🤖 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-reviewer that 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-auditor cannot accidentally write files; tools are declared per agent.
• Composability — chain them. Let tech-lead-orchestrator plan, 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	Svelte 5 runes and SvelteKit
flutter-pro	Flutter and Dart for fast native-feeling apps
swift-ios-pro	SwiftUI and structured concurrency on iOS
kotlin-android-pro	Kotlin and Jetpack Compose on Android
csharp-dotnet-pro	Modern C# and ASP.NET Core
laravel-pro	PHP and Laravel the way the framework intends
rails-pro	Ruby on Rails with convention over configuration
cpp-pro	Modern C++ with RAII discipline and sanitizers

☁️ Infrastructure & DevOps

Agent	What it does
devops-engineer	CI/CD pipelines and release automation
kubernetes-operator	K8s workloads, manifests, and troubleshooting
terraform-engineer	Modular, safe Terraform / OpenTofu
cloud-architect	Cloud infra on AWS / GCP / Azure with cost in mind
docker-specialist	Lean, secure Dockerfiles and Compose setups
sre-incident-responder	Incident triage and blameless postmortems
aws-specialist	AWS services, IAM, networking, and cost awareness
github-actions-specialist	Fast, secure GitHub Actions pipelines
serverless-architect	Event-driven serverless with idempotency built in
observability-engineer	Logs, metrics, traces, SLOs, and alerts that matter
cicd-pipeline-engineer	Build-once promotion pipelines with tested rollback

✅ Quality & Testing

Agent	What it does
code-reviewer	Thorough, prioritized code review
test-automator	Meaningful unit and integration tests
e2e-tester	Stable Playwright / Cypress end-to-end tests
performance-optimizer	Evidence-based performance fixes
debugger	Systematic root-cause debugging
refactoring-specialist	Safe, behaviour-preserving refactors
load-testing-engineer	Realistic load tests that name the bottleneck
visual-regression-tester	Flake-proof screenshot testing for UI changes

🛡️ Security

Agent	What it does
security-auditor	Defensive code review for common vulnerability classes
dependency-auditor	Third-party CVE and supply-chain review
secrets-detector	Detect committed secrets so they can be rotated and removed
threat-modeler	Design-time STRIDE threat modeling
auth-reviewer	Review authentication and authorization logic
container-security-auditor	Image, Kubernetes, and supply-chain hardening

📊 Data, AI & Databases

Agent	What it does
data-engineer	Reliable ETL / ELT and data pipelines
ml-engineer	Reproducible ML training, evaluation, and deployment
prompt-engineer	Testable LLM prompts and agent instructions
sql-pro	Query writing and performance tuning
postgres-expert	Deep PostgreSQL tuning and features
database-designer	Relational schema and data modeling
analytics-engineer	Trustworthy analytics models and metrics
llm-integration-engineer	LLM product features with evals and cost control
rag-architect	Retrieval-augmented generation that cites its sources
mongodb-expert	Document modelling and aggregation performance
redis-expert	Caching, queues, and realtime data in Redis
mlops-engineer	Reproducible training pipelines and monitored models
data-visualization-engineer	Honest, fast charts and dashboards

🧰 Developer Experience

Agent	What it does
documentation-writer	Clear developer and user documentation
api-documenter	Accurate API reference and OpenAPI specs
git-workflow-manager	Clean Git history and workflow
dependency-upgrader	Safe, staged dependency upgrades
dx-optimizer	Faster builds and a smoother local dev loop
monorepo-manager	Turborepo / Nx / workspaces with trustworthy caching

🎯 Specialized Domains

Agent	What it does
accessibility-auditor	WCAG accessibility audit and fixes
seo-optimizer	Technical and on-page SEO
blockchain-developer	Smart contracts and on-chain safety
game-developer	Gameplay systems and real-time performance
embedded-systems-engineer	Firmware for resource-constrained devices
stripe-payments-integrator	Payment flows and webhooks treated as safety-critical

🧭 Product & Orchestration

Agent	What it does
product-strategist	Scoping, specs, and ruthless prioritization
tech-lead-orchestrator	Decompose and sequence large tasks
code-archaeologist	Understand legacy and unfamiliar codebases
agent-workflow-designer	Design multi-agent workflows and pipelines

---

Contributing

New agents, improvements to existing prompts, and fixes are all welcome. Keep each agent focused, use the standard frontmatter, and write a system prompt you would actually want doing the job. See CONTRIBUTING.md for the short version.

License

CC0 1.0 Universal — public domain. Copy, adapt, and use these agents freely, no attribution required.

If these subagents save you time, a ⭐ helps other Gemini CLI users find them.