Supported AI Agents

Maistik Spec Kit works with 20+ AI coding agents. Claude Code is the recommended default — all examples in this documentation use Claude.

Agent List

Agent	Flag	Status
Claude Code	--ai claude	Recommended
Cursor	--ai cursor-agent	Supported
Gemini CLI	--ai gemini	Supported
Codex CLI	--ai codex	Supported
Windsurf	--ai windsurf	Supported
Amp	--ai amp	Supported
Qoder CLI	--ai qodercli	Supported
GitHub Copilot	--ai copilot	Supported
Roo Code	--ai roo	Supported
Kilo Code	--ai kilocode	Supported
Auggie CLI	--ai auggie	Supported
CodeBuddy CLI	--ai codebuddy	Supported
IBM Bob	--ai bob	Supported
opencode	--ai opencode	Supported
Qwen Code	--ai qwen	Supported
SHAI (OVHcloud)	--ai shai	Supported
Jules	--ai jules	Supported
Amazon Q CLI	--ai q	Partial
Antigravity (agy)	--ai agy	Supported
Generic	--ai generic	Bring your own

Using Claude Code

Claude Code is the recommended AI assistant for Spec Kit. After initialization:

# Initialize
specify init my-project --ai claude

# Open in Claude Code
cd my-project
claude

# Commands are available immediately
/speckit.constitution ...
/speckit.specify ...

Using Other Agents

Each agent gets its commands installed in the appropriate location:

# Cursor
specify init my-project --ai cursor-agent

# Gemini CLI
specify init my-project --ai gemini

# Windsurf
specify init my-project --ai windsurf

Generic / Bring Your Own Agent

For agents not in the list, use the generic option with a custom commands directory:

specify init my-project --ai generic --ai-commands-dir .myagent/commands/

The CLI copies command templates to your specified directory. Your agent needs to support reading Markdown command files from that path.

Agent Limitations

Agent	Limitation
Amazon Q CLI	Does not support custom arguments for slash commands. Commands work but cannot pass inline prompts.

Switching Agents

To add support for a different agent to an existing project:

specify init . --force --ai gemini

This merges the new agent's configuration without overwriting existing specs.

---

Swarm Agent Architecture

When using /speckit.swarm, tasks are dispatched to specialized Claude Code agents that work in parallel. Each agent has a focused responsibility and follows its own operating rules defined in .claude/agents/.

Specialist Agents

Agent	File	Responsibility
Infrastructure Agent	.claude/agents/infrastructure-agent.md	Project scaffolding, configuration files, environment setup, CI/CD, build tooling
Data Agent	.claude/agents/data-agent.md	Database schemas, migrations, ORM models, repositories, seed data, indexes
Logic Agent	.claude/agents/logic-agent.md	Service classes, use cases, domain rules, business validation, error handling
Interface Agent	.claude/agents/interface-agent.md	HTTP routes, CLI commands, request validation, response formatting, middleware
Test Agent	.claude/agents/test-agent.md	Tests derived from spec acceptance criteria, test pyramid (unit/integration/E2E)
Architect Agent	.claude/agents/architect-agent.md	System design, component boundaries, tech selection, data modeling, plan.md creation
Reviewer Agent	.claude/agents/reviewer-agent.md	Spec compliance review, code quality, constitution checks, merge recommendations
Security Agent	.claude/agents/security-agent.md	OWASP Top 10 audit, secrets detection, input validation, boundary analysis
Implementer Agent	.claude/agents/implementer-agent.md	General-purpose code implementation for tasks assigned by /speckit.implement

How Swarm Execution Works

Execution flow:

1. Task classification — Each task is assigned to a specialist agent based on keywords in the task description
2. Wave dispatch — Tasks with no dependencies run simultaneously in Wave 1; dependent tasks wait for later waves
3. Consensus checkpoint — After each wave, conflicts are detected (same file modified by multiple agents), resolved, and incomplete tasks are retried
4. Critical path — The longest sequential dependency chain runs in the main thread with full context for maximum correctness
5. Swarm log — A complete execution record is written to swarm-log.md

Task Classification Rules

Tasks are automatically classified by keyword matching:

Keywords	Agent
setup, initialize, configure, scaffold, env	Infrastructure Agent
model, entity, schema, migration, database	Data Agent
service, business logic, handler, use case	Logic Agent
endpoint, route, controller, CLI, API, view	Interface Agent
test, spec, contract test, integration test	Test Agent

When a task matches multiple patterns, the more specific agent wins (Interface > Logic > Data > Test > Infrastructure).

Installing Agent Definitions

Agent files are installed automatically when you run:

specify init my-project --ai claude

To update agent definitions in an existing project:

specify init . --force --ai claude

The agent files are placed in .claude/agents/ and referenced by /speckit.swarm during parallel execution.