you want and go.
Why subagents?
- Focus — a
code-reviewerthat only reviews beats a general assistant that sometimes reviews. - Isolated context — each subagent works in its own context window, so a big task does not pollute your main conversation.
- Tool scoping — a read-only
security-auditorcannot accidentally write files; tools are declared per agent. - Composability — chain them. Let
tech-lead-orchestratorplan, then hand pieces to the specialists.
Quick start
Install everything (user-level, available in every project):
git clone https://github.com/JosephHampton/awesome-gemini-cli-subagents.git
cd awesome-gemini-cli-subagents
./install-agents.sh # copies all agents to ~/.gemini/agents/
Install one agent manually:
# User-level (all projects)
cp agents/quality-testing/code-reviewer.md ~/.gemini/agents/
# Project-level (this repo only, shareable with your team)
mkdir -p .gemini/agents
cp agents/quality-testing/code-reviewer.md .gemini/agents/
Then in Gemini CLI just describe the work — "review this diff for security issues" — and it will delegate to the matching subagent. You can also list and manage them with the /agents command.
How a subagent is defined
Every agent is a Markdown file with YAML frontmatter. The body becomes the agent's system prompt:
---
name: security-auditor
description: Reviews code for security weaknesses and recommends fixes.
kind: local
tools:
- read_file
- grep_search
model: gemini-3-pro-preview
temperature: 0.2
max_turns: 20
---
You are a defensive security reviewer. Your job is to help teams find and
fix weaknesses in their own code before release...
| Field | Required | Purpose |
|---|---|---|
name | ✅ | Unique slug used as the tool name |
description | ✅ | When to use this agent (helps the router pick it) |
kind | local (default) or remote | |
tools | Tool allowlist; supports * and mcp_* wildcards | |
model | Pin a model, or inherit the session's | |
temperature | 0.0–2.0 | |
max_turns | Turn budget for the agent |
See the official Gemini CLI subagents docs for the full reference.
📚 The Collection
- 🏗️ Core Development · 🧠 Language Specialists · ☁️ Infrastructure & DevOps
- ✅ Quality & Testing · 🛡️ Security · 📊 Data, AI & Databases
- 🧰 Developer Experience · 🎯 Specialized Domains · 🧭 Product & Orchestration
🏗️ Core Development
| Agent | What it does |
|---|---|
backend-architect | Scalable backend systems, service boundaries, and data models |
frontend-developer | Accessible, performant UI components and state |
fullstack-engineer | End-to-end vertical feature slices across the whole stack |
api-designer | Clean, consistent REST / GraphQL / gRPC contracts |
microservices-architect | Service decomposition and distributed-system design |
mobile-app-developer | Cross-platform and native mobile features |
graphql-architect | GraphQL schemas, resolvers, and federation without N+1s |
electron-desktop-developer | Secure Electron apps with native-feeling UX |
browser-extension-developer | Manifest V3 extensions for Chrome and Firefox |
realtime-websocket-engineer | WebSocket / SSE systems that survive disconnects |
🧠 Language Specialists
| Agent | What it does |
|---|---|
typescript-pro | Advanced, type-safe TypeScript |
python-pro | Clean, idiomatic, fully typed Python |
go-pro | Idiomatic, concurrent Go |
rust-pro | Safe, idiomatic Rust that works with the borrow checker |
java-pro | Modern Java and the JVM ecosystem |
react-pro | Modern React patterns, hooks, and performance |
nextjs-pro | Next.js App Router, server components, and rendering strategy |
vue-pro | Vue 3 Composition API, Pinia, and Nuxt |
angular-pro | Modern Angular with signals and standalone components |
[svelte-pro](agent |