Supercharge Claude Code with specialized AI agents for any framework.
Scaffold projects, build features, review architecture, migrate legacy code, and hunt bugs — all following your project's conventions automatically.
Specialist Agent is a collection of AI agents, skills, and architectural conventions for Claude Code. It's not a library — it's a set of markdown instructions that make Claude Code work like a senior developer who knows your codebase.
Framework packs provide stack-specific agents, skills, and architecture patterns:
| Pack | Stack |
|---|---|
| Vue 3 | Vue 3 + TypeScript + Pinia + TanStack Vue Query |
| React | React 18 + TypeScript + Zustand + TanStack React Query |
| Next.js | Next.js 14+ (App Router) + TypeScript + Zustand + Server Components |
| SvelteKit | SvelteKit 2 + TypeScript + Svelte stores + load functions |
Agents consume tokens proportional to their complexity. Specialist Agent ships with Lite agents that use model: haiku — significantly cheaper per token.
| Mode | Model | Module scaffold | Code review | Bug investigation |
|---|---|---|---|---|
| Full | Sonnet/Opus | ~15-25k tokens | ~8-15k tokens | ~5-10k tokens |
| Lite | Haiku | ~5-10k tokens | ~3-5k tokens | ~2-5k tokens |
1. Install into your project
cd /path/to/your-project
npx specialist-agent init
The wizard asks:
2. Start using
claude
# Create a new project from scratch
"Use @starter to create a task-manager app with Vue + Express + PostgreSQL"
# Build a feature inside an existing project
"Use @builder to create an orders module with CRUD for GET/POST/PATCH/DELETE /v2/orders"
| Agent | What it does |
|---|---|
| @starter | Create a new project from scratch — any frontend + backend + database |
| @explorer | Explore unfamiliar codebases — onboarding, technical assessments, health score |
| Agent | What it does |
|---|---|
| @finance | Payment integration, billing, invoicing, tax calculation, financial reporting |
| @cloud | Cloud architecture, IaC (Terraform/Pulumi), serverless, containers, CI/CD |
| @security | Auth flows (OAuth/JWT), OWASP compliance, RBAC/ABAC, encryption |
| @designer | Design systems, responsive layouts, accessibility (WCAG), theming |
| @data | Database modeling, migrations, caching strategies, query optimization |
| @devops | Docker, Kubernetes, CI/CD pipelines, monitoring, logging |
| @tester | Test strategies, test suites, coverage analysis, mocking patterns |
| Agent | What it does |
|---|---|
| @builder | Create modules, components, services, hooks/composables, and tests |
| @reviewer | Review PRs, explore modules, check performance |
| @doctor | Investigate bugs by tracing through architecture layers |
| Agent | What it does |
|---|---|
| @migrator | Migrate legacy code to modern patterns (per framework pack) |
| @reviewer | Diagnose current state before migration, validate after |
# Start a new full-stack project
"Use @starter to create an e-commerce app with Vue + Fastify + PostgreSQL"
# Explore an unfamiliar codebase
"Use @explorer to assess this project — I just joined the team"
# Build features
"Use @builder to create the orders module with CRUD for /v2/orders"
# Review before merging
"Use @reviewer to review the changes in the payments module"
# Debug issues
"Use @doctor to investigate the 500 error on login"
# Migrate legacy code (6 phases with approval gates)
"Use @migrator to migrate the billing module from Options API to script setup"
Quick shortcuts you invoke with /skill-name in Claude Code.
