ln-642-layer-boundary-auditor

name	ln-642-layer-boundary-auditor
description	Checks layer boundary violations, transaction boundaries, session ownership, cross-layer consistency. Use when auditing architecture layers.
allowed-tools	Read, Grep, Glob, Bash, mcp__hex-graph__find_references, mcp__hex-graph__trace_paths, mcp__hex-graph__inspect_symbol, mcp__hex-graph__analyze_architecture, mcp__hex-line__read_file, mcp__hex-line__grep_search, mcp__hex-line__outline
license	MIT

Paths: File paths (references/, ../ln-*) are relative to this skill directory.

Layer Boundary Auditor

Type: L3 Worker

L3 Worker that audits architectural layer boundaries and detects violations.

Purpose & Scope

• Read architecture.md to discover project's layer structure
• Detect layer violations (I/O code outside infrastructure layer)
• Detect cross-layer consistency issues:
  • Transaction boundaries (commit/rollback ownership)
  • Session ownership (DI vs local)
• Check pattern coverage (all HTTP calls use client abstraction)
• Detect error handling duplication
• Return violations list to coordinator

Out of Scope (owned by ln-628-concurrency-auditor):

• Blocking I/O in async functions (sync open/read in async def)
• Fire-and-forget tasks (create_task without error handler)

Input (from ln-640)

- architecture_path: string    # Path to docs/architecture.md
- codebase_root: string        # Root directory to scan
- skip_violations: string[]    # Files to skip (legacy)
- output_dir: string           # e.g., ".hex-skills/runtime-artifacts/runs/{run_id}/audit-report"

# Domain-aware (optional, from coordinator)
- domain_mode: "global" | "domain-aware"   # Default: "global"
- current_domain: string                   # e.g., "users", "billing" (only if domain-aware)
- scan_path: string                        # e.g., "src/users/" (only if domain-aware)

When domain_mode="domain-aware": Use scan_path instead of codebase_root for all Grep/Glob operations. Tag all findings with domain field.

Workflow

Detection policy: use two-layer detection (candidate scan, then context verification); load references/two_layer_detection.md only when the verification method is ambiguous. Tool policy: follow host AGENTS.md MCP preferences; load references/mcp_tool_preferences.md and references/mcp_integration_patterns.md only when host policy is absent or MCP behavior is unclear.

Use hex-graph first when references, call paths, or architecture coupling materially improve the audit. Use hex-line first for local code and config reads when available. If MCP is unavailable, unsupported, or not indexed, continue with built-in Read/Grep/Glob/Bash and state the fallback in the report.

Phase 1: Discover Architecture

MANDATORY READ: Load ../ln-640-pattern-evolution-auditor/references/layer_rules.md -- use Architecture Presets (fallback), I/O Pattern Boundary Rules (Phase 2), Coverage Checks (Phase 4), Cross-Layer Consistency rules (Phase 3).

Read docs/architecture.md

Extract from Section 4.2 (Top-Level Decomposition):
  - architecture_type: "Layered" | "Hexagonal" | "Clean" | "MVC" | etc.
  - layers: [{name, directories[], purpose}]

Extract from Section 5.3 (Infrastructure Layer Components):
  - infrastructure_components: [{name, responsibility}]

IF architecture.md not found:
  Use fallback presets from layer_rules.md

Build ruleset:
  FOR EACH layer:
    allowed_deps = layers that can be imported
    forbidden_deps = layers that cannot be imported

Graph acceleration (if available): IF contextStore.graph_indexed OR .hex-skills/codegraph/index.db exists:

• Module coupling: analyze_architecture(path=scan_root, verbosity="full") -- use returned coupling metrics to identify tightly-coupled layers.
• Cross-layer calls: find_references(symbol) for transaction/session functions -- trace commit/rollback ownership across layers.
• Orchestration depth: trace_paths(name="ServiceFn", file="...", path_kind="calls", direction="forward", depth=3, path=scan_root) -- measure chain depth for flat orchestration check from a concrete service symbol.
• Empty trace_paths from a coarse or module-level selector is not enough to clear a layer; fall back to inspect_symbol or grep/manual review when the selector is broad.
• Fall back to grep-based detection below if graph unavailable.

Phase 2: Detect Layer Violations

scan_root = scan_path IF domain_mode == "domain-aware" ELSE codebase_root

FOR EACH violation_type IN layer_rules.md I/O Pattern Boundary Rules:
  grep_pattern = violation_type.detection_grep
  forbidden_dirs = violation_type.forbidden_in

  matches = Grep(grep_pattern, scan_root, include="*.py,*.ts,*.js")

  FOR EACH match IN matches:
    IF match.path NOT IN skip_violations:
      IF any(forbidden IN match.path FOR forbidden IN forbidden_dirs):
        violations.append({
          type: "layer_violation",
          severity: "HIGH",
          pattern: violation_type.name,
          file: match.path,
          line: match.line,
          code: match.context,
          allowed_in: violation_type.allowed_in,
          suggestion: f"Move to {violation_type.allowed_in}"
        })

Phase 3: Cross-Layer Consistency Checks

3.1 Transaction Boundary Violations

What: commit()/rollback() called at inconsistent layers (repo + service + API)

Detection:

repo_commits = Grep("\.commit\(\)|\.rollback\(\)", "**/repositories/**/*.py")
service_commits = Grep("\.commit\(\)|\.rollback\(\)", "**/services/**/*.py")
api_commits = Grep("\.commit\(\)|\.rollback\(\)", "**/api/**/*.py")

layers_with_commits = count([repo_commits, service_commits, api_commits].filter(len > 0))

Safe Patterns (ignore):

• Comment "# best-effort telemetry" in same context
• File ends with _callbacks.py (progress notifiers)
• Explicit # UoW boundary comment

Violation Rules:

Condition	Severity	Issue
layers_with_commits >= 3	CRITICAL	Mixed UoW ownership across all layers
repo + api commits	HIGH	Transaction control bypasses service layer
repo + service commits	HIGH	Ambiguous UoW owner (repo vs service)
service + api commits	MEDIUM	Transaction control spans service + API

Exception: Saga pattern / distributed transactions with explicit compensating actions -> downgrade CRITICAL to MEDIUM. UoW boundary documented with // architecture decision or ADR -> skip.

