review-code-quality

Enter planning mode. Deep-review the codebase for code quality and maintainability issues. Use maximum parallelism — spawn explore agents for independent areas.

What to Look For

Magic numbers and strings

• Hardcoded numeric values that should be named constants or config entries (e.g., Sleep(150) — what does 150 mean?)
• Hardcoded strings for states, modes, or message types that should be constants
• Exception: small, obvious values like 0, 1, true, false, or well-known Windows constants in DllCall are fine

DRY violations

• Same logic duplicated across multiple files or functions with slight variations
• Copy-paste code that has drifted (same origin, different bugs fixed in each copy)
• Similar switch/case blocks in multiple places that handle the same set of cases

Separation of concerns

• Business logic mixed with rendering or I/O
• Data transformation happening inside GUI code
• Config validation scattered across files instead of centralized
• But respect the architecture — producers living in src/core/ and touching window store data is by design, not a concern violation

Problematic function design

• Functions with too many parameters (>5) suggesting a missing options object
• Boolean parameters that control branching (suggests two separate functions)
• Functions that return different types depending on conditions (hard to reason about)
• Side effects hidden in what looks like a pure query function

Error handling

• Silent failures that swallow errors with no logging or recovery
• Inconsistent error handling patterns (some functions throw, some return false, some log)
• Try/catch blocks that catch too broadly

Naming

• Misleading names (function does more or less than the name suggests)
• Inconsistent naming conventions within a module
• Abbreviations that aren't obvious without context

CRITICAL — Mega-Function Guidelines

Do NOT flag functions for refactoring based on line count alone. Length is not a quality issue.

SKIP refactoring when the function has:

• Single responsibility — does one coherent thing with many steps (e.g., "process full state", "render overlay", "apply update")
• Linear/sequential logic — steps must happen in order, error handling spans multiple steps (update installers, init sequences, transaction-like operations)
• Performance-critical path — rendering loops, tick functions, hot paths. Function call overhead matters.
• Switch/case event handler — a 150-line switch handling 10 event types is clearer than 10 handler functions with dispatch logic
• Mirrors external structure — code structure mirrors an API, protocol, or data format
• Well-commented sections — clear section comments within a long function often beat extraction
• Tested and stable — working code with passing tests. Refactoring introduces regression risk for no functional gain.

RECOMMEND refactoring only when:

• Multiple unrelated responsibilities in one function
• Repeated logic — same 20+ lines in multiple places with slight variations
• Deep nesting — 4+ levels of indentation making flow hard to follow
• Difficult to test — can't unit test pieces in isolation when you need to
• Hard to extend — adding functionality requires understanding the entire function

Risk/reward check before recommending any refactor:

• What bugs or maintenance problems has this function actually caused?
• Will extracted functions be reused, or just called from one place?
• Does splitting add indirection that hurts readability?
• If the answer to "what problem does this solve?" is just "it's long" — skip it.

Explore Strategy

Split by area for parallel scanning:

• src/gui/ — GUI rendering, state machine, overlay, input
• src/core/ — Producers (WinEventHook, Komorebi, pumps)
• src/shared/ — Window list, config, IPC, blacklist, theme, stats
• src/editors/ — Config/blacklist editors
• src/pump/ — EnrichmentPump subprocess
• Root src/ files — Entry points, launcher, installation, update

Exclude src/lib/ (third-party code).

Use query_interface.ps1 to compare module public surfaces — similar API shapes across files can reveal DRY violations or misplaced responsibilities. Use query_function.ps1 <funcName> to read function bodies when evaluating DRY violations without loading full files. Use query_callchain.ps1 <funcName> -Reverse to trace callers when checking whether misplaced logic is called from the expected module or from elsewhere.

Validation

After explore agents report back, validate every finding yourself. Code quality is subjective — what looks like a concern violation may be an intentional design choice for performance or simplicity.

For each candidate:

1. Cite evidence: "I verified by reading file.ahk lines X–Y" with actual code quoted.
2. Trace downstream impact: For DRY violations, show both copies and where they've drifted. For magic numbers, show what would change if the value needed updating.
3. Counter-argument: "What would make this fix unnecessary or counterproductive?" — Is the "duplication" actually simpler than the abstraction? Is the magic number only used once and obvious in context?
4. Observed vs inferred: Did you find the pattern by reading the code, or infer it from file/function names?

Plan Format

Group by category:

Order by maintenance impact: issues that make future changes error-prone first, purely cosmetic issues last.

Ignore any existing plans — create a fresh one.