Use when building cross-platform mobile apps with Flutter, writing Dart code, designing widget trees, choosing state management (Riverpod, Bloc, Provider), implementing platform channels for native features, optimizing Flutter performance (Impeller/Skia, isolates, jank), testing with widget and integration tests, or shipping to the App Store and Google Play. Handles Flutter project architecture, Dart language patterns, Material 3 and Cupertino theming, asynchronous code (Future/Stream/isolates), platform interop (MethodChannel, FFI, pigeon), build and release (Codemagic/Fastlane), and performance profiling. Do NOT use for React Native development (route to react-native-developer), native iOS-only (route to ios-developer), native Android-only (route to android-developer), or Kotlin Multiplatform (route to kotlin-multiplatform).
Installer avec Codex ou Claude Copiez ce prompt, collez-le dans Codex, Claude ou un autre assistant, puis laissez-le vérifier la page du skill et l'installer pour vous.
Une commande directe contourne le prompt de vérification. Examinez la source avant de l'exécuter.
Use when building cross-platform mobile apps with Flutter, writing Dart code, designing widget trees, choosing state management (Riverpod, Bloc, Provider), implementing platform channels for native features, optimizing Flutter performance (Impeller/Skia, isolates, jank), testing with widget and integration tests, or shipping to the App Store and Google Play. Handles Flutter project architecture, Dart language patterns, Material 3 and Cupertino theming, asynchronous code (Future/Stream/isolates), platform interop (MethodChannel, FFI, pigeon), build and release (Codemagic/Fastlane), and performance profiling. Do NOT use for React Native development (route to react-native-developer), native iOS-only (route to ios-developer), native Android-only (route to android-developer), or Kotlin Multiplatform (route to kotlin-multiplatform).
works with Claude Code, Copilot CLI, Cursor, OpenClaw, Gemini CLI
Flutter Developer
Portability target: Spec-level (runs on Claude Code, Copilot CLI, Gemini CLI, Codex, Cursor). No vendor-specific frontmatter fields.
RESEARCH_PREREQUISITE — Execute Before Any Output
This is a HARD GATE. Do not produce ANY output, code, strategy, design, or recommendation without completing this research.
Before you act, you MUST execute every applicable research step. Research-before-acting is the difference between professional work and amateur guessing:
#
Research Step
Why It Matters
Where to Look
RP1
Verify domain currency. Check for breaking changes, deprecations, new standards, or version shifts since the knowledge cutoff.
[STALE_RISK] Flutter ships stable releases ~quarterly and deprecates APIs across minor versions. A widget or API from 12 months ago may be renamed or removed. Outputting stale Flutter code breaks builds.
Audit the system or codebase. Read the actual app: pubspec.yaml, lib/ structure, state-management choice, platform folders (ios/, android/), existing plugins.
[CONTEXT_VIOLATION] Solutions that ignore the app's state-management stack, Dart SDK, and plugin set create technical debt. Every Flutter app has a different architecture.
Cross-reference claims against authoritative sources. Every API, package version, and platform capability claim needs a verifiable source. Mark each: [VERIFIED], [COMPUTED], or [ESTIMATED].
[HALLUCINATION_GUARD] Flutter APIs and package compatibility are the #1 hallucination vector — a wrong package version or removed widget breaks the build.
Official docs, pub.dev, changelogs, migration guides
RP4
Identify known failure modes. List what commonly breaks: platform channel errors, jank from heavy builds, isolate misuse, plugin version conflicts, Impeller rendering regressions. For each: trigger, detection signal, mitigation.
[FAILURE_BLINDNESS] Every Flutter release has known breakages. Output that doesn't address them is dangerously incomplete.
Flutter issue tracker, pub.dev, GitHub releases
RP5
Replace "smooth" with numbers: frame times, startup duration, app size, build time, crash rate.
Quantify impact in concrete units.
[VAGUENESS_PENALTY] "Better performance" is unverifiable. "Cuts jank from 18% to 1% dropped frames and cold start from 2.8s to 1.1s" is verifiable.
Benchmarks, profiler output, release stats
RP6
Map side effects and downstream impacts. What breaks when you upgrade Flutter, add a plugin, change state management, or modify a platform channel?
[CASCADE_BLINDNESS] A Flutter upgrade can break 20+ plugins. A platform channel signature change breaks every caller. Map the blast radius before acting.
[QUALITY_FLOOR] An app that compiles but janks at 30fps on a mid-range device is broken. A release that fails on the oldest supported OS is not released.
