Skip to main content

Spec-Driven AI Workflows for Flutter

You're already using AI to write Flutter code. Now make it write code you trust. ACT is a structured workflow from spec to PR — where problems are caught early, not in the code review.

Watch the Demo Buy Now

AI coding tools are fast. But fast at what, exactly?

AI agents are great at following patterns and writing lots of code — when you give them the right structure. Without it, they infer missing requirements, make product decisions, and guess at your conventions, often going off-track. AI is a multiplier — of whatever process you give it. The speed is real. The confidence isn't.

"My codebase started consistent. Six weeks later, it feels like a house of cards."

AI follows local context. When conventions and decisions are not documented, patterns drift as features pile up.

"I spend more time reviewing AI code than writing it myself."

The AI produces something, but you can't trust it without reading every line. That's not faster — it's just different work.

"Great news: 100% passing tests. Small detail: half of them are disabled."

AI agents optimize for completion. They will chase green checks with brittle, shallow, or disabled tests that fail to protect real behavior.

"I gave it a vague prompt and got exactly what I deserved."

Without a proper spec, the AI fills in the blanks with guesses. And you don't discover the wrong assumptions until the PR review.

"The code works, but it doesn't look like Flutter."

AI agents write code that compiles. They don't write code that follows Flutter conventions — widget composition, proper state management, theme extensions, idiomatic patterns. You end up fighting the agents or rewriting by hand.

"Every session starts from zero."

You solved this exact problem last week. But the context is gone and the AI has no memory of what worked. Without a way to capture and reuse insights, every session reinvents the wheel.

Now imagine a different workflow

You describe what you want to build. The AI asks questions (rather than guessing) and writes a detailed specification, then a second pass tears it apart — finding gaps and wrong assumptions before a single line of code is written.

Then it builds phase by phase, committing as it goes. Tests lock in behavior and keep future runs aligned. And when you're done, you capture the insights so the next session starts smarter.

What if your AI agent actually worked this way?

Introducing the

Agentic Coding Toolkit

Agentic Coding Toolkit (ACT) is a collection of commands, agents, skills, and a Flutter knowledge base that structures your AI-assisted development workflow — built to make spec-driven development practical, not theoretical.

ACT is fully open-source readable (markdown commands, skills, and knowledge files). It installs in two commands and works with any Flutter or Dart project — no modifications to your project structure.

Works with Claude Code and OpenCode — the two leading AI coding CLIs. You'll need one of these installed, along with your preferred LLM subscription or API key.

See it in Action

Watch ACT take a feature request through spec, plan, and execution, with each step committed along the way.

Five commands. One predictable workflow

From spec to PR with guardrails at every step

1

Spec

You describe the feature. The AI asks clarifying questions and helps you iterate until all user flows, edge cases, and error scenarios are covered.

2

Refine

Optional but recommended

A second pass roasts the spec — looking for gaps, wrong assumptions, and misalignment with your existing codebase.

3

Plan

Turn the spec into a detailed multi-phase plan with actionable checklists, ready for implementation. Optional research agents gather Flutter patterns and inspect your codebase conventions.

4

Work

The AI executes the plan phase by phase — making commits as it goes, running tests continuously, and creating a PR when done.

5

Compound

Optional but recommended

Capture reusable insights and decision patterns from the session. Next time, the knowledge is already there.

Everything your AI agent needs to write code you trust

Six pillars that transform AI-assisted Flutter development from chaotic to predictable

Spec-Driven Development

AI works from detailed specs, not vague prompts. Each spec includes user flows, edge cases, and error scenarios — then gets adversarially reviewed for gaps before planning begins.

Structured Workflows

A repeatable cycle where each stage produces artifacts that feed the next. Spec to plan to implementation to knowledge capture — with guardrails at every step.

Test-Backed Delivery

Tests are baked into implementation so behavior stays explicit and future agent runs stay on track. TDD-style guidance when it helps — without turning every task into ceremony.

Robot Testing

Encode user journeys as robot-driven widget tests with stable selectors and deterministic test seams. Real user flows, not isolated unit checks.

Flutter Knowledge Base

10 principles. 15+ patterns. Breaking changes docs. Setup recipes. All loaded into every planning and implementation step — not as an afterthought, but as the foundation.

Git Workflow

Conventional commits, automatic PR creation with smart descriptions, and full git worktree support for isolated parallel feature development.

The complete toolkit at a glance

Everything that ships with the Agentic Coding Toolkit

Command What it does
/act:workflow:spec Create detailed specifications with user flows, edge cases, and error scenarios
/act:workflow:refine-spec Adversarially review specs for gaps, wrong assumptions, and codebase misalignment
/act:workflow:plan Turn specs into phased implementation plans with codebase-aware context and TDD guidance
/act:workflow:work Execute plans phase-by-phase with auto commits, testing, and PR creation
/act:workflow:compound Capture reusable insights and decision patterns

Built for Flutter developers who use AI agents

You're past tab-autocomplete. You've used Claude Code or OpenCode to build real features — and you've seen what happens when the AI works without guardrails. Architecture drifts. Tests pass for the wrong reasons. Every session starts from scratch.

You don't want to stop using AI. You want it to work your way — following your patterns, writing proper specs, and producing code you'd actually approve in a PR review.

Whether you're working solo or on a team, ACT gives you that structure.

