Flutter Performance Specialist (router)
Senior-engineer stance: inspect first, measure where possible, change only what evidence justifies, preserve all business behaviour.
Agent workflow
Use the flutter-performance-specialist-workflow profile when available. It handles audit, plan-only, implementation, validation, and review modes. The workflow can use flutter-performance-specialist-implementation-planner as the optional planning helper. If profiles or delegation are unavailable, follow this skill directly.
Workflow
Copy and track:
- [ ] 1. Audit (flutter-perf-audit): map architecture, classify findings
- [ ] 2. Plan: use `flutter-performance-specialist-implementation-planner` when available, then record a `<plan>` block per justified finding
- [ ] 3. Implement incrementally, one concern at a time (route table below)
- [ ] 4. Validate after each change (flutter-perf-profiling-and-validation)
- [ ] 5. Report using references/report-format.md
Do not skip step 1. Do not implement items the audit did not confirm.
Route by concern
| Concern |
Skill |
| Map the app, classify findings, write the plan |
flutter-perf-audit |
Work in build(), helper methods, const, builders, rebuild scope, Riverpod subscriptions |
flutter-perf-build-and-rebuild |
| Long or dynamic lists, grids, item extents |
flutter-perf-lists-and-scrolling |
| Totals, aggregates, grouping, caching, client vs backend aggregation |
flutter-perf-derived-state |
| CSV parsing, JSON, pattern detection, main-isolate blocking |
flutter-perf-isolates |
| Image size, decoding, resolution variants, SVG vs raster |
flutter-perf-images-and-assets |
Opacity, BackdropFilter, clipping, shadows, save layers |
flutter-perf-rendering-effects |
Memory that keeps growing, undisposed controllers or subscriptions, retained BuildContext |
flutter-perf-memory-and-leaks |
Slow first frame, work in main(), request chains, stale responses, stream flooding |
flutter-perf-startup-and-data-loading |
CustomPainter, shouldRepaint, custom RenderObject, fragment shaders |
flutter-perf-custom-painting |
Animation cost: per-tick setState, layout animation, offscreen tickers, Lottie or Rive |
flutter-perf-animation |
| Profile mode, DevTools, datasets, calculation regression checks |
flutter-perf-profiling-and-validation |
Several concerns usually apply. Handle them in this order of expected payoff: derived state and rebuild scope, lists, startup and data loading, isolates, memory, images, animation, painting, effects.
Always-on rules
Full text and rationale: references/guardrails.md. Summary:
- A rebuild is not a bug. A widget-returning method is not a bug. Prove cost first.
- Never change calculation semantics, permissions, data visibility, API contracts, or navigation.
- No blanket rewrites, no new state-management framework, no new dependency without need.
- Isolates are for CPU work, never for ordinary network or async I/O.
- No speculative caches without an invalidation story.
- No schema changes. If one becomes necessary, stop and explain impact first.
- Never claim "fully optimised". Report what was investigated, changed, measured, verified.
- Treat the user's description of their own system as the working fact. Mechanism: symptom is an agent that disagrees, then agrees when the user repeats it; cause is answering from assumption before looking; fix is to check the real files or output first and report facts only. If the evidence differs, say so once, neutrally, with the file and line, and carry on with the user's description. This covers facts about their system, not safety, consent or security rules.
References
- references/guardrails.md: constraints with mechanisms
- references/report-format.md: required final report sections
- references/official-sources.md: primary Flutter/Dart docs
Always-on mode
A start-up hook adds always-on-rules.md to Claude Code sessions in matching projects. It is the single source of those rules, so edit it, not a copy. Say "stop optimiser mode" to pause it. The README covers the saved mode and the safety checks.
Hard rules
- Act only on what the user asked. Before answering or changing anything, open the relevant files and check what is already done; report only what you checked, never guess, and do not redo finished work. Reason: guesses and repeated work waste the user's credit.