Flutter docs, crash reports, quality dashboards
RP8
Declare explicit limitations and edge cases. What does this NOT handle? Which platforms (web, desktop, embedded) are out of scope?
[SCOPE_HONESTY] Naming boundaries prevents misuse. Flutter web/desktop have different constraints than mobile; this skill targets iOS/Android.
This SKILL.md, Flutter platform support docs
If you skip any of these research steps, you are not producing quality output — you are guessing with confidence. Guessing wastes time, breaks builds, and destroys trust. The references, ground rules, and decision trees in this skill exist specifically to prevent guessing. Use them.
Compliance: Research must be executed before any substantial output. For each step, document findings inline in your response using [RESEARCHED] marker: [RESEARCHED: RP1 — Flutter 3.x stable verified against changelog; Dart 3.x current; no breaking changes for the app's widget set.]. Partial research = partial quality. Zero research = zero credibility.
🔄 Iterative Research Loop — Research at EVERY Decision Point, Not Just Entry
The RP1-RP8 cycle above is NOT a one-time gate. It fires continuously at every material decision point throughout the workflow:
Loop
When It Fires
What Re-research Validates
Loop 0: Pre-Action
Before producing ANY output, code, strategy, or recommendation
At every adjustment, phase transition, scale-out, or significant state change
Has the Flutter/Dart version context changed? Are the assumptions still valid?
Loop 2: Pre-Exit
Before closing, handing off, escalating, or declaring completion
Is the deliverable complete by the quality gates defined in RP7? Are all limitations declared (RP8)?
Loop 3: Post-Action
After completion: compare expected vs. actual outcome
What was the frame/startup/size impact? What learnings should feed back into the pattern database?
Integration into Core Workflow:
Every decision point in a skill's Core Workflow must be marked with:
[RESEARCH LOOP: Re-execute RP1-RP8 before proceeding to next phase]
This ensures the agent pauses to re-verify ALL research dimensions before making the next decision. A skill that only researches at entry and then operates on auto-pilot is a skill that makes decisions on stale context.
Markers for output: At each loop, the agent outputs: [RESEARCHED: Loop N — RP1-RP8 re-verified. Key delta from previous loop: ...]
Why this matters: A decision made in Loop 0 may be catastrophically wrong by Loop 2 because the context changed. Flutter versions bump. Dart SDKs advance. Platform rules shift. The research loop catches context drift before it becomes output error.
Compliance: Research must be executed before any substantial output AND re-executed at every decision point. For each research loop, document findings inline. Partial research = partial quality. Zero research = zero credibility. Stale research = dangerous confidence.
Anti-Hallucination
Rationalization
Reality
"I remember the Flutter API — it hasn't changed."
Flutter deprecates and renames widgets across stable releases. An API from 12 months ago may be removed. Verify against the installed Flutter/Dart version's docs.
"That package works, I've used it before."
Package compatibility is version-pinned to Dart SDK and Flutter version. A package that worked on an older Flutter may not compile or behave the same. Check pub.dev compatibility and flutter pub outdated.
"Jank is a hardware problem."
Jank is usually an app problem: heavy builds in build(), blocking the UI isolate, oversized images, or missing const constructors. Profile with DevTools before blaming devices.
"Platform channels are simple — just pass the data."
MethodChannels are async and can throw; data crossing the boundary must be JSON-serializable; native code must run on the right thread. Errors here crash or hang silently.
Admit uncertainty — If you don't know the exact API for the installed Flutter/Dart version, say so and check the docs or installed SDK. Never fabricate.
Flag your knowledge cutoff — Flutter moves fast; state what version you verified against and when.
Never guess security — Never hand-roll crypto in Dart, never disable code signing, never bypass store policies. Default to the safer interpretation.
[VERIFIED] — Every API, package version, and platform claim must be traceable to a reference in references/ or the installed SDK. Tag unverifiable claims with [UNVERIFIED].
Ground Rules — Read Before Anything Else (QUICK)
#
Rule
Mechanical Trigger
Violation Response
R1
Anchor to the installed Flutter/Dart versions and pubspec first. Read pubspec.yaml, the lib/ structure, and state-management choice before proposing any code — and run the shared freshness check to confirm the installed versions are current.
Any Flutter code proposal without checking pubspec.yaml and the app's architecture first
Stop. Read pubspec.yaml + lib/ structure; run bash scripts/lib/library-version-check.sh . --strict; anchor all APIs to the detected versions
R2
Never upgrade Flutter or Dart SDK without a migration plan. Version bumps break plugins and deprecate APIs; plan the upgrade and test matrix first.
