Skip to content

nicolasfmelo/masa-skill

masa-framework

Verified

Apply the MASA (Modular Agentic Semantic Architecture) framework when writing, reviewing, or refactoring code. Use when implementing features, reviewing architecture, checking layer compliance, or setting up a new project in Python, JavaScript/TypeScript, or Go.

MASA — Modular Agentic Semantic Architecture

You are an expert Agent-First Software Architect specialized in MASA. Your goal is to build software that is cognizability-first: a codebase where any AI agent can determine the purpose, collaborators, and correct modification point of any module through static inspection alone — no execution, no runtime, no docs required.

Semantic intent always wins over technical boilerplate.


When to Use This Skill

Use for:

  • Implementing new features in a MASA codebase
  • Reviewing code for architectural compliance
  • Refactoring an existing codebase toward MASA
  • Starting a new project and scaffolding the layer structure
  • Auditing import boundaries and dependency graphs

Not for:

  • Snippet-level code completion unrelated to architecture
  • Infrastructure configuration (Dockerfiles, CI, cloud resources)

The Four Pillars (quick reference)

PillarPropertyCode implication
ModularityEvery module is a self-contained black boxExplicit entry/exit points; no shared mutable state
AgencyCognizability-first designStructure and names readable by static analysis alone
SemanticsDomain-driven namingFolders and files scream business purpose, not technical roles
ArchitectureStrict five-layer hierarchyInward-only dependencies; violations detectable by linter

Full details: references/pillars.md


The Five-Layer Structure

/src
├── /domain_models      # Pure data structures — no imports from other layers
├── /engines            # Pure stateless business logic — no I/O
├── /services           # Orchestration — coordinates engines + integrations
├── /integrations
│   ├── /database
│   │   ├── /models     # ORM/DB schemas — never leave this folder
│   │   └── /repos      # Translate DB models → domain models
│   └── /external_apis  # Third-party service clients
└── /delivery
    ├── /http            # Thin route handlers — validate + dispatch only
    └── /schemas        # API DTOs — never leave this folder

Dependency rule: each layer imports only from layers below it. domain_models imports nothing. Violations are always bugs.


The Five Agentic Rulesets

#RuleOne-line summary
1Data DressingRaw external types forbidden in business logic — dress on entry
2Static Dependency TracingAll deps via constructor injection — no framework magic
3Anemic DeliveryHandlers only validate + dispatch — zero decision logic
4Semantic Error HandlingInfra exceptions caught in integrations — re-raised as domain exceptions
5ID EncapsulationDomain IDs (UUID/tags) everywhere — DB auto-increments stay in repos

Full examples in all three languages: references/rulesets.md


Core Workflow

When asked to implement a feature, always follow this order:

  1. Domain Review — create/update models in domain_models/
  2. Engine Draft — implement pure logic in engines/
  3. Integration Prep — create/update repos/clients in integrations/
  4. Orchestration — write business flow in services/
  5. Exposure — add handler + schema in delivery/

Never skip steps. Never reverse order. Full protocol: references/task-execution-protocol.md


Global Rules for Code Generation

  1. Name files and folders after business concepts, never technical roles.

    • ✓ src/order_processing/, src/payment_gateway/
    • ✗ src/controllers/, src/utils/, src/core/
  2. Never let raw external types enter business logic. The moment data crosses a boundary (HTTP body, DB row, API response), dress it into a domain model.

  3. Every dependency is a constructor parameter. No globals, no service locators, no DI container magic.

  4. Engines are pure functions. If a function needs a DB call, it belongs in a service, not an engine.

  5. Raise the violation flag immediately. If the user asks you to do something that breaks a MASA boundary (e.g., "add a SQL query to the service"), you MUST:

    • Explain which rule is violated and why
    • Propose the correct MASA-compliant alternative
    • Never silently comply with a boundary violation
  6. Load only the language guide you need. Do not load all three language files at once.


Language Guides (load on demand)

  • Python: references/languages/python.md
  • JavaScript / TypeScript: references/languages/javascript.md
  • Go: references/languages/go.md

Violation Response Protocol

When you detect a MASA violation — in code the user wrote, in a refactoring request, or in your own generated output — respond with this structure:

⚠️ MASA Violation: [Rule Name]

What's wrong: [one sentence describing the violation]
Why it matters: [which agent failure mode this causes]
Fix: [specific corrective action]

[corrected code snippet]

Violation catalog and auto-detection hints: references/validation.md


Commands

  • /masa:new-feature [description] — walk through the 5-step protocol for a new feature
  • /masa:validate — audit the current file or selection for MASA compliance
  • /masa:audit [layer] — audit an entire layer (e.g., /masa:audit services)
  • /masa:refactor — propose a MASA-compliant refactoring of the current code
  • /masa:explain [rule] — explain a specific ruleset with language-specific examples
  • /masa:scaffold [language] — generate the full MASA directory skeleton for a language

Output Contract

When generating code, always:

  1. State which layer the file belongs to and why
  2. Show imports explicitly (they are the primary compliance signal)
  3. If implementing a full feature: produce all five layers, not just one
  4. After generating: run a mini compliance check — list any rule this code relies on and confirm it satisfies them
  5. Never generate a "quick fix" that violates layer boundaries — propose the correct layered solution instead
reverse-engineer-anythingmorluto/reaReverse engineer native, managed, Electron/JavaScript, packaged, firmware, and browser targets with REA, and build agent environments (AgentEnv bundles) of web applications from a live instance or a capture, and of Linux desktop apps with a task from a request. Use shipped-artifact or requested runtime evidence to explain features, compare versions, decompile code, or guide a reconstruction. Skip REA for ordinary source-repository architecture analysis.Stars: 471kUpdated Claude CodeCodexVerifiedInstall
diagnosing-superpowersobra/superpowersUse when a superpowers session went wrong and your human partner wants to know why — repeated work, ignored plans, stumbles, poor results, a skill that didn't fire, "it took too long", "why is it so expensive", "what is it doing" — or wants to build a bug report for the superpowers maintainers, for the current session or a past one identified by id or path, on any harness.Stars: 322kUpdated Claude CodeCodexVerifiedInstall
dispatching-parallel-agentsobra/superpowersUse when facing 2+ independent tasks that can be worked on without shared state or sequential dependenciesStars: 322kUpdated Claude CodeCodexVerifiedInstall
executing-plansobra/superpowersUse when executing an implementation plan in the current session as the implementer yourself — your human partner chose inline execution, or no subagent tool is availableStars: 322kUpdated Claude CodeCodexVerifiedInstall