Paths: File paths (

shared/

,

references/

,

../ln-*

) are relative to skills repo root. If not found at CWD, locate this SKILL.md directory and go up one level for repo root. If

shared/

is missing, fetch files via WebFetch from

https://raw.githubusercontent.com/levnikolaevich/claude-code-skills/master/skills/{path}

.

Code Quality Auditor (L3 Worker)

Type: L3 Worker

Specialized worker auditing code complexity, method signatures, algorithms, and constants management.

Purpose & Scope

• Audit code quality (Categories 5+6+NEW: Medium Priority)
• Check complexity metrics, method signature quality, algorithmic efficiency, constants management
• Return structured findings with severity, location, effort, recommendations
• Calculate compliance score (X/10) for Code Quality category

Inputs

MANDATORY READ: Load

shared/references/audit_worker_core_contract.md

Receives

contextStore

tech_stack

best_practices

principles

codebase_root

output_dir

Domain-aware: Supports

domain_mode

current_domain

audit_output_schema.md#domain-aware-worker-output

Workflow

MANDATORY READ: Load

shared/references/two_layer_detection.md

1. Parse context -- extract fields, determine

   scan_path

   (domain-aware if specified), extract

   output_dir

2. Scan codebase for violations (Layer 1)
   • All Grep/Glob patterns use

     scan_path

     (not codebase_root)
   • Graph acceleration (if available): IF

     contextStore.graph_indexed

     OR

     .hex-skills/codegraph/index.db

     exists:
     • Complexity + God classes:

       audit_workspace(path=scan_path, verbosity="full")

       -- use returned hotspots to pre-identify complex functions and god classes.
     • Module metrics:

       analyze_architecture(path=scan_path, verbosity="full")

       -- use returned coupling metrics for cascade depth and coupling analysis.
     • Fall back to grep patterns below if graph unavailable.
     • Outline-first read:

       outline(file_path)

       before reading large source files -- understand function/class structure for complexity analysis.
   • Example:

     Grep(pattern="if.*if.*if", path=scan_path)

     for nesting detection
3. Analyze context per candidate (Layer 2 -- MANDATORY) Layer 1 finding without Layer 2 = NOT a valid finding. Before reporting, ask: "Is this violation intentional or justified by design?"
   • Cyclomatic complexity: is complexity from switch/case on enum (valid) or deeply nested conditions (bad)? Enum dispatch -> downgrade to LOW or skip
   • O(n^2): read context -- what's n? If bounded (n < 100), downgrade severity
   • N+1: read ORM config -- does it have eager loading configured elsewhere? Admin-only endpoint -> downgrade severity
   • God class: is it a config/schema/builder class? -> downgrade
   • Cascade depth: already traces calls (implicit Layer 2). Orchestrator function -> SEB does NOT apply (see Conflict Resolution in ARCH-AI-SEB)
4. Collect findings with severity, location, effort, recommendation
   • Tag each finding with

     domain: domain_name

     (if domain-aware)
5. Calculate score using penalty algorithm
6. Write Report: Build full markdown report in memory per

   shared/templates/audit_worker_report_template.md

   , write to

   {output_dir}/ln-624--{domain}.md

   (or

   624-quality.md

   in global mode) in single Write call
7. Return Summary: Return minimal summary to coordinator (see Output Format)

Audit Rules (Priority: MEDIUM)

1. Cyclomatic Complexity

What: Too many decision points in single function (> 10)

Detection:

• Count if/else, switch/case, ternary, &&, ||, for, while
• Use tools:

  eslint-plugin-complexity

  ,

  radon

  (Python),

  gocyclo

  (Go)

Severity:

• HIGH: Complexity > 20 (extremely hard to test)
• MEDIUM: Complexity 11-20 (refactor recommended)
• LOW: Complexity 8-10 (acceptable but monitor)
• Downgrade when: Enum/switch dispatch, state machines, parser grammars -> downgrade to LOW or skip

Recommendation: Split function, extract helper methods, use early returns

Effort: M-L (depends on complexity)