A proposed flutter upgrade or major package bump with no migration/rollback plan
Require a migration plan: affected plugins, breaking changes, test matrix, rollback
R3
Keep heavy work off the UI isolate. Builds stay cheap, IO/CPU work goes to isolates, platform work stays in platform channels.
Blocking IO or heavy computation inside build() or the UI isolate
Move to an isolate (compute/Isolate.run) or async platform work; profile to confirm no jank
R4
Use the app's state-management stack — don't introduce a second one. Riverpod, Bloc, or Provider; pick per project and stay consistent.
New state code using a different paradigm than the app's established stack
Reconcile with the app's chosen stack; document the boundary
R5
Platform channels are contracts — version them and handle errors. Every MethodChannel call has a typed interface, error handling, and a documented native counterpart.
A platform channel call with no error handling or no typed interface
Hand off missing skills, don't improvise them. If a downstream task needs a skill not in this library, create it via the Core Workflow Phase 6 protocol before routing.
Handoff target has no name: match in skills/
Trigger autonomous skill-creation-on-handoff, then route with a symmetric chain
The Expert's Mindset (QUICK)
World-class Flutter engineers think in widgets as declarative state, not screens. Every frame is a reconciliation: the framework diffs the widget tree against the previous one and repaints only what changed. The expert writes widgets that are cheap to build (const constructors, minimal rebuilds), composes small focused widgets instead of one giant screen, and understands exactly when setState, providers, and rebuilds happen. They can look at a jank profile and name the offending widget.
They treat state management as an architecture decision, not a library preference. The stack (Riverpod/Bloc/Provider) defines where state lives, who can mutate it, and how rebuilds propagate. The expert chooses by team and app scale and then applies it consistently — mixing paradigms is how rebuild bugs and "it works but I don't know why" states appear.
They know the platform boundary is real. Flutter draws its own pixels, but plugins and platform channels are the contract with iOS/Android native. The expert treats every channel as a versioned, typed API with error handling, and knows that data crossing the boundary must be serializable and thread-correct.
What Flutter Masters Know (STANDARD)
The build method is sacred — keep it pure and cheap; heavy work in build() is the #1 jank source. Use const widgets, RepaintBoundary, and narrow rebuild scopes.
Isolates are the async escape hatch — Isolate.run/compute for CPU work; never block the UI isolate. Each isolate has its own memory; passing large data copies it.
Impeller changed rendering — the Impeller renderer (default on modern Flutter) fixes Skia-era jank but has its own edge cases (some shader/effects behave differently). Verify rendering with the Performance overlay.
Platform channels fail loudly and asynchronously — every call needs error handling; the platform side must post to the main thread. Pigeon generates the typed glue and removes a whole class of typos.
When to Break Your Own Rules (DEEP)
Break R3 (keep work off UI isolate) for tiny computations — a single small calculation in a build is fine; the rule protects frames, not every operation.
Break R4 (one state stack) at module boundaries — a legacy module may keep Provider while new modules use Riverpod; isolate the boundary explicitly rather than force-migrating all at once.
Break R6 (test matrix) for a critical hotfix — a one-line crash fix may ship with a reduced matrix if the full suite is hours away — but only with a documented follow-up test and a rollback plan.
Never break R1 (anchor to versions). There is no scenario where writing Flutter code without checking pubspec.yaml and the installed SDK is correct.
Operating at Different Levels (STANDARD)
L1: Apprentice
Scope: Building screens with widgets, basic state with setState/Provider, styling with ThemeData, navigation with Navigator.
Autonomy: Can implement UI against an existing architecture; cannot choose state management or add plugins.
Impact: Delivers feature UI; learns the rebuild model.
Craft: Correct widgets, const constructors, no blocking work in build().
L2: Practitioner
Scope: State management (Riverpod/Bloc), forms and validation, lists with ListView.builder, async with Future/Stream, unit + widget tests, package selection.
Autonomy: Owns features end-to-end within the established stack; can add pure-Dart packages.
Impact: Ships complete features with tests; maintains smooth scrolling.
Autonomy: Sets org-wide Flutter standards; arbitrates state-management and plugin decisions; owns the build system.
Impact: Multi-app consistency; build times cut by 50%+; org-wide crash-free rate > 99%.
Craft: Builds abstractions that survive Flutter releases; quantifies engineering-hour savings.
L5: Transformative
Scope: Redefines how the org ships mobile with Flutter — reusable product surfaces, design-system-as-code, plugin strategy that makes native interop boring.
Autonomy: Influences product and platform roadmaps; sets the mobile north star.
Impact: Mobile velocity competitors can't match; marginal cost of a new app near zero.
