perf: reduce token usage

[RuiMiguel]

Problem

One quick analysis on layered-architecture skill showed up that nearly 150 lines can be reduced, layered-architecture/SKILL.md shows inline code that is nearly identical to what's already in its own reference files.

UserApiClient — shown in SKILL.md:146-171 AND worked-example.md:78-117
// Both files show UserApiClient with constructor-injected http.Client,
// getUser method, statusCode != 200 check, UserResponse.fromJson

User domain model — shown in SKILL.md:193-208 AND worked-example.md:222-242

UserRepository — shown in SKILL.md:212-231 AND worked-example.md:262-288

ProfileBloc _onLoadRequested — shown in SKILL.md:286-299 AND worked-example.md:373-386

main_development.dart bootstrap — shown in SKILL.md:308-345 AND worked-example.md:469-489

pubspec.yaml dependency graph — shown in SKILL.md:240-270 AND pubspec.md:1-104 (same patterns, different domain names:
user vs weather). Then worked-example.md shows them a third time.

The SKILL.md already links to these references, but then shows 80-90% of the same code inline.

Solution

Run analysis on skills to try to apply some best practices:

• duplicate reference files,
• near identical instructions,
• repeated patterns,
• bash heavy sections → scripts
• prose tightening
  It would be good to run same analysis over other skills

Trim inline examples in SKILL.md to short signature-level snippets (3-5 lines showing the key pattern) and rely on the existing reference links for full implementations. This would cut ~120-150 lines from layered-architecture/SKILL.md without losing any content — it's all still available in the reference files.

Requirements

• All CI/CD checks are passing.
• There is no drop in the test coverage percentage.

Additional Context

No response