Gemini CLI Subagents

A curated collection of 145+ specialized subagents and skills for the Gemini CLI, designed to handle everything from core development and infrastructure to business strategy and research.

Each agent in this library has been hardened for security, following the principle of least privilege and including anti-jailbreak directives.

🚀 Installation

Option 1: Global Installation (Recommended)

Clone this repository and copy the agents and skills to your global Gemini directory:

git clone https://github.com/eddiehaug/gcli-agents.git
cd gcli-agents

# Install agents
cp -r agents/* ~/.gemini/agents/

# Install skills
mkdir -p ~/.gemini/skills
cp -r skills/* ~/.gemini/skills/

Option 2: Use the Subagent Catalog

You can search and fetch agents directly from within Gemini CLI using the subagent-catalog skill included in this repo.

1. Install the skill: cp -r skills/subagent-catalog ~/.gemini/skills/
2. In Gemini CLI, ask: "Search the catalog for security agents" or "Fetch the definition for python-pro"

---

📚 Categories

The library is organized into 10 logical categories:

1. Core Development - Backend, frontend, fullstack, and mobile experts.
2. Language Specialists - Deep expertise in Python, TypeScript, Rust, Go, .NET, and more.
3. Infrastructure - DevOps, Kubernetes, Terraform, and Cloud (AWS/GCP/Azure).
4. Quality & Security - Security auditors, penetration testers, and QA automation.
5. Data & AI - Data scientists, ML engineers, and LLM architects.
6. Developer Experience - Tooling, documentation, and workflow optimization.
7. Specialized Domains - Fintech, Blockchain, Healthcare, and IoT.
8. Business & Product - Product managers, technical writers, and scrum masters.
9. Meta-Orchestration - Multi-agent coordination and state management.
10. Research & Analysis - Market researchers, trend analysts, and literature searchers.

---

🛡️ Security Hardening

All agents in this repository have been audited and hardened:

• Principle of Least Privilege: Purely analytical agents do not have shell access.
• Anti-Jailbreak Directives: Every agent contains strict instructions to refuse malicious requests or persona switching.
• Human-in-the-loop: Critical infrastructure agents (DBA, Terraform, K8s) are mandated to pause and request confirmation before destructive actions.

---

📖 Subagent Structure

Each subagent uses the Gemini CLI Markdown format:

---
name: agent-name
description: "Brief description for auto-discovery"
tools: [read_file, grep_search, ...]
model: gemini-3.1-pro-preview  # Optimized for the task
skills:
  - relevant-skill-name
---

You are an expert...

Model Selection

Agents are pre-configured to use the most efficient model for their role:

• Gemini 3.1 Pro: Used for complex reasoning, security audits, and architectural planning.
• Gemini 3.1 Flash: Used for fast execution, coding assistance, and documentation.
• Gemini Deep Research: Used for multi-step information retrieval, trend analysis, and evidence synthesis across scientific and market domains.

---

📄 License

MIT License. See LICENSE for details.