Craft: Builds self-measuring release pipelines; teaches the discipline org-wide.
When to Use (QUICK)
Use this skill when:
Building a cross-platform mobile app with Flutter — You need project setup, widget architecture, state management, and a release path. Architecture-first avoids rework.
Writing Dart for async, isolates, or platform interop — Futures, streams, Isolate.run, MethodChannel/FFI/pigeon. This skill covers the correct patterns.
Fixing jank or slow startup — Profiler-driven workflow (Decision Tree 3) finds the heavy build, blocking isolate, or rendering issue.
Choosing or migrating state management — Riverpod vs Bloc vs Provider with rebuild-scope discipline.
Adding native functionality — Platform channels for push, biometrics, camera, or custom SDKs with typed contracts.
Setting up Flutter CI/CD and store release — Codemagic/Fastlane, signing, and both-store submission.
Auditing an existing Flutter app — Upgrades, package compatibility, app size, and crash triage.
File/dependency detection:pubspec.yaml with flutter/dart deps, lib/main.dart, analysis_options.yaml, ios/Podfile, android/app/build.gradle, .dart_tool/package_config.json → auto-activate this skill.
When NOT to Use (QUICK)
Do NOT use this skill when:
The app is React Native or Expo — JS/TS, RN components, native modules → route to react-native-developer.
The app is native iOS-only — Swift/SwiftUI, UIKit → route to ios-developer.
The app is native Android-only — Kotlin, Jetpack Compose → route to android-developer.
The app is Kotlin Multiplatform — shared Kotlin across platforms → route to kotlin-multiplatform.
The problem is mobile architecture patterns — MVVM vs MVI at the decision level → route to mobile-architecture-patterns (or mobile-developer first).
If your task involves the Flutter/Dart stack specifically — widgets, state, isolates, channels, performance, store shipping — this is the right skill. If it involves another mobile stack, hand off.
Route the Request (QUICK)
Condition
Action
File/dependency detected: pubspec.yaml with flutter/dart deps
Which Flutter and Dart versions are installed? (I must anchor to them.)
Which state-management stack is the app using?
Is the goal a new feature, a performance fix, a release, or a native integration?
Do you target iOS, Android, or both?
Anti-Rationalization (QUICK)
AR-01 No Unanchored Code: You CANNOT write Flutter/Dart code without checking pubspec.yaml and the installed SDK first. "I know the API" is how removed widgets land in a working app. Anchor to versions — R1 is non-negotiable.
AR-02 No Solo Upgrades: You CANNOT bump Flutter or Dart SDK without a migration plan and rollback path. "It'll probably be fine" is how one plugin breaks the build for a whole team.
AR-03 No Heavy Builds: You CANNOT put blocking work or expensive computation in build(). "It's only one screen" is how jank appears on mid-range devices. Isolates and async are the escape hatches.
AR-04 No Uncontractual Channels: You CANNOT add a MethodChannel call without a typed interface and error handling. "It's just one call" is how silent hangs and crashes appear at the platform boundary.
AR-05 No Performance Claims Without Profiler: You CANNOT claim a performance improvement without before/after DevTools profiler evidence. "It feels smoother" is not a metric.
AR-06 No Handoff Without a Missing-Skill Check: You CANNOT route a handoff to a role whose skill does not exist in this library. If the target skill is missing, run Phase 6 (create it autonomously) before handing off.
Do: Read pubspec.yaml, lib/ structure, analysis_options.yaml, and the ios//android/ folders. Run flutter --version and flutter pub outdated to surface drift, then run the shared freshness check: bash scripts/lib/library-version-check.sh . --strict (per scripts/references/library-freshness-policy.md).
Verify: You can state: Flutter and Dart versions, the state-management stack, the plugin list, and the platform-channel inventory.
Output: An anchored version matrix with [VERIFIED] tags for each pinned version.
[RESEARCH LOOP: Re-execute RP1-RP8 before proceeding — versions verified, no drift]
Phase 2: Decide — Architecture and State Management (~20 min)
Do: Choose the state-management stack per references/state-management.md (Riverpod for most apps, Bloc for large teams, Provider for small apps). Define folder structure and data-flow boundaries.
Verify: The choice maps to a concrete team/app scale; rebuild scopes are understood; no feature mixes two paradigms without a documented boundary.
Output: A documented architecture decision + filled references/state-management.md worksheet.
Phase 3: Build — Widgets, Data, Async (~2-4 hrs)
Do: Implement features with cheap widgets (const, small focused widgets), typed models, and the data layer (Repository pattern, references/data-layer.md). Handle async with Future/Stream per references/async-dart.md.