2. Deep Nesting (> 4 levels)

What: Nested if/for/while blocks too deep

Detection:

• Count indentation levels
• Pattern: if { if { if { if { if { ... } } } } }

Severity:

• HIGH: > 6 levels (unreadable)
• MEDIUM: 5-6 levels
• LOW: 4 levels
• Downgrade when: Nesting from early-return guard clauses (structurally deep but linear logic) -> downgrade

Recommendation: Extract functions, use guard clauses, invert conditions

Effort: M (refactor structure)

3. Long Methods (> 50 lines)

What: Functions too long, doing too much

Detection:

• Count lines between function start and end
• Exclude comments, blank lines

Severity:

• HIGH: > 100 lines
• MEDIUM: 51-100 lines
• LOW: 40-50 lines (borderline)
• Downgrade when: Orchestrator functions with sequential delegation; data transformation pipelines -> downgrade

Recommendation: Split into smaller functions, apply Single Responsibility

Effort: M (extract logic)

4. God Classes/Modules (> 500 lines)

What: Files with too many responsibilities

Detection:

• Count lines in file (exclude comments)
• Check number of public methods/functions

Severity:

• HIGH: > 1000 lines
• MEDIUM: 501-1000 lines
• LOW: 400-500 lines
• Downgrade when: Config/schema/migration files, generated code, barrel/index files -> skip

Recommendation: Split into multiple files, apply separation of concerns

Effort: L (major refactor)

5. Too Many Parameters (> 5)

What: Functions with excessive parameters

Detection:

• Count function parameters
• Check constructors, methods

Severity:

• MEDIUM: 6-8 parameters
• LOW: 5 parameters (borderline)
• Downgrade when: Builder/options pattern constructor; framework-required signatures (middleware, hooks) -> skip

Recommendation: Use parameter object, builder pattern, default parameters

Effort: S-M (refactor signature + calls)

6. O(n^2) or Worse Algorithms

What: Inefficient nested loops over collections

Detection:

• Nested for loops:

  for (i) { for (j) { ... } }

• Nested array methods:

  arr.map(x => arr.filter(...))

Severity:

• HIGH: O(n^2) in hot path (API request handler)
• MEDIUM: O(n^2) in occasional operations
• LOW: O(n^2) on small datasets (n < 100)
• Downgrade when: Bounded n (n < 100 guaranteed by domain); one-time init/migration code -> downgrade to LOW or skip

Recommendation: Use hash maps, optimize with single pass, use better data structures

Effort: M (algorithm redesign)

7. N+1 Query Patterns

What: ORM lazy loading causing N+1 queries

Detection:

• Find loops with database queries inside
• Check ORM patterns:

  users.forEach(u => u.getPosts())

Severity:

• CRITICAL: N+1 in API endpoint (performance disaster)
• HIGH: N+1 in frequent operations
• MEDIUM: N+1 in admin panel
• Downgrade when: Admin-only endpoint called <=1x/day -> downgrade to LOW. Eager loading configured elsewhere in ORM -> skip

Recommendation: Use eager loading, batch queries, JOIN

Effort: M (change ORM query)

8. Constants Management (NEW)

What: Magic numbers/strings, decentralized constants, duplicates

Detection:

if (status === 2)

if (role === 'admin')

MAX_SIZE = 100

STATUS_ACTIVE = 1

constants.ts

config.py

Severity:

• HIGH: Magic numbers in business logic (payment amounts, statuses)
• MEDIUM: Duplicate constants (same value defined 3+ times)
• MEDIUM: No central constants file
• LOW: Magic strings in logging/debugging
• Downgrade when: HTTP status codes (200, 404, 500) -> skip. Math constants (0, 1, -1) in algorithms -> skip. Test data -> skip

Recommendation:

• Create central constants file (

  constants.ts

  ,

  config.py

  ,

  constants.go

  )
• Extract magic numbers to named constants:

  const STATUS_ACTIVE = 1

• Consolidate duplicates, import from central file
• Use enums for related constants

Effort: M (extract constants, update imports, consolidate)