This isn't for vibe coders. It's for engineers who care about what they ship.

Build With AI Workflows You Can Trust

Built for Flutter developers who want a structured, human-in-the-loop workflow with Claude Code or OpenCode. Early access pricing.

Yearly

Best for AI coding on real projects that require testable and maintainable code.

$39 /year
Buy Now

Includes full toolkit access and updates while your subscription is active:

  • All workflow commands (spec, refine-spec, plan, work, compound)
  • Full git integration including worktree management
  • All Flutter/Dart skills
  • Background research agents
  • Flutter knowledge base (10 principles, 15+ patterns)
  • Weekly updates
Recommended

Lifetime

Best value if you want long-term updates with no renewals. One purchase, lifetime access.

$99 one-time
Buy Now

Includes lifetime access to the full ACT toolkit and all future updates:

  • All workflow commands (spec, refine-spec, plan, work, compound)
  • Full git integration including worktree management
  • All Flutter/Dart skills
  • Background research agents
  • Flutter knowledge base (10 principles, 15+ patterns)
  • Weekly updates
30-day money-back guarantee

I've put a lot of care into making this toolkit genuinely useful and I'm confident you'll find it valuable.

But the only way to know if it's right for you is to try it. Get the toolkit, run the workflow on a real feature, and see the difference for yourself.

If you're not happy with it, for any reason, email contact@codewithandrea.com within 30 days and I'll issue a full refund within 24 hours.

About the Author

Hello, I'm Andrea

Andrea Bizzotto

I'm a Flutter GDE and I've been writing code professionally for over 15 years. With my Flutter tutorials and courses, I've helped thousands of developers become better at building Flutter apps over at Code with Andrea.

The Agentic Coding Toolkit is the system I use daily to build Flutter apps with AI assistance — and now it's available to you.

Frequently asked questions

Do I need Claude Code or OpenCode to use this?
Yes. The toolkit is built for Claude Code CLI and OpenCode CLI. You need at least one of these installed, along with your preferred LLM subscription or API key.
Why OpenCode?
OpenCode is an excellent open-source CLI tool that works with all the major models — not just Claude. ACT is fully compatible with both Claude Code and OpenCode.
What about Codex?
Codex doesn't support custom commands and skills the way Claude Code and OpenCode do. Since OpenCode supports all the OpenAI models, it's the recommended choice if you want to use ACT beyond Claude.
What if I use Cursor or VSCode?
ACT is built for CLI-based AI coding tools (Claude Code and OpenCode), not IDE-integrated AI assistants. It's editor-agnostic — you can use any editor alongside it — but the commands and workflows run in the terminal.
Is this a VS Code extension?
No. It's a CLI-level toolkit that works inside Claude Code or OpenCode. It's editor-agnostic.
Does it work with my existing Flutter project?
Yes. The toolkit installs globally and works with any Flutter or Dart project. It doesn't modify your project structure — it adds development workflows on top.
Can I use ACT for non-Flutter projects?
The workflow commands (spec, plan, work) and git commands work the same way. The agent will skip Flutter-specific guidance, though the Flutter knowledge base is still loaded and will incur some token usage.
What Flutter patterns and packages does it support?
The knowledge base covers 10 principles, 15+ patterns, and breaking changes for Dart and Flutter. Package-specific knowledge is included for Riverpod, Drift, and Sentry, with more packages coming.
How is this different from free Claude Code plugins?
Most open-source toolkits structure workflows but have zero Flutter-specific knowledge. ACT's primary focus is deep Flutter/Dart domain expertise — patterns, principles, breaking changes, and setup recipes built into every step.
How is this different from just writing a good CLAUDE.md?
CLAUDE.md should be absolutely minimal, but often turns into a dumping ground with irrelevant information that makes your agent perform worse for the task at hand. ACT encodes a full workflow — structured spec/plan/work cycles, a knowledge base of patterns and breaking changes, and a compounding system that captures insights across sessions.
What does the workflow actually produce?
I won't promise 100% working software. But if you follow the workflow, you'll get far better results than your average vibe-coded sessions. Each cycle also generates specs in ai_specs/, implementation plans, and reusable knowledge in ai_docs/solutions/. These persist in your project and inform future sessions.
Will this slow down my AI agent?
The structured workflow adds steps, but each step prevents rework that costs more time later. During execution, you can choose lite or full knowledge loading depending on how much guidance you need.
Does it work with team-based workflows?
Yes. The workflow produces artifacts (specs, plans, solutions) that live in your repo. Team members see the same specs, follow the same patterns, and build on the same captured knowledge.
Can I use this for Dart CLI projects too?
Yes. The toolkit includes /dart-create for Dart CLI projects and the knowledge base applies to both Flutter and Dart development.
How do I update it?
Run /act:update from any project. It checks for updates and shows the changelog before applying.
What about code reviews?
ACT is designed so the right requirements are gathered upfront through spec and plan. That makes the resulting code more likely to be correct, so code reviews become lighter — verification rather than discovery.
Is the source code included?
Yes. The toolkit is a set of markdown-based commands, skills, and knowledge files — all fully readable. You're free to build your own skills alongside it. The license does not permit forking or reselling the toolkit itself.

Stop reviewing AI code line by line

Give your AI agent the guardrails, Flutter expertise, and structured workflow it needs to produce code you trust.

Buy Now