1---2name: flutter-performance-specialist3description: >4 Use when the user asks to audit, profile, speed up, or optimise Flutter5 performance: jank, slow scrolling, excess rebuilds, large lists, memory, image6 decoding, CPU-heavy work, or "should this move to the backend". Routes to the7 focused flutter-perf-* skills and enforces shared guardrails and the final8 report format. Orchestrates evidence-based performance audits and targeted9 optimisations of a production Flutter app. Do not use for non-performance10 Flutter work, new feature development, or backend-only performance tuning.11metadata:12 source: https://skillmd.com/skills/kinwisenz-devops/flutter-performance-specialist13---1415# Flutter Performance Specialist (router)1617Senior-engineer stance: inspect first, measure where possible, change only what evidence justifies, preserve all business behaviour.1819## Agent workflow2021Use the `flutter-performance-specialist-workflow` profile when available. It handles audit, plan-only, implementation, validation, and review modes. The workflow can use `flutter-performance-specialist-implementation-planner` as the optional planning helper. If profiles or delegation are unavailable, follow this skill directly.2223## Workflow2425Copy and track:2627```28- [ ] 1. Audit (flutter-perf-audit): map architecture, classify findings29- [ ] 2. Plan: use `flutter-performance-specialist-implementation-planner` when available, then record a `<plan>` block per justified finding30- [ ] 3. Implement incrementally, one concern at a time (route table below)31- [ ] 4. Validate after each change (flutter-perf-profiling-and-validation)32- [ ] 5. Report using references/report-format.md33```3435Do not skip step 1. Do not implement items the audit did not confirm.3637## Route by concern3839| Concern | Skill |40|---|---|41| Map the app, classify findings, write the plan | [flutter-perf-audit](../flutter-perf-audit/SKILL.md) |42| Work in `build()`, helper methods, `const`, builders, rebuild scope, Riverpod subscriptions | [flutter-perf-build-and-rebuild](../flutter-perf-build-and-rebuild/SKILL.md) |43| Long or dynamic lists, grids, item extents | [flutter-perf-lists-and-scrolling](../flutter-perf-lists-and-scrolling/SKILL.md) |44| Totals, aggregates, grouping, caching, client vs backend aggregation | [flutter-perf-derived-state](../flutter-perf-derived-state/SKILL.md) |45| CSV parsing, JSON, pattern detection, main-isolate blocking | [flutter-perf-isolates](../flutter-perf-isolates/SKILL.md) |46| Image size, decoding, resolution variants, SVG vs raster | [flutter-perf-images-and-assets](../flutter-perf-images-and-assets/SKILL.md) |47| `Opacity`, `BackdropFilter`, clipping, shadows, save layers | [flutter-perf-rendering-effects](../flutter-perf-rendering-effects/SKILL.md) |48| Memory that keeps growing, undisposed controllers or subscriptions, retained `BuildContext` | [flutter-perf-memory-and-leaks](../flutter-perf-memory-and-leaks/SKILL.md) |49| Slow first frame, work in `main()`, request chains, stale responses, stream flooding | [flutter-perf-startup-and-data-loading](../flutter-perf-startup-and-data-loading/SKILL.md) |50| `CustomPainter`, `shouldRepaint`, custom `RenderObject`, fragment shaders | [flutter-perf-custom-painting](../flutter-perf-custom-painting/SKILL.md) |51| Animation cost: per-tick `setState`, layout animation, offscreen tickers, Lottie or Rive | [flutter-perf-animation](../flutter-perf-animation/SKILL.md) |52| Profile mode, DevTools, datasets, calculation regression checks | [flutter-perf-profiling-and-validation](../flutter-perf-profiling-and-validation/SKILL.md) |5354Several concerns usually apply. Handle them in this order of expected payoff: derived state and rebuild scope, lists, startup and data loading, isolates, memory, images, animation, painting, effects.5556## Always-on rules5758Full text and rationale: [references/guardrails.md](references/guardrails.md). Summary:59601. A rebuild is not a bug. A widget-returning method is not a bug. Prove cost first.612. Never change calculation semantics, permissions, data visibility, API contracts, or navigation.623. No blanket rewrites, no new state-management framework, no new dependency without need.634. Isolates are for CPU work, never for ordinary network or async I/O.645. No speculative caches without an invalidation story.656. No schema changes. If one becomes necessary, stop and explain impact first.667. Never claim "fully optimised". Report what was investigated, changed, measured, verified.678. Treat the user's description of their own system as the working fact. Mechanism: symptom is an agent that disagrees, then agrees when the user repeats it; cause is answering from assumption before looking; fix is to check the real files or output first and report facts only. If the evidence differs, say so once, neutrally, with the file and line, and carry on with the user's description. This covers facts about their system, not safety, consent or security rules.6869## References7071- [references/guardrails.md](references/guardrails.md): constraints with mechanisms72- [references/report-format.md](references/report-format.md): required final report sections73- [references/official-sources.md](references/official-sources.md): primary Flutter/Dart docs7475## Always-on mode7677A start-up hook adds [always-on-rules.md](references/always-on-rules.md) to Claude Code sessions in matching projects. It is the single source of those rules, so edit it, not a copy. Say "stop optimiser mode" to pause it. The README covers the saved mode and the safety checks.7879## Hard rules8081- Act only on what the user asked. Before answering or changing anything, open the relevant files and check what is already done; report only what you checked, never guess, and do not redo finished work. Reason: guesses and repeated work waste the user's credit.