name	yourssh-refactor-sweep
description	Use when the user wants to review and refactor/optimize functions across the yourssh codebase — e.g. "review và refactor", "tối ưu code", "clean up duplication", "find dead code", or a broad "optimize the whole project" request scoped to app/lib or packages/.

YourSSH Refactor Sweep

Review the Flutter codebase for optimization opportunities and apply the safe, behavior-preserving ones. Optimize for high-confidence wins; defer risky changes for explicit sign-off.

When to use

User asks to "review và refactor", "tối ưu", "clean up", "find dead code" over an area or the whole project.
After a feature lands and the code needs a tidy pass.

Do not use when: the user wants a single targeted bug fix (use superpowers:systematic-debugging) or a behavioral change/new feature (use superpowers:brainstorming).

Four categories to scan

Performance — excess notifyListeners(), work in build()/getters called per frame, repeated I/O, O(n) lookups that should be a Map/Set, redundant parsing.
Duplication — repeated logic → shared helper/extension/top-level function.
Readability — long functions, deep nesting, huge build() → split into methods or extracted widgets.
Dead code — unused private methods/fields/imports, unreachable branches.

Workflow

Scope first. If the request is broad, use AskUserQuestion to pin down: which categories, and which area (app/lib/services, app/lib/providers, whole app/lib, or packages/). Don't sweep blind — it burns tokens.
Map size. find app/lib -name '*.dart' | xargs wc -l | sort -rn | head -40 to find the heavy files.
Fan out (parallel). Dispatch Explore subagents — one per area (services / providers / widgets) — to return concrete findings: file:line + category + one-line problem + suggested fix. Tell them: high-confidence only, no style nitpicks, prioritize by impact.
Filter to safe wins. Apply ONLY behavior-preserving changes. Verify each finding by reading the actual code before editing (agents over-report). Good defaults:
- Merge identical methods; extract repeated predicates/loops into a helper.
- Guard notifyListeners() with an equality check (if (_x == v) return;).
- Replace long switch/.any() per-frame scans with Map/Set lookups.
- Extract a large ListView.builder row into a StatelessWidget (better element reuse); split a >150-line build() into private _buildX methods or sub-widgets.
- Pull repeated InputDecoration/styling into a helper.
Defer risky changes — surface them in the summary instead of doing them silently: public API/signature changes (e.g. connection pooling), widget splits tangled with lifecycle state (didUpdateWidget, post-frame callbacks), anything that could shift behavior.
Verify before claiming done (mandatory): cd app && flutter analyze → must be clean, then flutter test → all pass. Never report success without both.

Output

Group applied changes by category with file:line. List deferred high-value items separately with the reason they need sign-off. Ask before doing the risky batch.

Notes

Repo output (code, comments) is English-only; chat replies in Vietnamese.
Keep edits idiomatic to surrounding code (match naming, comment density).

YourSSH Refactor Sweep

Review the Flutter codebase for optimization opportunities and apply the safe, behavior-preserving ones. Optimize for high-confidence wins; defer risky changes for explicit sign-off.

When to use

User asks to "review và refactor", "tối ưu", "clean up", "find dead code" over an area or the whole project.

After a feature lands and the code needs a tidy pass.

Do not use when: the user wants a single targeted bug fix (use superpowers:systematic-debugging) or a behavioral change/new feature (use superpowers:brainstorming).

Four categories to scan

Performance — excess notifyListeners(), work in build()/getters called per frame, repeated I/O, O(n) lookups that should be a Map/Set, redundant parsing.

Duplication — repeated logic → shared helper/extension/top-level function.

Readability — long functions, deep nesting, huge build() → split into methods or extracted widgets.

Dead code — unused private methods/fields/imports, unreachable branches.

Workflow

Scope first. If the request is broad, use AskUserQuestion to pin down: which categories, and which area (app/lib/services, app/lib/providers, whole app/lib, or packages/). Don't sweep blind — it burns tokens.

Map size. find app/lib -name '*.dart' | xargs wc -l | sort -rn | head -40 to find the heavy files.

Fan out (parallel). Dispatch Explore subagents — one per area (services / providers / widgets) — to return concrete findings: file:line + category + one-line problem + suggested fix. Tell them: high-confidence only, no style nitpicks, prioritize by impact.

Filter to safe wins. Apply ONLY behavior-preserving changes. Verify each finding by reading the actual code before editing (agents over-report). Good defaults:

Merge identical methods; extract repeated predicates/loops into a helper.
Guard notifyListeners() with an equality check (if (_x == v) return;).
Replace long switch/.any() per-frame scans with Map/Set lookups.
Extract a large ListView.builder row into a StatelessWidget (better element reuse); split a >150-line build() into private _buildX methods or sub-widgets.
Pull repeated InputDecoration/styling into a helper.

Defer risky changes — surface them in the summary instead of doing them silently: public API/signature changes (e.g. connection pooling), widget splits tangled with lifecycle state (didUpdateWidget, post-frame callbacks), anything that could shift behavior.

Verify before claiming done (mandatory): cd app && flutter analyze → must be clean, then flutter test → all pass. Never report success without both.

Output

Group applied changes by category with file:line. List deferred high-value items separately with the reason they need sign-off. Ask before doing the risky batch.

Notes

Repo output (code, comments) is English-only; chat replies in Vietnamese.

Keep edits idiomatic to surrounding code (match naming, comment density).

yourssh-refactor-sweep