Building Plugins

Plugins are the most powerful extension mechanism in Gee-Code — a single plugin can bundle multiple skills, custom agents, slash commands, and MCP server configurations.

Plugin Structure

.gee/plugins/my-plugin/
├── plugin.json              # Metadata (optional)
├── README.md                # Documentation (optional)
├── .mcp.json                # MCP server configs (optional)
├── skills/
│   └── my-skill/
│       ├── SKILL.md         # Skill instructions
│       ├── scripts/         # Python helpers
│       ├── references/      # Reference docs
│       └── assets/          # Binary assets
├── agents/
│   └── reviewer.md          # Standalone agent
└── commands/
    └── deploy.md            # Slash command

Step by Step

1. Create the directory structure

mkdir -p .gee/plugins/my-toolkit/skills/analyze/
mkdir -p .gee/plugins/my-toolkit/agents/
mkdir -p .gee/plugins/my-toolkit/commands/

2. Add a skill

Create .gee/plugins/my-toolkit/skills/analyze/SKILL.md:

---
name: analyze
description: Analyze code quality and suggest improvements
agents:
  - name: code-analyzer
    description: Specialized code analysis agent
    model: sonnet
    system_prompt: You are a code quality expert...
---

# Analyze Skill

When activated, analyze the codebase for:
1. **Security issues** — SQL injection, XSS, auth flaws
2. **Performance** — N+1 queries, unnecessary allocations
3. **Style** — Inconsistent naming, dead code

3. Add helper scripts

Create .gee/plugins/my-toolkit/skills/analyze/scripts/scan.py:

"""Security pattern scanner."""
import re, sys, json

PATTERNS = {
    "sql_injection": r"f['\"].*SELECT.*{.*}",
    "eval_usage": r"\beval\s*\(",
    "hardcoded_secret": r"(password|secret|key)\s*=\s*['\"][^'\"]+['\"]",
}

def scan_file(filepath):
    with open(filepath) as f:
        content = f.read()
    findings = []
    for name, pattern in PATTERNS.items():
        for match in re.finditer(pattern, content):
            findings.append({"type": name, "line": content[:match.start()].count('\n') + 1})
    return findings

if __name__ == "__main__":
    print(json.dumps(scan_file(sys.argv[1]), indent=2))

4. Add a standalone agent

Create .gee/plugins/my-toolkit/agents/reviewer.md:

---
name: reviewer
description: Code review specialist
model: sonnet
system_prompt: |
  You are a thorough code reviewer. Focus on:
  - Logic errors and edge cases
  - Security vulnerabilities
  - Performance implications
  Always provide specific line references and suggested fixes.
---

5. Add MCP server config (optional)

Create .gee/plugins/my-toolkit/.mcp.json:

{
  "servers": [
    {
      "name": "project-db",
      "command": "npx",
      "args": ["-y", "@modelcontextprotocol/server-postgres"],
      "env": { "DATABASE_URL": "postgresql://localhost:5432/mydb" }
    }
  ]
}

6. Sync to server

SyncPlugin(plugin_name="my-toolkit")

This uploads all components in one operation.

Plugin Search Paths

Plugins are discovered in two locations:

1. Project-level: .gee/plugins/{name}/ — specific to this project
2. Global: ~/.gee-code/plugins/{name}/ — available everywhere

Validation

Before syncing, validate individual skills within the plugin:

ValidateSkill(skill_path=".gee/plugins/my-toolkit/skills/analyze")

Tips

• Start with a skill, then promote to a plugin when you need agents or commands
• Bundle related agents in skill frontmatter when they’re tightly coupled to a skill
• Use references for stable domain knowledge, scripts for dynamic processing
• Binary assets in assets/ are automatically base64-encoded during sync

Next Steps

• Creating Skills — skill development guide
• Creating MCP Servers — add external tool capabilities
• Validation — code validation setup