yuri-code-intelligence

Yuri Code Intelligence — Unified Refactoring + Review Layer

Synthesized from OpenClaw corpus: method-decomposition-refactoring, class-responsibility-realignment, web-presentation-pattern-selector, code-review-skill-agent

Doctrine

Refactoring without diagnosis is noise. Diagnosis without a repair plan is observation. Intelligence means: see what is wrong, know what to apply, execute with tests passing at every step.

Phase 1: Smell Diagnosis

Before any tool call or edit, answer:

Smell	Signal	Technique
Long Method	Method body requires a comment to understand a block	Extract Method, Replace Temp with Query
Dual-purpose Temp	A variable is assigned for two different purposes	Split Temporary Variable
Feature Envy	A method uses another class's data more than its own	Move Method, Extract Method
Large Class	Class has too many instance variables or responsibilities	Extract Class, Extract Subclass
Inappropriate Intimacy	Two classes know too much about each other's internals	Move Method + Move Field
Web Presentation Pattern	Inconsistency between how data is formatted/rendered	Pattern Selector (see Phase 3)

Phase 2: Refactoring Execution (Fowler Catalog)

Method-Level Techniques

Extract Method: When a code fragment can be grouped behind a name.

1. Create a new method named after the fragment's intent (what, not how)
2. Move the fragment into the new method
3. Pass any referenced local variables as parameters
4. If one variable is assigned: return it
5. Compile + test after every extraction

Replace Temp with Query: When a temp holds a side-effect-free expression.

1. Mark temp final — confirm single assignment
2. Extract right-hand side into a private query method
3. Replace all references to temp with method call
4. Remove temp declaration
5. Compile + test

Split Temporary Variable: When a temp is assigned twice for different purposes.

1. Rename at first declaration to reflect first use, mark final
2. Update references up to second assignment
3. At second assignment, declare new temp with original name
4. Compile + test. Repeat for further assignments.

Class-Level Techniques

Extract Class: When a class is doing the work of two.

1. Decide how to split responsibilities
2. Create a new class. Move relevant fields to it.
3. Create a reference from old class to new.
4. Move methods that reference the new class's fields.
5. Decide on exposure: delegate through old class or expose new class directly.

Move Method: When a method uses more data from another class than its own.

1. Examine methods called by candidate method — move them together or before
2. Declare the method in the target class
3. Copy the method body, adjust to reference target class's fields
4. In source class: delegate to target, or remove if caller can reference target directly
5. Compile + test at every step

Phase 3: Web Presentation Pattern Selection

When the task involves rendering, formatting, or displaying data in a UI:

Context	Pattern	When to Use
Complex conditional rendering	Strategy	Variation is fixed at construction time
Decorator/wrapper UI	Decorator	Behaviors layered dynamically
Data transformation pipeline	Pipeline	Sequential transforms with composable stages
Theme-driven display	Template Method	Skeleton fixed, subclass fills in the details
Component state management	State	Object behavior changes with internal state

Phase 4: Code Review Checklist

Run after any non-trivial change:

Correctness

• [ ] Does the code do what it claims to do?
• [ ] Are edge cases (empty input, nulls, boundary values) handled?
• [ ] Is error handling specific and actionable?

Security

• [ ] Are user inputs validated at system boundaries?
• [ ] Are secrets and tokens absent from code (env-var only)?
• [ ] Are SQL/shell commands parameterized (not concatenated)?

Clarity

• [ ] Can a new reader understand a method from its name alone?
• [ ] Are variable names intention-revealing (not i, j, tmp, data)?
• [ ] Do comments explain WHY, not WHAT?

Performance

• [ ] Are there N+1 query patterns? (loop that queries inside)
• [ ] Are expensive operations computed once, not repeated?
• [ ] Are results cached where the input is stable?

Logging

• [ ] Do logs not include PII, secrets, or stack traces in production?
• [ ] Are log levels used correctly (debug/info/warn/error)?

Delivery

Always return:

1. Smell diagnosis (what patterns were found)
2. Refactoring prescription (which techniques, in which order)
3. Step-by-step execution with test command between each step
4. Review checklist result for the final state

Session Notes

2026-05-29

• session: 349m | peak ctx: 71% | compacts: 4
• tools: Bash×268, Read×133, Edit×104, TodoWrite×12, Write×8, StructuredOutput×8, Workflow×2, ToolSearch×1, AskUserQuestion×1
• corrections: none
• errors: none

2026-05-17

• Created from corpus synthesis: 4 code quality skills merged into Musubi-native form
• Sources: method-decomposition-refactoring (Fowler catalog, Long Method), class-responsibility-realignment (Feature Envy, Extract Class), web-presentation-pattern-selector (UI rendering patterns), code-review-skill-agent (security, clarity, logging checklist)
• Corrections: none
• Errors: none