| Skill | What it does |
|---|---|
/dev-create-module [name] |
Full module scaffold (delegates to agents) |
/dev-create-component [name] |
Component with script setup template |
/dev-create-service [resource] |
Types + contracts + adapter + service |
/dev-create-composable [name] |
Composable with server state integration |
/dev-create-test [file] |
Tests for adapter, composable, or component |
/dev-generate-types [endpoint] |
Types + contracts + adapter from endpoint/JSON |
| Skill | What it does |
|---|---|
/review-review [scope] |
Full code review (automated + manual) |
/review-check-architecture [module] |
Automated conformance checks |
/review-fix-violations [module] |
Auto-fix violations by priority |
| Skill | What it does |
|---|---|
/migration-migrate-component [file] |
Options API → script setup |
/migration-migrate-module [path] |
Full module migration (6 phases) |
| Skill | What it does |
|---|---|
/docs-onboard [module] |
2-minute module overview for onboarding |
All agents enforce the four-layer architecture defined in each pack:
Service (HTTP only) → Adapter (parse) → Orchestration → Component (UI)
Each pack adapts the orchestration layer:
| Pack | Orchestration | Client State | File Extension |
|---|---|---|---|
| Vue 3 | Composables + Vue Query | Pinia | .vue |
| React | Hooks + React Query | Zustand | .tsx |
| Next.js | Hooks + Server Actions | Zustand | .tsx |
| SvelteKit | Stores + load functions | Svelte stores | .svelte |
Full guide:
docs/ARCHITECTURE.md— the source of truth that all agents read.
For cost-conscious usage or simpler tasks:
npx specialist-agent init # select "Lite" mode in wizard
| Aspect | Full | Lite |
|---|---|---|
| Model | Sonnet/Opus (default) | Haiku |
| First action | Reads ARCHITECTURE.md | Rules inline |
| Validation | tsc, build, vitest | Skipped |
| Size | ~80-120 lines | ~30-50 lines |
| Cost | ~5-25k tokens | ~2-10k tokens |
specialist-agent/
├── packs/
│ ├── vue/ ← Vue 3 pack
│ ├── react/ ← React pack
│ ├── nextjs/ ← Next.js pack
│ └── svelte/ ← SvelteKit pack
│ ├── agents/ ← Full agents (4)
│ ├── agents-lite/ ← Lite agents (4)
│ ├── skills/ ← Skills (12)
│ ├── ARCHITECTURE.md ← Architecture patterns
│ └── CLAUDE.md ← Pack-specific config
│
├── agents/ ← Framework-agnostic agents (9 + 9 lite)
│
├── cli/ ← CLI installer
│ └── index.mjs ← npx specialist-agent init
│
├── docs/ ← VitePress documentation
└── package.json
For up-to-date library documentation when asking Claude about APIs, you can add the Context7 MCP server. This benefits you as a developer — agents work fully without it.
Add to your Claude Code MCP config (~/.claude/mcp.json):
{
"mcpServers": {
"context7": {
"command": "npx",
"args": ["-y", "@upstash/context7-mcp@latest"]
}
}
}
Create .claude/agents/my-agent.md following the 5-part blueprint (Mission, Workflow, Output, Rules, Handoff):
---
name: my-agent
description: "MUST BE USED to [do X] when [trigger]. Use PROACTIVELY when [condition]."
tools: Read, Write, Edit, Bash, Glob, Grep
---
# Title
## Mission
One sentence describing what this agent does.
## Workflow
1. Step one
2. Step two
## Output
After completing work, provide:
- What was done (files created/modified)
- Key decisions and rationale
- Validation results
- Recommendations for next steps
## Rules
- Rule one
- Rule two
## Handoff Protocol
- If [condition] → suggest @agent-name
Create .claude/skills/my-skill/SKILL.md:
---
name: my-skill
description: "What this skill does and when to use it"
user-invocable: true
argument-hint: "[arguments]"
allowed-tools: Read, Write, Edit, Bash, Glob, Grep
---
Instructions for the skill.
Target: $ARGUMENTS
## Steps
1. ...
Edit docs/ARCHITECTURE.md — all agents read this file before acting. Changes take effect immediately.
Full docs are available via VitePress:
cd specialist-agent
npm install && npm run docs:dev
Contributions are welcome!
git checkout -b feature/my-feature
git commit -m 'feat: add my feature'
git push origin feature/my-feature
# Open a Pull Request
MIT — Use freely. See LICENSE.
Built for developers who use Claude Code every day.
GitHub · Docs · Issues