Code Quality Checker

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 Checker

Type: L3 Worker Category: 5XX Quality

Analyzes Done implementation tasks with quantitative Code Quality Score based on metrics, MCP Ref validation, and issue penalties.

Inputs

Resolution: Story Resolution Chain. Status filter: In Progress, To Review

MANDATORY READ: Load shared/​references/​mcp_tool_preferences.md and shared/​references/​mcp_integration_patterns.md - use hex-graph as the primary path for clone, architecture, and semantic quality evidence when the project is indexed. Fall back to Grep/​Read only when graph is unavailable or unsupported.

Purpose & Scope

• Load Story and Done implementation tasks (exclude test tasks)
• Calculate Code Quality Score using metrics and issue penalties
• MCP Ref validation: Verify optimality, best practices, and performance via external sources
• Check for DRY/​KISS/​YAGNI violations, architecture boundary breaks, security issues
• Produce quantitative verdict with structured issue list; never edits Linear or kanban

Code Metrics

Code Quality Score

Formula: Code Quality Score = 100 - metric_penalties - issue_penalties

Issue penalties by severity:

Score interpretation:

Issue Prefixes

OPT- subcategories:

ARCH- subcategories:

PERF- subcategories:

MNT- subcategories:

When to Use

• All implementation tasks in Story status = Done
• Before tech debt cleanup and inline agent review

Workflow (concise)

MANDATORY READ: Load shared/​references/​input_resolution_pattern.md

1. Resolve storyId: Run Story Resolution Chain per guide (status filter: [In Progress, To Review]).

2. Load Story (full) and Done implementation tasks (full descriptions) via Linear; skip tasks with label "tests".

3. Collect changed files (changed_files[]): MANDATORY READ: Load shared/​references/​git_scope_detection.md

   • IF invoked by ln-510: use changed_files[] from coordinator context → proceed to Enrich step in guide
   • IF invoked standalone: run full algorithm from guide

4. Two-Layer Detection (MANDATORY): MANDATORY READ: Load shared/​references/​two_layer_detection.md All threshold-based findings require Layer 2 context analysis. Layer 1 finding without Layer 2 = NOT a valid finding. Before reporting any metric violation, ask: "Is this violation intentional or justified by design?" See Exception column in metrics below.

5. Calculate code metrics:

   • Cyclomatic Complexity per function (target ≤10; Exception: enum/​switch dispatch, state machines, parser grammars → downgrade to LOW)
   • Function size (target ≤50 lines; Exception: orchestrator functions with sequential delegation)
   • File size (target ≤500 lines; Exception: config/​schema/​migration files, generated code)
   • Nesting depth (target ≤3)
   • Parameter count (target ≤4; Exception: builder/​options patterns)

6. MCP Ref Validation (MANDATORY for code changes — SKIP if --skip-mcp-ref flag passed): MANDATORY READ: Load shared/​references/​research_tool_fallback.md

   Fast-track mode: When invoked with --skip-mcp-ref, skip this entire step (no OPT-, BP-, PERF- checks). Proceed directly to step 6 (static analysis). This reduces cost from ~5000 to ~800 tokens while preserving metrics + static analysis coverage.

   Level 1 — OPTIMALITY (OPT-):

   • Extract goal from task (e.g., "user authentication", "caching", "API rate limiting")
   • Research alternatives: ref_search_documentation("{goal} approaches comparison {tech_stack} 2026")
   • Compare chosen approach vs alternatives for project context
   • Flag suboptimal choices as OPT- issues

   Level 2 — BEST PRACTICES (BP-):

   • Research: ref_search_documentation("{chosen_approach} best practices {tech_stack} 2026")
   • For libraries: query-docs(library_id, "best practices implementation patterns")
   • Flag deviations from recommended patterns as BP- issues

   Level 3 — PERFORMANCE (PERF-):

   • PERF-ALG: Analyze algorithm complexity (detect O(n²)+, research optimal via MCP Ref)
   • PERF-CFG: Check library configs (connection pooling, batch sizes, timeouts) via query-docs
   • PERF-PTN: Research pattern pitfalls: ref_search_documentation("{pattern} performance bottlenecks")
   • PERF-DB: Check for N+1, missing indexes via query-docs(orm_library_id, "query optimization")

   Triggers for MCP Ref validation:

   • New dependency added (package.json/​requirements.txt changed)
   • New pattern/​library used
   • API/​database changes
   • Loops/​recursion in critical paths
   • ORM queries added

7. Analyze code for static issues (assign prefixes): MANDATORY READ: Load shared/​references/​clean_code_checklist.md, shared/​references/​destructive_operation_safety.md

