Setting up AI DevKit for opencode

Learn how to set up AI DevKit for opencode so your team can use the same operating model across AI coding agents.

AI DevKit adds project-local setup, memory, communication, workflow skills, and verification to opencode so every project follows the same playbook.

• Project initialization and environment selection
• AI editor files like rules, agents, and commands
• Memory, communication, verification, and skills for repeatable workflows AI DevKit is a control plane for AI coding agents. It gives supported tools like Cursor, Claude Code, Codex, Gemini CLI, opencode, Pi, GitHub Copilot, and others one setup model, one console, local-first memory, cross-agent communication, workflow skills, and verification.

Use it when your AI coding setup has grown from one assistant into multiple agents, terminals, config files, and memory gaps. Initialize the project once, then give every supported coding agent the same operating model.

Workflow skills are part of AI DevKit, but they are not the whole product. The bigger idea is one operating layer for setup, supervision, communication, local-first memory, and verification across the agents you already use.

When working with AI coding agents, you often find yourself:

• Managing multiple terminal sessions with no shared control surface
• Copy-pasting prompts, logs, and test output between agents
• Repeating the same instructions across sessions
• Losing context between conversations
• Struggling to maintain consistency across tools and features

AI DevKit solves these problems by giving your coding agents:

• One local setup — .ai-devkit.json reconciles supported agent files from one source of truth
• Agent console — See and supervise running local agent sessions
• Agent send — Route prompts, logs, and stdin to the right running agent
• Local memory — Decisions and patterns that persist across sessions
• Workflow skills — Reusable process guidance tailored to multi-agent coding
• Skills — Reusable instruction packs your agents can load for domain-specific work
• Plugins — Optional npm packages that add CLI commands
• Verification gates — Fresh evidence before completion claims

Before you begin, make sure you have:

• Node.js (version 20.20.0 or higher)
• npm or npx (comes with Node.js)
• At least one supported AI coding agent or environment, such as Cursor, Claude Code, Codex, Gemini CLI, opencode, Pi, GitHub Copilot, Devin, or Antigravity

Install AI DevKit globally using npm:

npm install -g ai-devkit

Or use it directly with npx (no installation required):

npx ai-devkit@latest init

Navigate to your project directory and run:

npx ai-devkit@latest init

You'll be prompted to select which AI environments you use (Cursor, Claude Code, etc.). AI DevKit will then:

1. Create workflow docs — A configured AI docs directory, docs/ai/ by default, for requirements, design, planning, implementation, and testing
2. Set up AI environment files — Configuration, skills, and MCP servers where supported
3. Save your preferences — Stored in .ai-devkit.json for future updates

After initialization, open the local agent console:

Start one supported coding agent in this project first, such as Claude Code, Codex, Gemini CLI, or opencode. Then run:

ai-devkit agent list
ai-devkit agent console

Then send one task or log stream to a running agent:

ai-devkit agent send "summarize current branch and test status" --id <agent-name>
npm test 2>&1 | ai-devkit agent send --id <agent-name> --stdin

This is the core loop: see your agents, send work to the right session, and keep memory and verification available when the task needs them.

After initialization, you'll have workflow docs your agent can use as durable context. The default path is docs/ai/; projects can customize it in .ai-devkit.json.

docs/ai/
├── requirements/    # What you're building and why
├── design/          # Architecture and technical decisions
├── planning/        # Task breakdown and timeline
├── implementation/  # Implementation notes and guides
├── testing/         # Test strategy and cases
├── deployment/      # Deployment procedures
└── monitoring/      # Monitoring and observability

This structure gives your coding agents a clear handoff between phases instead of relying on chat history.

AI DevKit installs skills into your AI environment. Skills are reusable capability packs the agent can load when your request matches a workflow or domain.

Terminal commands still start with ai-devkit; skills are used by the agent inside your coding assistant.

Skill	Purpose
dev-lifecycle	Orchestrate worktree setup, requirements, design, planning, implementation, testing, and review
dev-worktree, dev-requirements, dev-design, dev-planning, dev-implementation, dev-testing, dev-review, dev-pr	Run focused lifecycle and publish-for-review phases directly
dev-commit	Commit only intended, verified changes with a conventional message
tdd	Add or change behavior test-first
verify	Require fresh command output before completion claims
structured-debug	Debug issues with reproduction, hypotheses, fixes, and verification
document-code	Document and understand existing code
memory	Store and retrieve reusable project knowledge

For detailed usage, see Development with AI DevKit.

Here's how a typical workflow might look:

1. In your terminal:
   $ npx ai-devkit@latest init
   $ ai-devkit agent console

2. In your AI coding agent:
   > Use the dev-lifecycle skill to start requirements for user authentication with OAuth

   AI: "What feature would you like to build?"
   You: "Add user authentication with OAuth"

   AI guides you through requirements -> design -> planning -> implementation -> verification -> review

1. Explore your AI editor — Ask the agent to use dev-lifecycle on a small feature
2. Open the agent console — Operate running agents
3. Read the workflows guide — Development with AI DevKit
4. Set up memory — Give your AI long-term memory
5. Install skills — Extend your AI's capabilities or browse AI coding agent skills
6. Install plugins — Add optional CLI commands

• Check the Supported Agents page for environment-specific setup
• Browse the Roadmap to see what's coming
• Open an issue on GitHub for bugs or questions

Once AI DevKit is set up in opencode, follow the getting started guide and ask the agent to use the dev-lifecycle skill so it plans before editing code.