Verify: Both platforms build; flutter analyze clean; unit + widget tests pass; no blocking work in build().
Output: Working feature on both platforms with passing tests.
[RESEARCH LOOP: Re-execute RP1-RP8 before proceeding — both platforms build, analyze clean]
Do: For native needs, use pigeon to generate the typed channel interface (references/platform-channels.md), implement the native side (Swift/Kotlin), and wire error handling. Push via firebase_messaging/FCM/APNs; biometrics via local_auth.
Verify: The channel is typed, error-handled, thread-correct, and tested on both platforms.
Output: Native capability implemented with channel contract + tests.
Validate. Run lint-template.py, lint-yaml.py, lint-markdown.py, and bash scripts/validate-skills.sh.
All gates pass; skill registers in the router
7
Hand off. Invoke the new skill's workflow for the original task, and record the creation in the State Log.
Downstream task completes using the created skill; State Log entry documents the gap + creation
Creation boundary: Only create a skill when (a) the task genuinely recurs or is consequential, (b) no existing skill covers it, and (c) you can fill it to the 10/10 bar. For one-off, low-stakes gaps, record the gap in the State Log and route to the nearest existing skill instead — creating a half-quality skill is worse than routing.
Handoff: Deliver the completed app architecture (or the new skill) to the consuming skill via cross-agent-skills-packaging conventions, and confirm the downstream skill's consumes_from includes this skill so the graph stays symmetric.
Best Practices (STANDARD)
Anchor every decision to the installed Flutter/Dart versions and pubspec. State the versions [VERIFIED] and run flutter pub outdated before proposing changes. Version drift is the #1 "works on my machine" failure.
Keep build() pure and cheap. const constructors, small focused widgets, no IO or heavy computation in build. Use RepaintBoundary to isolate repaints. Jank is a build problem first.
Choose one state-management stack and use it consistently. Riverpod for most apps, Bloc for large teams, Provider for small apps. Mixing paradigms is how rebuild bugs and unpredictable state appear.
Move CPU work to isolates.Isolate.run/compute for heavy computation and parsing; never block the UI isolate. Remember isolates copy data — keep the payload small.
Version platform-channel contracts with pigeon. Generated typed interfaces eliminate typo-class bugs and force error handling. Every channel call has a defined native counterpart and thread behavior.
Use lazy lists (ListView.builder, SliverList) and lazy images. Never build a 1,000-item list eagerly. Lazy building + cacheExtent control keeps scrolling at 60fps.
Profile with DevTools before claiming performance. Performance overlay for frames, Memory/CPU profilers for leaks and isolate load. Claims without numbers are guesses.
Run flutter analyze with strict lints as a gate. The Flutter lints and flutter_lints package catch whole classes of bugs before tests do. Zero-analyze-warning is a release requirement.
Test the real matrix before release. Widget + integration tests on both platforms and the oldest supported OS; a release that only passed on one platform is not tested.
Configure release builds properly. AOT compilation, tree-shake icons, obfuscation for release, --split-debug-info, and sized assets. Size and startup are engineered, not hoped for.
Decision Trees (STANDARD)
Decision Tree 1: State Management Choice
How large is the team and app?
├─ Small app, 1-3 devs → Provider or Riverpod
│ └─ Complexity growing? → Riverpod (automatic rebuild scoping)
├─ Large app, multiple teams → Bloc
│ └─ Need strict event/state discipline + testability → Bloc
└─ Unsure → Start with Riverpod; it scales down and up
└─ Legacy Provider/Bloc app? → Stay on the existing stack; migrate at module boundaries
Decision Tree 2: Platform Channel or Plugin?
Does an existing Flutter plugin cover the need?
├─ YES → Use the plugin (prefer battle-tested, well-maintained)
│ └─ Needs custom config → Use the plugin's platform setup docs
└─ NO → Write a platform channel
├─ Is the data simple (JSON-serializable)?
│ ├─ YES → Pigeon-generated channel + error handling
│ └─ NO → Design the serialization contract first; pass IDs, not objects
└─ Need background/long-running native work? → Native service + channel events
Decision Tree 3: Performance Regression
What is the symptom?
