flutter-ai-rules: 36 Auto-Loading Flutter and Dart Skills for AI Coding Agents

What flutter-ai-rules provides and who it is for

flutter-ai-rules is a repository of 36 skill files covering Flutter, Dart, and Firebase topics for AI coding agents. Each skill is a folder containing a SKILL.md file. When an AI coding agent encounters a relevant task, it reads the skill's description, decides the task matches, and loads the guidance automatically. The developer does not need to paste anything into a rules file or remember to mention a documentation source.

The README describes this as the key advantage: Firebase Auth guidance appears when the developer touches authentication code, and Riverpod guidance appears when they touch state providers, without any explicit user action per session. The collection covers Flutter and Dart foundations, four state management libraries (BLoC, Riverpod, Provider, and ChangeNotifier), testing (unit, widget, E2E with Patrol, Mockito, and Mocktail), and thirteen Firebase services. It is positioned as a non-opinionated collection: every item in the collection is sourced from official Flutter, Dart, Firebase, and package documentation.

How skill auto-loading works in a supporting agent

A skill is a folder with a SKILL.md file. The file describes what the skill covers and when it is relevant. The agent reads each skill's description when it starts or when it begins a task, and decides which skills apply based on that description and the current task context. The developer's repository can hold skills in an agent-specific directory such as .claude/skills/, .cursor/skills/, .codex/skills/, .agents/skills/, or .windsurf/skills/.

The auto-loading mechanism means skill content is pulled in on demand rather than being permanently loaded in the context window. A project with all 36 skills installed does not load every skill on every message; each one is loaded when the agent determines it is relevant. This is different from adding a single large rules document that is always present in context. The README notes that with a .skills/ folder, a developer can also reference a skill explicitly: for example, by asking the agent to read @.skills/bloc/SKILL.md when reviewing bloc-related code.

Installing via the Skills CLI

The quickest installation path uses the Skills CLI via npx:

npx skills add evanca/flutter-ai-rules

The README states this requires Node 22.20 or newer. On older versions, the CLI exits before doing anything with the message `SyntaxError: ... does not provide an export named 'styleText'`. Every other install route listed in the README works on any Node version.

To review available skills before installing:

npx skills add evanca/flutter-ai-rules --list

To install only a specific skill:

npx skills add evanca/flutter-ai-rules --skill flutter-best-practices

The Skills CLI discovers the skill folders under skills/ in the repository and installs them for each supported agent it detects in the current project.

Installing as a plugin for Claude Code, Codex, and Cursor

The repository also functions as a plugin marketplace for agents that support that model. Plugin manifests live at .claude-plugin/, .codex-plugin/, .cursor-plugin/, .agents/plugins/, and plugin.json, all pointing at the same skills/ directory.

For Claude Code:

/plugin marketplace add evanca/flutter-ai-rules
/plugin install flutter-ai-skills@flutter-ai-rules

Skills then load as flutter-ai-skills:bloc, flutter-ai-skills:riverpod, and so on.

For Codex:

codex plugin marketplace add evanca/flutter-ai-rules
codex plugin install flutter-ai-skills@flutter-ai-rules

For Cursor, the README describes two options: running /add-plugin inside Cursor, or importing the repository as a team marketplace under Dashboard, Settings, Plugins, Import from Repo, pointing to https://github.com/evanca/flutter-ai-rules.

For Antigravity, the install uses git clone followed by the agy plugin command:

git clone --depth 1 https://github.com/evanca/flutter-ai-rules.git
agy plugin install ./flutter-ai-rules

Manual installation and vendoring into a project

For agents that do not support plugins, or for projects that need a vendored copy, the README provides a manual approach. Skills can be copied or symlinked individually from any cloned copy of the repository into the agent's skills directory. To vendor the entire set into a project using a one-line shell command:

git clone --depth 1 https://github.com/evanca/flutter-ai-rules.git temp_repo && mkdir -p .skills && cp -r temp_repo/skills/* .skills && rm -rf temp_repo

This clones the repository with minimal history, copies all skill folders into .skills/, and then removes the temporary clone. The README notes that Windsurf and Devin lack a plugin manifest, but Cascade already scans .agents/skills/ and ~/.agents/skills/, so the manual copy path is the supported install route for those agents.

What the 36 skills cover

The skills are organized into four groups. The Flutter and Dart foundations group has eight skills: flutter-best-practices, effective-dart (for naming and doc comments), dart-3-updates (records, patterns, sealed classes, switch expressions), flutter-app-architecture, architecture-feature-first, flutter-errors (for RenderFlex overflows and unbounded constraints), flutter-use-column-row-first (for Row, Column, Expanded, Flexible layouts), and flutter-pre-caching (for fonts, images, animations, and config).

The state management group covers four libraries: bloc, riverpod, provider, and flutter-change-notifier.

The testing group has five skills: testing (unit, widget, and golden tests), mockito, mocktail, patrol-e2e-testing (for tests involving native permissions or system dialogs), and code-review.

The Firebase group has the most skills, covering thirteen services: authentication, Cloud Firestore, Realtime Database, Cloud Storage, Analytics, Crashlytics, Cloud Messaging, In-App Messaging, Remote Config, App Check, Cloud Functions, firebase-ai (for Gemini text generation via the firebase_ai package), and image generation via Firebase AI.

GitHub Copilot is not supported, and the Node version matters

The README explicitly states that GitHub Copilot does not read SKILL.md files at all. The collection is not usable with Copilot in its skill form. The README points Copilot users to Rules and combined sets, which are described as a legacy approach for agents that consume a flat rules file rather than individual skill files.

The Skills CLI's Node 22.20 requirement is a hard minimum. The README does not suggest a workaround for older Node installations; it notes that all other install routes work without any Node version constraint. Developers who cannot upgrade Node can use the plugin install path, the manual copy command, or the git clone approach shown above.

The repository has no GitHub releases. Updates are tracked in the repository's default branch. Agents installed via the plugin marketplace approach can update the skill set in place, while projects that used the manual copy route require a re-run of the copy command to pick up new skills.

Comparison to maintaining a CLAUDE.md or .cursorrules file manually

The standard approach to giving an AI coding agent Flutter context is to write guidance directly into a CLAUDE.md file, a .cursorrules file, or an equivalent rules document for the agent in use. This is a single file that is always loaded, regardless of what the developer is working on. A Flutter developer who adds a few paragraphs about BLoC conventions, a few about Firebase Auth, and a few about testing will end up with a file that is loaded in full even when the current task has nothing to do with those topics.

flutter-ai-rules separates concerns into 36 individual skills that are loaded on demand. The README frames this as a replacement for pasting guidance into a rules file or remembering to mention documentation sources. The trade-off is that the auto-loading behavior depends on the agent's ability to match a task to a skill description correctly, which the developer cannot directly observe. A large CLAUDE.md is always present and predictable; a skill set is contextually loaded and may miss a relevant skill if the agent's matching logic does not recognize the task.

Editorial conclusion

flutter-ai-rules suits Flutter developers who use an AI coding agent and want context-appropriate guidance delivered automatically rather than manually curated in a rules file. GitHub Copilot is explicitly not supported. The Skills CLI requires Node 22.20 or newer; any other install route has no Node requirement. All 36 skills are derived from official documentation, which means the guidance is verifiable and does not introduce personal conventions that conflict with a team's existing style.

Frequently asked questions

Official sources

1. evanca/flutter-ai-rules on GitHub
2. Issues
3. License: MIT
4. README