Tagline
A repo-based AI development office for building Flutter apps with specialized agents, durable memory, and release discipline.
Loading portfolio
Open-source AI dev officeFlutter + agentic development systemPersonal open-source project2026
AI Dev Team Flutter: An AI-Operated Flutter Studio
A public MIT-licensed Flutter workspace that treats AI agents like a product team, using role contracts, branch ownership, repo-backed memory, and FVM quality gates to move ideas toward production main.
システム設計図
Description
The Problem Domain
AI coding sessions often become one overloaded conversation: product intent, UI decisions, implementation details, tests, review, and release risk all compete for the same context. The result can be impressive demos, but weak continuity, vague ownership, and unclear proof that the work is production-ready.
Technical Implementations
Office Assistant: Designed a natural-language front door that can route unstructured requests into product, design, architecture, Flutter implementation, QA, code review, and release roles.
Role Contracts: Defined each agent with a mission, branch, file ownership, handoff path, and boundaries so parallel work can stay scoped instead of colliding across the repo.
Repo Memory: Stored decisions in Markdown docs, feature folders, handoffs, outboxes, commits, and a lightweight status index so future sessions can resume from evidence instead of hidden chat history.
Flutter Quality Gates: Standardized the workflow around FVM, formatting, Flutter analysis, tests, web builds, and release handoffs before work is treated as production-ready.
What It Solves
The project gives AI-assisted development a repeatable operating model: split complex work by role, preserve context in the repo, verify Flutter behavior through quality gates, and make progress status readable without scanning every generated file or replaying old chats.
Project story
AI workflow architecture
Turning AI coding from one chat into an operating system
I built AI Dev Team Flutter because one powerful assistant is useful, but large app work needs memory, ownership, review, and release discipline. The project asks a practical question: what if the repo itself became the office where AI agents work?
Why I built it
The pain was not generating code. The pain was continuity: sessions forget why decisions were made, status checks become expensive, and one context window tries to be product manager, designer, engineer, QA, and release lead at the same time.
How I designed it
I split work into role contracts and made the repository the source of truth. Branch names, file ownership, feature docs, handoffs, outboxes, and quality gates all exist so agent output can be inspected and resumed by a later session.
What it proved
The office is already tied to actual Flutter product work: the Minimal Timer app shipped to main, newer timer features are tracked in the status index, and the docs distinguish shipped work from items that still need verification.
Open-source Flutter AI-office with role contracts, repo memory, FVM gates, and shipped app work.
主なポイント
- Built a public Flutter AI-office architecture around role-specific agents, branch ownership, handoffs, and release gates.
- Kept the user-facing entrypoint natural: plain tasks activate the Office Assistant instead of requiring manual workflow commands.
- Documented a branch model with production `main`, reusable `org/main`, integration branches, and scoped role branches.
- Pinned the Flutter workflow to FVM and recorded standard gates for format, analyze, tests, and web builds.
- Added a branch-aware status protocol through `docs/features/status-index.md` so progress checks stay lightweight and read-only.
- Shipped the Minimal Timer app to `main`, with later timer features tracked through the same feature-status system.
- Made the workflow portable across tools through native sub-agents where available and packet-based fallback prompts elsewhere.
担当範囲
- Designed the overall AI-office operating model and role taxonomy.
- Wrote the reusable documentation system for feature planning, async handoffs, status checks, and release decisions.
- Set up Flutter-specific tooling expectations with FVM, Dart/Flutter gates, MCP configuration, and official skill usage.
- Used the system to coordinate real Flutter app work instead of leaving it as a prompt-only concept.
関連プロジェクト
2018年-現在 / 公開済みプロダクト
Alt Games Portal
プロンプトや画像からゲームを生成できるエージェント型AIゲームジェネレーターを備えた公開済みFlutterゲームポータル。元はネイティブAndroidアプリWeb Games Portalとして始まりました。
2026 / 最近のGitHubプロジェクト
Brobot Companion Lab
ローカルOllamaモデルから返る構造化JSONアクションで動く、ElectronとThree.jsのコンパニオンロボットプロトタイプ。
2021 / プロチームプロジェクト
Sweed Food Delivery App
おすすめ機能、決済、配達場所対応を備え、地域の食品を注文できるReact Nativeフードデリバリーアプリ。