Skip to main content

​
What is Agent Integration?

Spec Kit provides a unified framework for integrating with AI coding agents through standardized slash commands and workflows. This integration allows developers to use their preferred AI assistant while maintaining consistent spec-driven development practices across teams.

​
How It Works

When you initialize a Spec Kit project with specify init, the CLI automatically:
  1. Creates agent-specific directories - Sets up the correct folder structure for your chosen AI agent (e.g., .claude/commands/, .windsurf/workflows/, .github/agents/)
  2. Generates command files - Installs slash commands like /speckit.specify, /speckit.plan, and /speckit.implement in the appropriate format for your agent
  3. Configures script executables - Sets up shell scripts (bash/PowerShell) that agents can invoke to automate repository operations
  4. Establishes project structure - Creates the .specify/ directory with templates, memory, and scripts

​
Agent Categories

Spec Kit supports two primary categories of AI agents:

CLI-Based Agents

Require a command-line tool installation. Examples include Claude Code, Gemini CLI, Cursor, opencode, and Amazon Q Developer CLI. These agents typically execute commands through their CLI interface.

IDE-Based Agents

Built into integrated development environments. Examples include GitHub Copilot (VS Code), Windsurf IDE, IBM Bob, and Kilo Code. These agents work within their native IDE environment.

​
Command File Formats

Different agents use different file formats for their commands:
Used by Claude Code, Cursor, opencode, Windsurf, Codex, Amazon Q, Amp, SHAI, IBM Bob, and Roo Code.Example structure:
  • YAML frontmatter with description field
  • Uses $ARGUMENTS for argument passing
  • Script paths use .specify/ prefix
  • Stored in agent-specific directories (.claude/commands/, .cursor/commands/, etc.)
---
description: "Create feature specifications from requirements"
---

# Specify Command

You are helping create a feature specification...

Execute: bash .specify/scripts/create-new-feature.sh $ARGUMENTS

​
Directory Conventions

Each agent follows specific directory conventions for organizing command files:
Agent TypeDirectory PatternExample
Standard commands.<agent>/commands/.claude/commands/, .cursor/commands/
Workflows.<agent>/workflows/.windsurf/workflows/, .kilocode/workflows/
Prompts.<agent>/prompts/.codex/prompts/, .amazonq/prompts/
GitHub Copilot.github/agents/.github/agents/
GenericUser-specifiedVia --ai-commands-dir flag

​
The Spec-Driven Development Workflow

Regardless of which agent you use, all agents follow the same core workflow:
1

Constitution

Establish project principles and development guidelines using /speckit.constitution
2

Specify

Define requirements and user stories using /speckit.specify
3

Clarify (Optional)

Resolve ambiguities through structured questioning using /speckit.clarify
4

Plan

Create technical implementation plans with your tech stack using /speckit.plan
5

Tasks

Generate actionable task breakdowns using /speckit.tasks
6

Analyze (Optional)

Validate cross-artifact consistency using /speckit.analyze
7

Implement

Execute all tasks to build the feature using /speckit.implement

​
Agent Skills (agentskills.io)

Spec Kit supports the agentskills.io specification for installing reusable agent skills. When you initialize a project with the --ai-skills flag: 
specify init my-project --ai claude --ai-skills
The CLI automatically:
  • Creates an agent-specific skills/ directory (e.g., .claude/skills/, .gemini/skills/)
  • Converts each Spec Kit command into a standalone skill with a SKILL.md file
  • Adds enhanced descriptions and metadata following the agentskills.io spec
  • Enables skill discovery and reuse across projects
Benefits:
  • Portable skills that work across compatible agents
  • Better discoverability through structured metadata
  • Version control and sharing of agent capabilities
  • Compatibility declarations for project requirements

​
Choosing the Right Agent

Consider these factors when selecting an AI agent:
  • VS Code users: GitHub Copilot, Cursor, or Roo Code
  • Standalone CLI: Claude Code, Gemini CLI, opencode, or Amazon Q
  • Specialized IDEs: Windsurf, IBM Bob, or Kilo Code
Choose an agent that your entire team can access and use consistently. This ensures everyone follows the same spec-driven workflow.
  • Cross-platform: Most CLI agents work on Linux, macOS, and Windows
  • Script preference: Choose sh for Unix-like systems or ps for PowerShell environments
  • Custom agents: Use --ai generic --ai-commands-dir <path> for unsupported agents
  • Extended capabilities: Some agents offer additional features beyond slash commands

​
Next Steps

Supported Agents

View the complete list of 15+ supported AI agents with installation links and compatibility notes

Configuration Guide

Learn how to configure projects for different agents and customize the integration

Build docs developers (and LLMs) love

Get started for freeTalk to us