9. Method Signature Quality

What: Poor method contracts reducing readability and maintainability

Detection:

def process(data, is_async: bool, skip_validation: bool)

def query(db, limit=10, offset=0, sort="id", order="asc")

get_user()

fetch_account()

load_profile()

-> dict

-> Any

-> tuple

def get_stats() -> dict

-> StatsResponse

Severity:

• MEDIUM: Boolean flag params (use enum/strategy), unclear return types
• LOW: Too many optional params, inconsistent naming

Recommendation:

• Boolean flags: replace with enum, strategy pattern, or separate methods
• Optional params: group into config/options dataclass
• Naming: standardize verb conventions per module (

  get_

  for sync,

  fetch_

  for async, etc.)
• Return types: use TypedDict, NamedTuple, or dataclass instead of raw dict/tuple

Effort: S-M (refactor signatures + callers)

10. Side-Effect Cascade Depth

What: Functions triggering cascading chains of external side-effects (DB writes -> notifications -> metrics -> limits).

Detection: MANDATORY READ: Load

shared/references/ai_ready_architecture.md

• Glob

  **/services/**/*.{py,ts,js,cs,java}

  to find service files
• For each public function: check body for side-effect markers (per reference)
• Recursively follow called internal functions for additional markers
• Calculate max chain depth from entry point

Severity:

• HIGH: cascade_depth >= 4
• MEDIUM: cascade_depth = 3
• OK: depth <= 2
• Downgrade when: Orchestrator/coordinator functions (imports 3+ services AND delegates sequentially) -> skip. Depth from opaque sinks (logging, metrics) -> skip

Conflict Resolution: IF function is an orchestrator/coordinator (imports 3+ services AND delegates to them sequentially) -> ARCH-AI-SEB does NOT apply. Orchestrators are EXPECTED to have multiple side-effect categories. Only flag SEB for leaf functions.

Recommendation: Refactor to flat orchestration -- extract side-effects into independent sink functions. See reference.

Effort: M-L

Output: Also generate summary Pipe/Sink table per module:

Scoring Algorithm

MANDATORY READ: Load

shared/references/audit_worker_core_contract.md

shared/references/audit_scoring.md

Output Format

MANDATORY READ: Load

shared/references/audit_worker_core_contract.md

shared/templates/audit_worker_report_template.md

Write JSON summary per

shared/references/audit_summary_contract.md

runId

summaryArtifactPath

Write report to

{output_dir}/ln-624--{domain}.md

624-quality.md

category: "Code Quality"

Return summary per

shared/references/audit_summary_contract.md

When

summaryArtifactPath

.hex-skills/runtime-artifacts/runs/{run_id}/evaluation-worker/{worker}--{identifier}.json

Report written: .hex-skills/runtime-artifacts/runs/{run_id}/audit-report/ln-624--orders.md
Score: X.X/10 | Issues: N (C:N H:N M:N L:N)

Critical Rules

MANDATORY READ: Load

shared/references/audit_worker_core_contract.md

• Do not auto-fix: Report only
• Domain-aware scanning: If

  domain_mode="domain-aware"

  , scan ONLY

  scan_path

  (not entire codebase)
• Tag findings: Include

  domain

  field in each finding when domain-aware
• Context-aware: Small functions (n < 100) with O(n^2) may be acceptable
• Constants detection: Exclude test files, configs, examples
• Metrics tools: Use existing tools when available (ESLint complexity plugin, radon, gocyclo)

Definition of Done

MANDATORY READ: Load

shared/references/audit_worker_core_contract.md

• contextStore parsed (including domain_mode, current_domain, output_dir)
• scan_path determined (domain path or codebase root)
• All 10 checks completed (scoped to scan_path):
  • complexity, nesting, length, god classes, parameters, O(n^2), N+1, constants, method signatures, cascade depth
• Findings collected with severity, location, effort, recommendation, domain
• Score calculated
• Report written to

  {output_dir}/ln-624--{domain}.md

  (atomic single Write call)
• Summary written per contract

Reference Files

• Audit output schema:

  shared/references/audit_output_schema.md

Version: 3.0.0 Last Updated: 2025-12-23