Agents Reference

27+ specialized AI agents organized by purpose.

Cross-Cutting Features

All agents include:

• Verification Protocol — Proof-based verification before claiming completion
• Anti-Rationalization — Tables preventing common shortcuts and excuses
• Lite Mode — Cost-optimized Haiku variants (60-80% cheaper)
• Handoff Protocol — Structured agent-to-agent delegation

Quick Reference

Need	Agent
New project	@starter
New feature	@builder
Code review	@reviewer
Fix bug	@doctor or @debugger
Plan feature	@planner
Write tests	@tdd or @tester
API design	@api
Performance	@perf
Security	@security
Payments	@finance
Infrastructure	@cloud or @devops
Database	@data
Translations	@i18n
Documentation	@docs
Refactoring	@refactor
Dependencies	@deps
Compliance	@legal
Impact analysis	@ripple
Explore codebase	@explorer or @scout

---

Core Agents

@starter

Create projects from scratch.

"Use @starter to create an e-commerce app with Next.js + PostgreSQL"

Supports: Vue, React, Next.js, SvelteKit, Angular, Astro, Nuxt, Express, Fastify, PostgreSQL, MongoDB, and more.

---

@builder

Build modules, components, services, and tests.

# Full module
"Use @builder to create a products module with CRUD"

# Component
"Use @builder to create a ProductCard component"

# Service
"Use @builder to create the orders service for /v2/orders"

Creates: Types, adapters, services, hooks/composables, components, tests.

---

@reviewer

Review code before merge. Three checks in one:

1. Spec compliance — Does it meet requirements?
2. Code quality — Is it clean and tested?
3. Architecture fit — Does it follow patterns?

"Use @reviewer to review the auth module"

Output: Pass/Fail verdict with specific issues.

---

@doctor

Investigate bugs by tracing through layers.

"Use @doctor to investigate the 500 error on login"

Traces: Component → State → Adapter → Service → API

Finds root cause, not symptoms.

---

@migrator

Modernize legacy code in 6 phases.

"Use @migrator to migrate src/legacy/billing/"

Phases: Analysis → Structure → Types → Services → State → Components

Approval required between phases.

---

Workflow Agents

@planner

Plan features based on complexity.

Complexity	Planning
Trivial	Skip
Simple	Quick plan
Medium	Detailed plan
Complex	Full design

"Use @planner to plan user authentication"

---

@executor

Execute plans with checkpoints and quality gates.

• Tracks progress and costs
• Creates git checkpoints
• 5 quality gates with mandatory evidence
• Supports rollback

"Use @executor to implement the auth plan"

---

@tdd

Test-Driven Development.

1. RED — Write failing test
2. GREEN — Make it pass
3. REFACTOR — Improve code

"Use @tdd to implement calculateDiscount"

No code without failing test first.

---

@debugger

4-phase systematic debugging.

1. Gather — Collect evidence
2. Analyze — Form hypothesis
3. Test — Verify hypothesis
4. Implement — Fix and validate

"Use @debugger to fix the stale data issue"

---

@pair

Real-time pair programming.

• Thinks out loud
• Catches mistakes early
• Suggests improvements

"Use @pair while I work on the checkout flow"

---

Specialist Agents

@api

Design REST and GraphQL APIs.

"Use @api to design the orders API with OpenAPI spec"

Creates: OpenAPI specs, GraphQL schemas, endpoint documentation.

---

@perf

Optimize performance.

"Use @perf to analyze and optimize the dashboard"

Analyzes: Bundle size, runtime, network, rendering.

---

@security

Application security.

"Use @security to implement JWT auth with refresh tokens"
"Use @security to audit for OWASP vulnerabilities"

Covers: Auth, RBAC/ABAC, encryption, vulnerability scanning.

---

@finance

Financial systems.

"Use @finance to integrate Stripe payments"

Covers: Payments, subscriptions, invoicing, reporting.

---

@cloud

Cloud architecture.

"Use @cloud to set up AWS with Terraform"

Covers: AWS, GCP, Azure, Terraform, Pulumi, serverless.

---

@data

Database engineering.

"Use @data to design the schema with Prisma"

Covers: Schema design, migrations, caching, query optimization.

---

@devops

DevOps and infrastructure.

"Use @devops to create Docker and Kubernetes config"

Covers: Docker, K8s, CI/CD, monitoring, logging.

---

@i18n

Internationalization.

"Use @i18n to add multi-language support"

Covers: Translations, locale management, RTL support.

---

@docs

Documentation generation.

"Use @docs to generate API documentation"

Creates: README, API docs, inline documentation.

---

@refactor

Code refactoring.

"Use @refactor to clean up the utils module"

Applies: Extract method, remove duplication, improve naming.

---

@deps

Dependency management.

"Use @deps to audit and update dependencies"

Covers: Security audit, outdated check, unused deps.

---

@legal

Data privacy compliance.

"Use @legal to review for GDPR compliance"

Covers: GDPR, LGPD, CCPA, consent management.

---

@tester

Testing specialist.

"Use @tester to create tests for the orders module"

Creates: Unit tests, integration tests, E2E tests.

---

@designer

UI/UX implementation.

"Use @designer to create a design system with dark mode"

Covers: Design tokens, responsive layouts, accessibility.

---

@ripple

Cascading effect analysis.

"Use @ripple to analyze the impact of renaming UserService"
"Use @ripple before changing the API response shape"

Analyzes: Direct dependents, indirect dependents, test coverage, breaking changes. Produces safe change plans ordered by dependency.

---

Support Agents

@scout

Quick project analysis. Recommends which agents to use.

"Use @scout to analyze this project"

Ultra-light. ~500 tokens.

---

@analyst

Convert business requirements to technical specs.

"Use @analyst to convert these requirements to a technical spec"

---

@orchestrator

Coordinate multiple agents.

"Use @orchestrator to build the feature with parallel agents"

---

@memory

Session memory management.

"Use @memory to save this decision"
"Use @memory to recall previous decisions"

---

@explorer

Explore unfamiliar codebases.

"Use @explorer to map this codebase — I'm new here"

Output: Health score, structure map, recommendations.

---

Verification Protocol

All key agents verify claims with evidence before marking work complete.

CLAIM: "Tests pass"
PROOF: Must show actual test output (PASS/FAIL)

CLAIM: "Build succeeds"
PROOF: Must show build command output

CLAIM: "Bug is fixed"
PROOF: Must show test that reproduces bug now passing

No "should work" or "probably fine". Run the command, show the output.

See the /verify skill for the full verification framework.

---

Anti-Rationalization

Key agents include rationalization prevention tables:

Excuse	Reality
"It's obvious, no need to test"	Obvious to you, not to the computer. Run the test.
"Just this once"	"Just this once" is how every bad habit starts.
"It should work"	"Should" is not evidence. Prove it.
"I'll verify later"	Later never comes. Verify now.

If you catch yourself thinking "just this once" — stop and follow the process.

---

Lite Versions

All agents have lite versions using the Haiku model.

Aspect	Full	Lite
Model	Sonnet/Opus	Haiku
Cost	Higher	60-80% less
Best for	Complex tasks	Quick tasks

Choose mode during installation:

npx specialist-agent init  # Select "Lite"