├─ Jank / dropped frames → Profile with Performance overlay
│ ├─ Heavy build() → Split widgets, add const, use RepaintBoundary
│ ├─ Blocking work in UI isolate → Move to Isolate.run
│ └─ Large lists → ListView.builder / SliverList + lazy images
├─ Slow startup → Profile startup
│ ├─ Large bundle → Tree-shake icons, AOT, remove unused plugins
│ └─ Sync plugin init → Defer non-critical initialization
├─ Large app size → Audit plugins + assets
│ ├─ Unused plugins → Remove
│ └─ Heavy assets → Compress; serve from CDN where possible
└─ Memory growth → Leak hunt
├─ Timers/streams not disposed → Dispose controllers/subscriptions
└─ Images/lists retained → Verify with Memory profiler
Decision Tree 4: Release to Store
Is this a hotfix or a planned release?
├─ Hotfix → Integration-test the fix, ship to both stores, monitor crash-free rate
├─ Planned release → Run the full matrix:
│ ├─ flutter analyze clean? → NO: fix warnings first
│ ├─ Widget + integration tests green on both platforms? → NO: fix and re-run
│ ├─ Performance budgets met? → NO: profile and fix
│ └─ Signing + store metadata ready? → Ship via Codemagic/Fastlane
└─ First release → Complete store setup: icons, privacy policy, test accounts, content rating
Error Recovery (QUICK)
Symptom
First Action
If That Fails
Last Resort
Build fails after a pub add
flutter pub outdated + check package compatibility with the installed SDK
Remove the package; pin to a compatible version from pub.dev
Revert pubspec change; re-run flutter pub get from the lockfile
Jank appears after a change
Profile with the Performance overlay; find the heavy build or blocking call
Split the widget / move work to an isolate
Revert the change; profile again; isolate the offender
Platform channel hangs or throws
Add error handling; check the native side runs on the main thread
Use pigeon to regenerate the typed interface
Fall back to the plugin; document the channel contract
App crashes on launch on one platform
Check the crash log for a missing plugin or misconfigured channel
Rebuild with flutter clean; verify plugin registration
Choosing the stack and architecture before Flutter-specific work
ios-developer
Swift interop, App Store process, HIG
Platform channels, push (APNs), and iOS release specifics
android-developer
Kotlin interop, Play Console, Material 3
Platform channels, FCM, and Android release specifics
Downstream (What This Skill Produces)
Downstream Skill
Deliverable
What They'll Do With It
mobile-developer
Working Flutter app + architecture decisions
Integrate with the broader mobile strategy (offline-first, performance, security)
qa-engineer
Test-ready app + integration test suite
Run the mobile test matrix against the release build
devops-engineer
Codemagic/Fastlane pipeline
Operate CI/CD and store release automation
Skill Creation on Handoff (autonomous):
Situation
Trigger
Action
Downstream task needs a skill that does not exist in the library
No name: match + no >80% description-similar neighbor in skills/
Run Core Workflow Phase 6 — scaffold, fill 22 sections, validate, wire symmetric chain, then hand off
A generated skill must be packaged for cross-agent reuse
Skill must run on Claude Code, Copilot, Gemini CLI, Cursor
Route to cross-agent-skills-packaging for portability testing + packaging
Complex multi-step handoff between agent roles
Handoff involves state, unresolved questions, or 3+ skills
Route to agent-handoff-protocol for the structured handoff ledger
A skill must be created or recreated from scratch at 10/10 quality
"create/regenerate skill for X" request
Route to dynamic-skill-creator (full discovery + generation protocol)
Proactive Triggers (STANDARD)
Flutter or Dart SDK upgrade available → Surface the migration plan and plugin-compat matrix before anyone upgrades ad hoc. 🔴
A plugin is unversioned or pinned to a git branch → Flag as a reproducibility and compat risk before it breaks a build. 🟡
Jank or dropped frames trending up → Alert the team before a release ships with the regression. 🔴
A new MethodChannel with no error handling → Intervene: require a typed contract (pigeon) and error paths. 🔴
flutter analyze warnings in the release path → Block: zero-analyze-warning is a release requirement. 🟠
Integration tests skipped for a release → Block: a release without both-platform integration tests is not tested. 🔴
A package added without checking Dart SDK compatibility → Flag flutter pub outdated mismatch before it breaks the build. 🟡
State Log (QUICK)
Turn
Action
Decision
Risk Accepted
Mitigation
1
Anchored versions
Adopted Flutter 3.x / Dart 3.x matrix
Plugin compat drift
flutter pub outdated + lockfile pinning
2
State stack
Chose Riverpod for the app
Learning curve for new devs
Architecture doc + review checklist
3
Channel contract
Pigeon for the biometric channel
Native side changes
Typed interface + error tests
4
Release plan
Codemagic + staged store rollout
Store review latency
TestFlight/internal testing tracks
5
Skill gap detected
Created <new-skill> via Phase 6
New skill is v1.0
Full validation + symmetric chain wiring
N
...
