feat: add enforce_architecture MCP tool — architectural guardrails for AI agents

[bntvllnt]

Summary

Add an enforce_architecture MCP tool that validates proposed changes against architectural rules. dependency-cruiser does this for humans — nobody does it for AI agents via MCP.

Motivation

• AI agents generate code in isolation, ignoring inter-file dependency constraints
• 45% of AI-generated code contains vulnerabilities, often from violating architectural boundaries
• dependency-cruiser has 18K+ GitHub stars proving demand for boundary enforcement
• Risk-tiered enforcement is the enterprise consensus for AI code governance

Proposed API

tool: enforce_architecture
input: {
  rules?: Array<{
    from: string,      // glob pattern: "src/controllers/**"
    to: string,        // glob pattern: "src/models/**"
    allowed: boolean   // false = this dependency is forbidden
  }>,
  filePath?: string,   // check specific file against rules
  autoDetect?: boolean // infer rules from existing module structure
}
output: {
  violations: Array<{
    from: string,
    to: string,
    rule: string,
    severity: "error" | "warning"
  }>,
  passed: boolean,
  suggestions: string[]
}

Features

1. User-defined rules: explicit allow/deny patterns
2. Auto-detected rules: infer boundaries from existing module structure + cohesion metrics
3. Pre-flight check: agent validates before writing imports
4. CI integration: run as part of quality gates

Acceptance Criteria

• Validates file imports against architectural rules
• Auto-detects module boundaries from existing graph
• Returns PASS/FAIL with violation details
• Integrates with module cohesion metrics
• Tests covering rule violations and edge cases

Priority

Short-term — Unique capability no competitor offers via MCP.