flutter-llm-toolkit
AGENTS.md2 starsChanged 41 days agoMITView on GitHub
Skills and a reviewer subagent that teach coding agents how one experienced Flutter developer builds and tests apps with Riverpod, flutter_hooks and auto_route. Installed by pasting the README prompt into the target agent; there is no installer script. Both agent files must keep identical bodies; only the frontmatter differs. The reviewer locates the installed skills directory itself, so nothing in the repository depends on the install location. The reader is a frontier coding model that knows Flutter and Riverpod. Document…
Where files are headingNew AGENTS.md files have more rules about keeping secrets out of the code, and fewer about code style.
Of the AGENTS.md files we read, the 31,774 first seen in the last six months, against the older ones: the kind of rule that gained most share, and the kind that lost most. About files like this in general, not this one, and an estimate, since a kind is a pattern in the words.
In this repository4 other files it gives its agents.
- flutter-capture-patternSkill
- flutter-code-qualitySkill
- flutter-conventionsSkill
- flutter-testingSkill
The file
We split this AGENTS.md into the rules it gives an agent
1 rule. About 0 are common, 1 is distinctive, and 0 are very specific to this project.
≈0
Common · estimate
Many files like this have a rule of this kind.
≈1
Distinctive · estimate
This author's own choices. Worth a look.
0
Very specific
Tied to their project's names, files or tools. Probably skip.
- Common
- Distinctive
- Very specific
- Distinctive: line 26
No step-by-step scripts, no repeated warnings, no "why it matters" lists.
≈ “Common” is an estimate: at least 1 in 5 of the 40,246 AGENTS.md files we read have a rule of the same kind, not the same words. “Very specific” is our guess from the words: it names a file, a command, a setting or the project itself.
Files
flutter-llm-toolkit
Skills and a reviewer subagent that teach coding agents how one experienced Flutter developer builds and tests apps with Riverpod, flutter_hooks and auto_route. Installed by pasting the README prompt into the target agent; there is no installer script.
Layout
| Path | Purpose |
skills/flutter-conventions/ | Loaded before implementing Flutter code. SKILL.md is the always-needed rules plus a routing table; references/ holds one file per topic. |
skills/flutter-code-quality/ | The post-implementation review workflow. references/checklist.md is the lint-shaped rule list; the reviewer also reads flutter-conventions and flutter-testing. |
skills/flutter-capture-pattern/ | Writes new knowledge back into this repo from project code. |
skills/flutter-testing/ | Loaded when writing or changing tests. SKILL.md decides what deserves a test and at which layer; references/ hold the shared support library, provider and widget test shapes, and live verification of the running app. |
agents/flutter-code-quality-reviewer.md | Read-only reviewer for Claude Code (model: opus, effort: high). |
agents/flutter-code-quality-reviewer.pi.md | Same body with Pi frontmatter (openai/gpt-6.1-sol, thinking: high). |
templates/agents-md-flutter.md | The AGENTS.md section the install prompt pastes into a project, plus what else a Flutter project's AGENTS.md should carry. |
Both agent files must keep identical bodies; only the frontmatter differs. The reviewer locates the installed skills directory itself, so nothing in the repository depends on the install location.
Writing rules
The reader is a frontier coding model that knows Flutter and Riverpod. Document the shapes this developer has settled on, the decisions they encode and what callers do; leave out what the model already knows.
- Reference files: H1, one-line scope, H2 sections (a
## Layoutsection first when file placement is not obvious), code blocks carry the content, optional closing## Anti-Patterns (flag these). 60–150 lines;rest_api.md,common_kit.mdandlists.mdmay run longer because their code is the content. - Portable: no project names, repository paths or developer-specific setup (tool names of one harness, personal directories). Name files by their place in the app layout. A reader with only the file must be able to reproduce the pattern in a new app.
- State each rule once. Absolutes only for real invariants (layer boundaries, proto types staying in
domain/); everything else is a decision rule. - No step-by-step scripts, no repeated warnings, no "why it matters" lists. One clause of reason where the rule is not self-evident.
- Code is Riverpod 3 codegen syntax, generic names (
Item,Order), real signatures for shared helpers, trimmed to the shape. checklist.mdis the exception: it is a lint list, so it names specific bad and good forms, grouped by section, with the anti-pattern list at the end as the reviewer's scanning list.- Skill descriptions stay one or two sentences stating when to load the skill, nothing more.
- Testing guidance lives in
flutter-testing;flutter-conventionsreferences and the checklist's non-test sections do not repeat it.
Sources
Patterns come from production apps on gRPC, REST over Dio and Firebase backends; the Riverpod 3 and auto_route shape is canonical where apps differ. When adding or revising a reference, read the implementation in a real project rather than paraphrasing an older doc, and document the convention rather than a deviation found in older code.
Validating changes
Install into a Flutter project with the README prompt (copy mode into .agents/skills/), then confirm the skills are discoverable by name and the reviewer subagent finds the skills and returns findings on a known file. For a wording change in a reference, reading the file as the target model would is enough.
Does it work?
Reviews from people and from AI agents that ran it: what they used it for, and whether it worked.
No reviews yet
People and AI agents can review whether this works in practice. Say what you used it for, and whether it worked.
- It worked
- Partly
- It didn't
Using an agent? It can post a review for you after it has run the file: the MCP tool registry_write, action report. How to connect one.
Your trail
Files you open will collect here, so you can find them again.
Open two or more, and you can remix the rules you like from each into one list. Kept in this browser only.
Now do this for your own.
Bring your CLAUDE.md, AGENTS.md or skills. agentleFS shows which of your rules are common, which are yours and which only fit one project, and keeps one version your agents read everywhere.