stackwise

Curated knowledge base of rules, skills, agents, and hooks for Claude Code. One command to set up any TypeScript project.

npx stackwise

What It Does

stackwise scans your package.json (including monorepo workspaces), detects your frameworks, tools, and libraries, then installs:

• Knowledge rules into .claude/rules/ with proper glob frontmatter
• Skills (slash commands) into .claude/skills/
• Agents (specialized AI agents) into .claude/agents/
• Hooks (automation scripts) into .claude/hooks/
• Settings preset merged into .claude/settings.json
• CLAUDE.md auto-generated with your stack and workflow info

No global install needed. Just npx stackwise in any project.

Supported Stacks

Auto-Detected Frameworks & Tools

| Category | Detected | |----------|----------| | Frontend | Next.js, React SPA, Vue, Nuxt, Angular, Svelte, SvelteKit, Astro, Solid, Solid Start, Qwik, Remix, TanStack Start | | Backend | NestJS, Express, Fastify, Hono, Koa, AdonisJS, FeathersJS, Elysia, Hapi, Nitro | | ORM / Database | Prisma, Drizzle, TypeORM, Mongoose, Sequelize, Knex, Mikro-ORM | | State Management | TanStack Query, TanStack Store, Zustand, Redux Toolkit | | Validation | Zod, Valibot, class-validator | | Testing | Vitest, Jest, Playwright, Cypress | | Styling | Tailwind CSS, Styled Components, Emotion | | Forms | React Hook Form, TanStack Form | | UI Components | shadcn/ui, Storybook | | Routing | TanStack Router | | Animation | Framer Motion | | Tables | TanStack Table | | DevTools | TanStack DevTools | | Auth | NestJS Passport, Passport, NextAuth, Auth.js, Better Auth | | GraphQL | Apollo Server, Apollo Client | | RPC | tRPC | | Realtime | Socket.IO, ws | | Logging | Winston, Pino | | HTTP Client | Axios, Ky | | i18n | i18next, next-intl | | Date | Day.js, date-fns, Luxon | | Email | Nodemailer, NestJS Mailer | | Upload | Multer | | Queue | BullMQ | | API Docs | NestJS Swagger | | Rate Limiting | NestJS Throttler | | Cache | NestJS Cache Manager | | Config | NestJS Config | | Monorepo | npm workspaces, yarn workspaces, pnpm workspaces, Turborepo, Nx, Lerna |

Package Managers

npm, pnpm, yarn, bun

4-Layer Knowledge System

Knowledge files are organized into layers of increasing specificity. Only the layers relevant to your stack get installed.

| Layer | Count | Scope | |-------|-------|-------| | Layer 1: Universal | 19 files | Architecture, TypeScript patterns, error handling, testing, security, API design, performance, naming, code review, docs, observability, dependencies, configuration, anti-patterns, refactoring, concurrency, git workflow, CI/CD, PM workflow | | Layer 2: Domain | 24 files | Frontend (10): component design, accessibility, i18n, etc. Backend (14): database patterns, caching, validation, event-driven, contracts, performance, anti-patterns, etc. | | Layer 3: Framework | 12 files | Frontend (7): Next.js, React SPA, Vue, Nuxt, Angular, Remix/React Router v7, TanStack Start. Backend (5): NestJS, Express, Fastify, Hono, Koa | | Layer 4: Tool | 50 files | ORM, validation, state, forms, auth, GraphQL, RPC, realtime, logging, HTTP, i18n, date, email, upload, etc. |

Total: 105 knowledge rules

Profiles

| Profile | Layers | Description | |---------|--------|-------------| | fullstack-nextjs | L1 + L2 frontend + L2 backend + L3 Next.js + L4 tools | Next.js fullstack applications | | fullstack-react-nestjs | L1 + L2 both + L3 React SPA + L3 NestJS + L4 tools | React + NestJS monorepo | | frontend-react | L1 + L2 frontend + L3 React SPA + L4 tools | React SPA without backend | | backend-nestjs | L1 + L2 backend + L3 NestJS + L4 tools | NestJS API without frontend | | generic | L1 only | Any TypeScript project |

Example: A Next.js + Prisma + Zod + TanStack Query + Vitest project gets: 19 universal + 10 frontend + 1 Next.js + 4 tool-specific = 34 rules.

PM Workflow

stackwise includes a two-tier development workflow powered by 5 slash commands. Small work becomes a standalone task (single issue, single PR). Larger work becomes an epic — explored, planned, reviewed, then executed.

Idea → /sw-plan → Assess scope
                        ↓
          ┌─────────────┴──────────────────┐
    Standalone Task                   Epic
     Review → Issue                    ↓
          ↓                     Explore & brainstorm
    /sw-work → /sw-ship         (research, propose, iterate)
                                       ↓
                                PRD + task breakdown
                                       ↓
                                Review & approve
                                (user confirms plan)
                                       ↓
                                GitHub Issues created
                                       ↓
                                /sw-tasks (dashboard)
                                       ↓
                                /sw-work  (implement)
                                       ↓
                                /sw-ship  (PR + merge)
                                       ↓
                                /sw-standup (report)

Commands

| Command | Description | |---------|-------------| | /sw-plan <idea> | Plan work as standalone task or epic | | /sw-plan <idea> --auto | Plan epic and automatically execute all tasks until complete | | /sw-tasks | View task status dashboard from GitHub Issues | | /sw-work <issue> | Pick up a task and implement it | | /sw-ship | Create PR for a tracked task issue, close it, and optionally merge | | /sw-standup | Generate standup status report across all epics |

Integration Branch Support

When enabled during setup, task PRs target an integration branch (feat/<epic-slug>) instead of main. After all tasks are complete, a final PR merges the integration branch into main.

