Creating Skills

Skills are the primary extension mechanism for Gee-Code. A skill is a set of instructions loaded on demand via a slash command.

Quick Start

1. Scaffold the skill

InitSkill(name="my-skill")

This creates:

.gee/skills/my-skill/
├── SKILL.md
├── scripts/
├── references/
└── assets/

2. Write the instructions

Edit .gee/skills/my-skill/SKILL.md:

---
name: my-skill
description: Generates API endpoint boilerplate
activation: /api-gen
args: "[endpoint-name]"
user_invocable: true
---

# API Generator

When invoked, generate a REST API endpoint following the project's patterns.

## Steps

1. Read the existing route files to understand patterns
2. Create a new route file with the endpoint
3. Add request validation
4. Add tests for the new endpoint

## Conventions

- Use the same error handling pattern as existing endpoints
- Follow the project's naming conventions
- Include OpenAPI docstrings

3. Sync to server

SyncSkill(skill_name="my-skill")

The skill is now available as /api-gen.

SKILL.md Frontmatter

Field	Required	Description
name	Yes	Skill identifier
description	Yes	Short description (max 1024 chars)
activation	No	Slash command trigger (default: /name)
args	No	Argument hint shown in help
user_invocable	No	Whether users can invoke directly (default: true)
agents	No	Bundled agent definitions
mcp_servers	No	Required MCP server configurations

Adding Scripts

Python scripts in scripts/ can be executed by the AI during skill execution:

scripts/analyze.py

import sys
import json

def analyze_routes(project_path):
    """Find all route definitions in the project."""
    # ... analysis logic
    return results

if __name__ == "__main__":
    results = analyze_routes(sys.argv[1])
    print(json.dumps(results))

The AI calls it with:

RunSkillScript(skill_name="my-skill", script_name="analyze.py", args="/path/to/project")

Adding References

Reference docs in references/ are available to the AI when the skill is active:

references/
├── api-patterns.md      # Your API conventions
└── error-codes.md       # Standard error codes

These provide domain-specific knowledge the AI can reference during skill execution.

Bundling Agents

Skills can include agent definitions in the frontmatter:

---
name: full-stack
description: Full-stack development skill
agents:
  - name: frontend
    description: React component specialist
    model: claude-sonnet
    system_prompt: |
      You are a React specialist. Focus on component architecture...
  - name: backend
    description: API endpoint specialist
    model: claude-sonnet
    system_prompt: |
      You are a backend specialist. Focus on API design...
---

These agents are created when the skill is installed and available when the skill is active.

Validation

Before syncing, validate your skill:

ValidateSkill(skill_path=".gee/skills/my-skill")

This checks:

• SKILL.md frontmatter is valid
• Python scripts have correct syntax
• Directory structure is correct

Skill Lifecycle

Tool	Description
InitSkill	Scaffold a new skill directory
ValidateSkill	Check skill for errors
SyncSkill	Upload to Gee server
UpdateSkill	Update an existing skill
DeleteSkill	Remove a skill you created
UninstallSkill	Remove a marketplace skill

Next Steps

• Skills & Plugins — overview of the skill system
• Building Plugins — multi-component plugins
• Creating Agents — standalone agent definitions