gsAgents - Multi-Provider Agent Compiler

A unified agent and skill compilation system that transforms universal JSON configurations into provider-specific formats for Claude Code, OpenCode, GitHub Copilot, and Gemini CLI.

Repository Structure

gsAgent/
├── agents/                 # Universal agent JSON configurations
├── skills/                 # Universal skill JSON configurations
├── compiler/               # Compilation engine and templates
│   ├── compile.py          # Main compilation script
│   ├── templates/          # Provider-specific templates
│   └── schema/            # JSON schemas
├── output/                 # Compiled provider-specific outputs
├── .github/workflows/      # GitHub Actions for deployment
└── AGENTS.md              # Universal instructions

Quick Start

1. Initialize deployment structure:

   ./init-deploy.sh

2. Add your agent configurations:

   • Place agent JSON files in agents/ directory
   • Place skill JSON files in skills/ directory

3. Compile and validate:

   python3 compiler/compile.py --validate
   python3 compiler/compile.py --all

4. Deploy:

   • Commit and push to trigger GitHub Actions
   • Agents will be compiled and deployed to separate branches

Compilation Process

The compiler transforms universal JSON configurations into provider-specific formats:

Unified JSON Format

{
  "name": "security-auditor",
  "description": "Security-focused code auditor",
  "tools": ["read", "grep", "glob", "bash"],
  "permissions": {
    "edit": "deny",
    "bash": "ask"
  },
  "model": "sonnet",
  "color": "red",
  "temperature": 0.5,
  "providers": {
    "claude": true,
    "opencode": true,
    "copilot": true,
    "gemini": false
  }
}

Provider-Specific Output

Claude Code (.claude/agents/)

---
name: security-auditor
description: Security-focused code auditor
tools: [Read, Grep, Glob, Bash]
color: red
---

[Prompt content]

OpenCode (.opencode/agents/)

---
description: Security-focused code auditor
tools:
  read: true
  grep: true
  glob: true
  bash: true
model: sonnet
temperature: 0.5
---

[Prompt content]

GitHub Copilot (.github/agents/)

---
name: security-auditor
description: Security-focused code auditor
tools: ['read', 'grep', 'glob', 'bash']
---

[Prompt content]

GitHub Actions Workflow

The repository includes a CI/CD pipeline that:

1. Triggers on push to main branch
2. Compiles agents for each provider (claude, copilot, opencode)
3. Deploys to separate branches:
   • agents-claude - Claude Code agents
   • agents-copilot - GitHub Copilot agents
   • agents-opencode - OpenCode agents
4. Creates release artifacts for each provider

Supported Providers

• Claude Code - .claude/ directory
• OpenCode - .opencode/ directory
• GitHub Copilot - .github/ directory
• Gemini CLI - .gemini/ directory (skills only)

Commands

# Validate all configurations
python3 compiler/compile.py --validate

# Compile all agents and skills
python3 compiler/compile.py --all

# Compile specific agent
python3 compiler/compile.py --agent security-auditor

# Compile for specific provider
python3 compiler/compile.py --provider claude --all

# Compile only agents
python3 compiler/compile.py --agents-only

# Compile only skills
python3 compiler/compile.py --skills-only

File Naming Conventions

• Agents: agents/{agent-name}.json
• Skills: skills/{skill-name}.json
• Compiled agents: output/.{provider}/agents/{agent-name}.{ext}
• Compiled skills: output/.{provider}/skills/{skill-name}/SKILL.md

Provider-Specific Notes

• Claude: Uses PascalCase for tools, supports color and model
• OpenCode: Uses YAML format with boolean values for tools
• Copilot: Uses quoted strings for tools, supports handoffs and MCP servers
• Gemini: Skills only, no agent support

Deployment Branches

Each provider gets its own deployment branch:

• agents-claude - Claude Code agents
• agents-copilot - GitHub Copilot agents
• agents-opencode - OpenCode agents

Requirements

• Python 3.7+
• pip packages: jinja2, schema
• GitHub repository with Actions enabled

License

This repository follows the standard open-source licensing model for AI agent configurations.