Skip to content

AI Agents

11 specialized sub-agents for focused code review, security analysis, architecture audits, migration generation, and quality scoring


AI Agents

Massu AI includes 11 specialized sub-agents that bring domain expertise to specific review dimensions. Unlike general-purpose AI review, each agent is tuned for a single concern — security, architecture, UX, patterns, or schema integrity — producing deeper, more actionable feedback.

Why Specialized Agents Matter

A general code review catches surface issues. A specialized security reviewer thinks like an attacker. A specialized architecture reviewer thinks about coupling, cohesion, and scalability. By decomposing review into focused passes, each agent goes deeper than a single general review ever could.

Massu's agents run in parallel during commands like /massu-review (which invokes 7 agents simultaneously) and /massu-loop (which uses the plan auditor and pattern reviewer at checkpoints).

Available Agents

Security Reviewer

Adversarial security-focused review that hunts for vulnerabilities. Checks OWASP categories, authentication flows, input validation, secrets exposure, SQL injection, XSS, CSRF, and authorization bypass.

Used by: /massu-review, /massu-push

Architecture Reviewer

Reviews code for architectural issues: coupling between modules, cohesion within modules, scalability patterns, dependency direction violations, and design principle compliance.

Used by: /massu-review, /massu-refactor

UX Reviewer

Evaluates user experience quality: WCAG accessibility, keyboard navigation, error messaging, loading states, empty states, and interaction patterns.

Used by: /massu-review, /massu-loop-playwright

Plan Auditor

Verifies implementation plans for completeness. Checks every deliverable against proof, identifies coverage gaps, missing verification steps, unclear acceptance criteria, and unaddressed edge cases.

Used by: /massu-plan, /massu-plan-audit, /massu-golden-path

Blast Radius Analyzer

When changing constants, exports, config keys, or tool names, this agent greps the entire codebase for all references and categorizes each as CHANGE (must update), KEEP (with reason), or INVESTIGATE (needs human decision). Zero INVESTIGATE items allowed before implementation starts.

Used by: /massu-loop, /massu-refactor

Pattern Reviewer

Automated pre-commit review that checks pattern compliance: ESM import style, config access via getConfig(), tool registration in tools.ts, hook stdin/stdout format, database access patterns, and memDb lifecycle management.

Used by: /massu-commit, /massu-push

Schema Sync Verifier

Compares database schemas across all environments (development, staging, production) and reports mismatches in tables, columns, types, constraints, indexes, and RLS policies.

Used by: /massu-push, /massu-release

Migration Writer

Generates SQL migrations following project patterns: proper ordering relative to existing migrations, RLS policies on new tables, rollback scripts, idempotent operations, and correct column types.

Used by: /massu-new-feature, /massu-loop

Output Scorer

Scores implementation quality across 5 dimensions — correctness, completeness, pattern compliance, security, and maintainability — and returns a structured scorecard with specific improvement suggestions.

Used by: /massu-loop, /massu-golden-path

Help Sync

Compares documentation against actual codebase features. Reports undocumented features, stale documentation that references removed code, missing examples, and broken links.

Used by: /massu-docs, /massu-review

Competitive Scorer

When multiple implementation approaches exist, this agent scores each across multiple quality dimensions to identify the strongest option with data-driven reasoning.

Used by: /massu-golden-path (competitive mode)

How Agents Are Invoked

Agents are invoked automatically by workflow commands. You don't call them directly — they're wired into the governance pipeline:

  • /massu-review spawns 7 agents in parallel (security, architecture, UX, patterns, performance, accessibility, AI-specific)
  • /massu-plan uses the plan auditor to verify coverage
  • /massu-commit runs the pattern reviewer before allowing commits
  • /massu-loop uses blast radius analysis and output scoring at checkpoints
  • /massu-push runs schema sync verification before push

All agents are included free in the open-source version.