   • For large code files, use outline(file_path) before targeted reads.
   • SEC-: hardcoded creds, unvalidated input, SQL injection, race conditions
   • SEC-DESTR-: unguarded destructive operations — use code-level guards table from destructive_operation_safety.md (loaded above). Check all 5 guard categories (DB, FS, MIG, ENV, FORCE).
   • MNT-: DRY violations (MNT-DRY-: duplicate logic), dead code (MNT-DC-: per checklist), complex conditionals, poor naming
   • MNT-DRY- cross-story hotspot scan: Grep for common pattern signatures (error handlers: catch.*Error|handleError, validators: validate|isValid, config access: getSettings|getConfig) across ALL src/ files (count mode). If any pattern appears in 5+ files, sample 3 files (Read 50 lines each) and check structural similarity. If >80% similar → MNT-DRY-CROSS (medium, -10 points): Pattern X duplicated in N files — extract to shared module.
   • MNT-DRY- preferred (hex-graph): If hex-graph indexed, use audit_workspace(path=scan_path, verbosity="minimal", limit=5, clone_member_limit=3). Each clone group with 2+ members in different files = MNT-DRY-CROSS. Raise limits only when the bounded preview is insufficient. Use returned hotspot and clone context for priority. Fall back to Grep pattern scan above if hex-graph unavailable.
   • MNT-DC- cross-story unused export scan: For each file modified by Story, count export declarations. Then Grep across ALL src/ for import references to those exports. Exports with 0 import references → MNT-DC-CROSS (medium, -10 points): {export} in {file} exported but never imported — remove or mark internal.
   • OPT-OSS- cross-reference ln-645 (static, fast-track safe): IF docs/​project/​.audit/​ln-640/​*/​645-open-source-replacer*.md exists (glob across dates, take latest), check if any HIGH-confidence replacement matches files changed in current Story. IF match found → create OPT-OSS-{N} issue with module path, goal, recommended package, confidence, stars, license from ln-645 report. Severity: high if >200 LOC, medium otherwise. This check reads local files only — no MCP calls — runs even with --skip-mcp-ref.
   • ARCH-: layer violations, circular dependencies, guide non-compliance
   • ARCH-LB-: layer boundary violations (HTTP/​DB/​FS calls outside infrastructure layer)
   • ARCH-TX-: transaction boundary violations (commit() across multiple layers)
   • ARCH-DTO-: missing DTOs (4+ repeated params), entity leakage (ORM entities returned from API)
   • ARCH-DI-: direct instantiation in business logic (no DI container or mixed patterns)
   • ARCH-CEH-: centralized error handling absent or bypassed
   • ARCH-SES-: session ownership conflicts (DI + local session in same module)
   • ARCH-AI-SEB: side-effect breadth (3+ categories in one leaf function; orchestrator functions exempt — see Conflict Resolution in table above)
   • ARCH-AI-AH: architectural honesty (read-named function with hidden writes)
   • ARCH-AI-FO: flat orchestration (leaf service importing 3+ services; orchestrator imports exempt)
   • ARCH-EVENT-: event channel mismatch — Grep for NOTIFY|pg_notify|\.publish\(|\.emit\( (publishers) and LISTEN|\.subscribe\(|\.on\( (subscribers) in changed_files[]. Cross-reference channel name strings. <!-- Defense-in-depth: also checked by ln-652 Rule 6 -->
   • MNT-GOD-: god classes (>15 methods or >500 lines per class)
   • MNT-SIG-: method signature quality (boolean flags, unclear returns)
   • MNT-ERR-: error contract inconsistency (mixed raise/​return patterns in same service)

8. Calculate Code Quality Score:

   • Start with 100
   • Subtract metric penalties (see Code Metrics table)
   • Subtract issue penalties (see Issue penalties table)

9. Output verdict with score and structured issues. MANDATORY READ: Load references/​output_schema.md Format output per schema. Add Linear comment with findings.

Critical Rules

• Read guides mentioned in Story/​Tasks before judging compliance.
• MCP Ref validation: For ANY architectural change, MUST verify via ref_search_documentation before judging.
• Context7 for libraries: When reviewing library usage, query-docs to verify correct patterns.
• Language preservation in comments (EN/​RU).
• Do not create tasks or change statuses; caller decides next actions.

Runtime Summary Artifact

MANDATORY READ: Load shared/​references/​quality_summary_contract.md, shared/​references/​quality_worker_runtime_contract.md

Runtime profile:

• family: quality-worker
• worker: ln-511
• summary kind: quality-worker
• payload fields used by coordinators: worker, status, verdict, score, issues, warnings

Invocation rules:

• standalone: omit runId and summaryArtifactPath
• managed: pass both runId and exact summaryArtifactPath
• always write the validated summary before terminal outcome

Definition of Done

• Story and Done implementation tasks loaded (test tasks excluded)
• Code metrics calculated (Cyclomatic Complexity, function/​file sizes)
• MCP Ref validation completed (OPT-, BP-, PERF- categories)
• ARCH- subcategories checked (LB, TX, DTO, DI, CEH, SES, EVENT); MNT- subcategories checked (DC, DRY, GOD, SIG, ERR)
• Issues identified with prefixes and severity, sources from MCP Ref/​Context7
• Code Quality Score calculated
• Output formatted per references/​output_schema.md
• Linear comment posted with findings

Reference Files

• Git scope detection: shared/​references/​git_scope_detection.md
• Code metrics: references/​code_metrics.md (thresholds and penalties)
• Guides: docs/​guides/
• Templates for context: shared/​templates/​task_template_implementation.md
• Clean code checklist: shared/​references/​clean_code_checklist.md
• Research tool fallback: shared/​references/​research_tool_fallback.md

---

Version: 5.1.0 Last Updated: 2026-03-15