Decision governance, fully equipped
From constitutional law system to compliance monitoring. 40+ AI tools and 10 industry templates to manage architecture decisions.
Constitutional Law System
The foundation of the entire system. Architecture decisions are treated as constitutional law — immutable, structured, and can only be created by humans.
- Decisions are immutable — cannot be modified or deleted once created
- Evolution only through supersedes with semantic versioning (v1.0.0 → v2.0.0)
- AI has ZERO approval authority — only humans can create and approve
- Every decision has an owner (created_by) who is accountable
- Standard format: domain, aspect, scope, blast_radius, constraints, invariants
- Append-only log — complete audit trail of every decision ever made
4×4 Decision Matrix
Every decision is classified in a 4 Domain × 4 Aspect matrix, producing 16 categories that cover all architecture aspects.
- INT (Intent & Direction) — Vision, goals, and strategic product direction
- ARCH (Architecture) — Technical structure, patterns, and tech stack
- CTL (Control & Policy) — Rules, standards, and policies to follow
- EVO (Evolution) — Changes, migrations, and system evolution roadmap
- Aspects A01–A16 — Vision, Scope, Stakeholder, Spec, Pattern, Component, Integration, Contract, Standard, Compliance, Monitoring, Enforcement, Strategy, Migration, Versioning, Deprecation
- Unique code per decision (e.g., ARCH-A06-001-v1.0.0) for clear reference
3-Gate Validation Pipeline
Every decision passes through 3 validation stages before being stored. No shortcuts.
- Gate 1 (Deterministic) — Schema validation, field rules, format checking. HARD block on failure
- Gate 2 (AI Heuristic) — Quality scoring, conflict detection, duplicate check. SOFT warning
- Gate 3 (Human Approval) — Human review. HARD block — AI cannot approve
- Integrated pipeline: draft → validate → approve → store
- Automatic duplicate detection based on semantic similarity
- Alignment check against existing decisions before creation
Smart Retrieval & MICS Engine
AI assistants can automatically retrieve relevant decisions based on code context. MICS (Multi-stage Intelligent Context Selection) ensures the right context is sent to AI.
- Context-aware search — analyzes file path, content, and scope for relevance
- MICS 7-stage pipeline — intent detection, scope resolution, decision retrieval, conflict check, priority ranking, context assembly, response formatting
- Trigger-based suggestions — decisions appear automatically when patterns are detected
- Knowledge graph — visualize decision relationships, pattern detection, and relationship traversal
- Hybrid search — keyword (Meilisearch) + semantic (vector embeddings) for maximum accuracy
- Hot decisions cache — frequently accessed decisions get priority
Constraint Enforcement
Decision constraints can be automatically enforced through regex checks, import rules, and CI/CD integration.
- Regex constraint checking — automatic pattern matching against source code
- Import rule enforcement — dependency and import pattern validation
- Violation tracking — every violation recorded with decision reference and severity
- mantra_review tool — 3-layer checking (regex → import → manual checklist)
- mantra_diff_review — parse git diff, check only changed lines
- mantra_scan — find // mantra:CODE annotations, validate coverage
10 Industry Templates
MANTRA can be used across industries. Each template adapts terminology, taxonomy, and field guidance to match the domain.
- Software Engineering — Default template for software architecture decisions (API, database, infrastructure)
- Healthcare — Clinical decision governance, medical protocol tracking, HIPAA/health regulation compliance
- Government — Policy governance, regulatory compliance, inter-agency decision coordination
- Finance — Financial regulation compliance, risk management framework, audit trail requirements
- Education — Curriculum decision governance, institutional policy management
- Manufacturing, Research, NGO, Startup, Corporate Law — 5 additional templates adapted per domain
40+ MCP Tools
Deep integration with Claude Code, Cursor, Windsurf, and other AI assistants through Model Context Protocol. 40+ tools covering the entire decision lifecycle.
- Decision Management — create, retrieve, list, group, draft wizard, simulate impact
- Compliance — review, diff_review, scan, multi_review (security + architecture + compliance perspectives), batch review
- Governance — approve, reject, pending_approvals, compliance_score, decision_health, violation tracking
- Intelligence — semantic_search, check_alignment, cross_references, get_patterns, explain
- Documentation — generate_document (14 types), generate_rules (9 IDE formats), get_checklist
- Learning — log_activity, search_memory, get_task_context, feedback, watch_start (continuous monitoring)
Document & IDE Export
Generate documents and IDE rules directly from decisions. One source of truth, many output formats.
- 14 document types — PRD, Tech Spec, API Spec, Security Spec, Test Plan, ADR, Exec Summary, User Manual, Feature Spec, Threat Model, Deployment Guide, Runbook, FAQ, Changelog
- 9 IDE rule formats — .cursorrules (Cursor), CLAUDE.md (Claude Code), Copilot, Windsurf, Aider, Cline, and universal format
- Decision export — JSON, Markdown, ADR format, SQL dump for backup and portability
- Generated directly from decision data — always in sync, never outdated
- Per-domain templates — output adapts to your industry's terminology and format
- Bulk generation — generate all documents at once for your entire scope of decisions
Decision Projections & Knowledge Graph
View decisions from multiple perspectives. Knowledge graph visualizes relationships and detects patterns across decisions.
- Timeline view — decision evolution over time
- Scope projection — decisions grouped by domain, aspect, or area
- Knowledge graph — node-edge relationship visualization with automatic pattern detection
- Tech stack analysis — decisions grouped by technology used
- Blast radius mapping — impact analysis and dependency tracking for each decision
- Relationship projection — supersedes chains, cross-references, and dependency graphs