Recommendation: Choose single UoW owner (service layer recommended), remove commit() from other layers

Effort: L (requires architectural decision + refactoring)

3.2 Session Ownership Violations

What: Mixed DI-injected and locally-created sessions in same call chain

Detection:

di_session = Grep("Depends\(get_session\)|Depends\(get_db\)", "**/api/**/*.py")
local_session = Grep("AsyncSessionLocal\(\)|async_sessionmaker", "**/services/**/*.py")
local_in_repo = Grep("AsyncSessionLocal\(\)", "**/repositories/**/*.py")

Violation Rules:

Condition	Severity	Issue
di_session AND local_in_repo in same module	HIGH	Repo creates own session while API injects different
local_session in service calling DI-based repo	MEDIUM	Session mismatch in call chain

Recommendation: Use DI consistently OR use local sessions consistently. Document exceptions (e.g., telemetry)

Effort: M

3.3 Flat Orchestration Violations

What: Service-layer functions calling other services that call yet other services -- deep orchestration chains.

Detection: MANDATORY READ: Load references/ai_ready_architecture.md -- map service imports, find chain depth.

Violation Rules:

Condition	Severity	Issue
Service chain >= 3 (A->B->C->D)	HIGH	Deep orchestration
Service chain = 2 (A->B->C)	MEDIUM	Consider flattening

Recommendation: Extract orchestrator calling all services at same level. Each service becomes a sink.

Effort: L

---

Phase 4: Check Pattern Coverage

# HTTP Client Coverage
all_http_calls = Grep("httpx\\.|aiohttp\\.|requests\\.", codebase_root)
abstracted_calls = Grep("client\\.(get|post|put|delete)", infrastructure_dirs)

IF len(all_http_calls) > 0:
  coverage = len(abstracted_calls) / len(all_http_calls) * 100
  IF coverage < 90%:
    violations.append({
      type: "low_coverage",
      severity: "MEDIUM",
      pattern: "HTTP Client Abstraction",
      coverage: coverage,
      uncovered_files: files with direct calls outside infrastructure
    })

# Error Handling Duplication
http_error_handlers = Grep("except\\s+(httpx\\.|aiohttp\\.|requests\\.)", codebase_root)
unique_files = set(f.path for f in http_error_handlers)

IF len(unique_files) > 2:
  violations.append({
    type: "duplication",
    severity: "MEDIUM",
    pattern: "HTTP Error Handling",
    files: list(unique_files),
    suggestion: "Centralize in infrastructure layer"
  })

Phase 5: Calculate Score

MANDATORY READ: Load references/audit_worker_core_contract.md and references/audit_scoring.md.

Phase 6: Write Report

MANDATORY READ: Load references/templates/audit_worker_report_template.md.

Write JSON summary per references/audit_summary_contract.md. In managed mode the caller passes both runId and summaryArtifactPath; in standalone mode the worker generates its own run-scoped artifact path per shared contract.

# Build markdown report in memory with:
# - AUDIT-META (standard penalty-based: score, counts)
# - Checks table (io_isolation, http_abstraction, error_centralization, transaction_boundary, session_ownership)
# - Findings table (violations sorted by severity)
# - DATA-EXTENDED: {architecture, coverage}

IF domain_mode == "domain-aware":
  Write to {output_dir}/642-layer-boundary-{current_domain}.md
ELSE:
  Write to {output_dir}/642-layer-boundary.md

Phase 7: Return Summary

Report written: .hex-skills/runtime-artifacts/runs/{run_id}/audit-report/642-layer-boundary-users.md
Score: 4.5/10 | Issues: 8 (C:1 H:3 M:4 L:0)

Critical Rules

Apply the already-loaded references/audit_worker_core_contract.md.

• Read architecture.md first - never assume architecture type
• Skip violations list - respect legacy files marked for gradual fix
• File + line + code - always provide exact location with context
• Actionable suggestions - always tell WHERE to move the code
• No false positives - verify path contains forbidden dir, not just substring

Definition of Done

Apply the already-loaded references/audit_worker_core_contract.md.

• Architecture discovered from docs/architecture.md (or fallback used)
• All violation types from layer_rules.md checked
• Cross-layer consistency checked:
  • Transaction boundaries analyzed (commit/rollback distribution)
  • Session ownership analyzed (DI vs local)
• Coverage calculated for HTTP abstraction + 2 consistency metrics
• Violations list with severity, location, suggestion
• If domain-aware: all Grep scoped to scan_path, findings tagged with domain
• Report written to {output_dir}/642-layer-boundary[-{domain}].md (atomic single Write call)
• Summary written per contract

Reference Files

• Layer rules: ../ln-640-pattern-evolution-auditor/references/layer_rules.md
• Scoring impact: use references/audit_scoring.md; pattern scoring manuals are owned by ln-641-pattern-analyzer

---

Version: 2.1.0 Last Updated: 2026-02-08