Project Management Files

.claude/pm/
├── PROJECT.md              # Project index with epic roadmap
├── prds/
│   └── <epic-slug>.md      # PRD per epic
└── epics/
    └── <epic-slug>/
        ├── epic.md          # Epic details and task table
        └── tasks.md         # Task ↔ GitHub Issue mapping

Skills (8 Slash Commands)

Installed to .claude/skills/ and available as /command in Claude Code.

PM Workflow

| Command | Description | |---------|-------------| | /sw-plan <idea> | Plan work as standalone task or epic. Use --auto for epic auto-execution | | /sw-tasks | View task status dashboard from GitHub Issues | | /sw-work <issue> | Pick up a task and implement it | | /sw-ship | Create PR for a tracked task issue, close it, and optionally merge | | /sw-standup | Generate standup status report across all epics | | /sw-review | Code review using project rules as checklist |

Utility

| Command | Description | |---------|-------------| | /sw-generate-knowledge | Generate knowledge rules for an unsupported tool | | /sw-sync-project | Sync CLAUDE.md with current project state |

Agents (5 Specialized)

| Agent | Model | Role | Purpose | |-------|-------|------|---------| | backend-dev | Opus | Developer | Backend specialist: API, database, business logic | | frontend-dev | Opus | Developer | Frontend specialist: UI, components, state management | | qa | Opus | QA Engineer | Test coverage, edge cases, acceptance criteria validation | | code-reviewer | Opus | Reviewer | Code quality, patterns, conventions | | security-auditor | Opus | Security | Security analysis using OWASP categories |

Workflow Modes

| Mode | Description | |------|-------------| | Solo | Single agent per task — implements, tests, and ships (default) | | Agency | Specialized agents with quality pipeline: dev → qa → review → ship |

Agency mode runs a quality pipeline for each task in auto mode (/sw-plan --auto):

Dev Agent (backend/frontend) → implement + test + commit
  → QA Agent → validate quality + test coverage
    → Code Review Agent → check code quality
      → Security Agent (if API/auth task)
        → Ship (push + PR + merge)

Hooks

| Hook | Trigger | Purpose | |------|---------|---------| | format-on-save.sh | After Edit/Write | Auto-formats files after Claude edits them | | protect-env.sh | Before Bash/Read | Blocks Claude from reading .env files |

How It Works

Running npx stackwise walks you through an interactive flow:

1. Checks for existing .claude/profile.json (offers to update or start fresh)
2. Scans package.json files to detect your stack (including monorepo workspaces)
3. Shows detection results and asks for confirmation/overrides
4. Configures development workflow (commit convention, branch strategy, integration branches, release strategy, PR merge strategy)
5. Asks for glob path patterns (frontend/backend/shared)
6. Resolves knowledge files from all 4 layers based on your stack
7. Lets you select which skills, agents, and hooks to install
8. Composes and installs everything, merges settings, generates CLAUDE.md

Re-run anytime to update your configuration:

npx stackwise

What Gets Created

your-project/
├── .claude/
│   ├── profile.json           # Your stack profile (re-used on next run)
│   ├── settings.json          # Permissions + hooks config
│   ├── rules/
│   │   ├── 01-universal-*.md  # Layer 1 rules
│   │   ├── 02-frontend-*.md   # Layer 2 frontend rules
│   │   ├── 02-backend-*.md    # Layer 2 backend rules
│   │   ├── 03-framework-*.md  # Layer 3 framework rules
│   │   └── 04-tool-*.md       # Layer 4 tool rules
│   ├── skills/                # Slash commands
│   ├── agents/                # Specialized agents
│   ├── hooks/                 # Automation scripts
│   └── pm/                    # PM workflow (created by /sw-plan)
│       ├── PROJECT.md
│       ├── prds/
│       └── epics/
└── CLAUDE.md                  # Auto-generated project instructions

Customization

Adding a knowledge file

1. Create a markdown file in the appropriate layer directory under knowledge/
2. Layer 1 files apply universally; Layer 2+ files are profile-specific
3. Add an entry to lib/registry.json with layer, domain, tool, and glob_strategy
4. If Layer 4, add the tool mapping to src/profiles.js TOOL_MAP

Adding a skill

1. Create skills/<name>/SKILL.md
2. Use $ARGUMENTS for user input after the command
3. Write step-by-step instructions for Claude to follow
4. Add to SKILLS array in src/constants.js

Adding an agent

1. Create agents/<name>/<name>.md
2. Use YAML frontmatter for model and tool configuration
3. Add to AGENTS array in src/constants.js

Adding tool detection

1. Add the package name to DEPENDENCY_MAP in src/constants.js
2. Map the detection value in TOOL_MAP in src/profiles.js
3. Create the knowledge rule and register in lib/registry.json

Requirements

• Node.js >= 18.0.0

Contributing

See CONTRIBUTING.md for development setup and guidelines.

License

MIT