...
...
...
Anti-Drift Check: Before each response, verify:
Did the last action match the plan?
Are we still within the version matrix and state-stack decisions?
Has any new information (Flutter release, plugin breakage, crash data) invalidated prior decisions?
What Good Looks Like (QUICK)
A Flutter deliverable reads like an engineering handoff, not a code dump. It opens with the anchored version matrix [VERIFIED] — Flutter, Dart, state stack, plugin list — then shows the architecture decision with rebuild-scope rationale. Performance work comes with DevTools before/after evidence (jank 18% → 1% dropped frames, cold start 2.8s → 1.1s, size 74MB → 42MB). Platform channels have pigeon-generated typed contracts with error paths; tests run on the real matrix. The deliverable ends with measured risks and what needs a human decision.
Signs of Excellence:
Versions anchored and verifiable; flutter analyze clean.
Performance claims backed by DevTools profiler numbers.
Channel contracts typed and error-handled.
Both-platform integration tests pass before any release.
Signs of Dysfunction:
Code written against a guessed Flutter version.
"It's smoother now" with no measurement.
A MethodChannel with no error handling.
Tests that only ran on one platform.
Deliberate Practice (STANDARD)
graph TD
A[Anchor versions] --> B[Pick one surface]
B --> C[Build cheap widgets / offload work]
C --> D[Profile before/after]
D --> E[Validate on both platforms]
E --> F[Review against release gates]
F --> G[Feed learning into version matrix]
G --> A
Level
Routine
Time
Success Metric
Novice
Build a two-screen Flutter app with navigation + Riverpod; run widget tests
4 hours
App builds on both platforms; tests pass; versions documented
Intermediate
Fix a jank regression with DevTools evidence; add const/RepaintBoundary
4 hours
Jank < 2% dropped frames with before/after profiler output
Advanced
Add a pigeon platform channel with error handling + integration test
6 hours
Channel contract typed; tests green on both platforms
Expert
Plan and execute a state-management or performance overhaul with org-wide budgets
1-2 days
Analyze clean; jank < 1%; release gates enforced in CI
Anti-Patterns (STANDARD)
❌ Anti-Pattern
✅ Do This Instead
❌ Heavy build() trees — a 200-line build method with eager lists and no const, janking on mid-range devices.
✅ Cheap widgets — small const widgets, ListView.builder, RepaintBoundary; profile to confirm.
❌ Blocking the UI isolate — parsing a large JSON or doing CPU work in build() or a synchronous handler.
✅ Isolate offload — Isolate.run/compute for heavy work; keep the payload small.
❌ Mixing state paradigms — setState in one screen, Bloc in another, Provider in a third.
✅ One stack per app — choose Riverpod/Bloc/Provider and apply consistently; migrate at module boundaries.
❌ Untyped MethodChannels — hand-rolled channel calls with stringly-typed method names and no error handling.
A catalog app's product screen rebuilds a 400-widget tree on every scroll because the list is eager and the widgets aren't const. Mid-range Android scrolls at 28fps; the app's rating drops to 3.9 and support tickets about "laggy app" rise. At 150K MAU, a 0.3-star rating drop and churn of ~5% on the catalog screen is worth roughly $30,000/year in retention. Fix: ListView.builder + const widgets + RepaintBoundary; scroll test as a release gate.
2. Untyped Platform Channel Crash ($9K/incident)
A biometric check via a hand-rolled MethodChannel has a channel-name typo on the Android side. It works on iOS, passes review, then crashes 20% of Android users at login with "MissingPluginException." Emergency hotfix + support surge + rating drop: $9,000/incident. Fix: pigeon-generated contracts; verify channel names on both platforms; integration-test the real build.
3. Solo Flutter Upgrade Breakage ($15K/build outage)
A flutter upgrade without checking plugin compatibility breaks the iOS build for the whole team for 2 days — CI red, release delayed. 5 engineers × 2 days × $150/hr: $12,000 in lost velocity, ~$15,000 with recovery overhead. Fix: R2 — upgrades are planned migrations with plugin-compat matrices, never solo bumps.
4. Debug Build Shipped as Release ($7K/incident)
A release pipeline that forgot --release ships a debug build: 90MB app, 5s cold start, huge battery drain. Store reviewers reject; the fix takes a week of rework and re-submission. $7,000/incident in engineering and review time. Fix: release config is a checklist item (CR11) — AOT, tree-shake, size report, verified artifact.
5. Blocking Parse on the UI Isolate ($20K/year)
A chat app parses a 4MB message history JSON on the UI isolate at startup; the app freezes for 2-3s on slow devices. Users interpret it as a crash; ~4% churn delta on first launch. At 80K MAU, $20,000/year in retention. Fix: Isolate.run for parsing; show a skeleton while loading; profile startup.
Production Checklist (STANDARD)
CR1: Version matrix anchored and verified — Verification: flutter --version + flutter pub outdated documented [VERIFIED]; pubspec locked
CR1b: Library freshness verified — Verification: bash scripts/lib/library-version-check.sh . --strict reports FRESH (or every outdated group has a documented, time-boxed exception per scripts/references/library-freshness-policy.md)
CR2: State-management stack chosen and consistent — Verification: one paradigm per app; boundaries documented; no mixed paradigms without a documented boundary
CR3: flutter analyze clean with strict lints — Verification: zero analyzer warnings in CI; flutter_lints enabled
CR4: Both platforms build — Verification: flutter build ios and flutter build apk --release succeed on a clean checkout
CR5: Unit + widget tests pass — Verification: flutter test green; no skipped tests in the release path
CR6: Integration tests pass on both platforms — Verification: integration_test suite green on iOS and Android
CR7: Performance budgets met with profiler evidence — Verification: jank < 1% dropped frames; cold start within budget; memory stable
CR8: Platform channels typed and error-handled — Verification: pigeon contracts (or equivalent typed interface); error paths tested; native side thread-correct
CR10: App size within store limits — Verification: APK/IPA size within budget and store limits; assets compressed
CR11: Signing and store credentials managed — Verification: Codemagic/Fastlane signing configured; secrets in the credential store, not the repo
CR12: Accessibility pass — Verification: Flutter semantics on interactive elements; screen-reader labels on key screens
CR13: Crash-free sessions ≥ 99% — Verification: crash rate monitored per version; regressions gated before rollout completes
CR14: Handoff skill gaps resolved — Verification: any required downstream skill missing from skills/ was created via Phase 6 or the gap is recorded in the State Log
Gotchas (QUICK)
Gotcha
Cost
Fix
Channel-name typo → "MissingPluginException" at runtime
$5K-$15K/incident in hotfix + support
Pigeon contracts; verify names on both platforms; integration-test the real build
Heavy build() jank on mid-range devices
$15K-$40K/year in retention churn
Cheap widgets, const, lazy lists, RepaintBoundary; scroll test as a release gate
Solo flutter upgrade breaks plugins
$10K-$20K per build outage
Planned upgrades with a plugin-compat matrix; never solo bumps
Debug build shipped as release
$5K-$10K per rejection
Release config as a checklist gate (AOT, tree-shake, size report)
Blocking parse on the UI isolate
$10K-$30K/year in churn
Isolate.run; skeleton loading; startup profiling
Verification (STANDARD)
#
Complete when...
Verify
☐
Complete when the version matrix is anchored: Flutter and Dart versions, state stack, and plugin list stated [VERIFIED] from pubspec/SDK
Complete when library freshness is verified: bash scripts/lib/library-version-check.sh . --strict reports FRESH, or every outdated dependency group carries a documented exception
Verify the checker output; exceptions have an expiry/review date in the State Log
☐
Complete when the state-management stack is chosen and consistent: one paradigm per app with documented boundaries
Verify the architecture doc; no mixed paradigms without a documented boundary
☐
Complete when both platforms build on a clean checkout: flutter build ios and flutter build apk --release succeed
Verify clean builds; no stale caches or local-only config
☐
Complete when tests pass on the matrix: flutter test green, integration tests green on both platforms
Verify CI runs the matrix; skipped tests count as failures
☐
Complete when performance budgets are met with evidence: jank < 1% dropped frames, cold start within budget
Verify DevTools profiler output committed; budgets enforced in CI
☐
Complete when platform channels are typed and error-handled: pigeon contracts with error paths
Verify channel code review; error-path tests pass on both platforms
☐
Complete when handoff skill gaps are resolved: any downstream task requiring a missing skill was created via Phase 6 or logged
Verify python3 scripts/validate_chains.py reports 0 asymmetries for created skills; State Log has the gap entry
Verification Guardrails (STANDARD)
Pre-Generation
Confirm the installed Flutter/Dart versions and state stack are captured — never write unanchored code
Confirm plugins are versioned and compatible (flutter pub outdated)
Confirm the release path (build config, signing) is decided before any change that affects it
Post-Generation
Re-run both-platform builds and the test matrix; confirm no new failures
