backend-scaffold（后端脚手架）

backend-scaffold

Generate production-ready backend code scaffolds from upstream specs — multi-stack, tech.md-adaptive.

Context Files

• $JAAN_CONTEXT_DIR/tech.md - Tech stack context (CRITICAL — determines framework, DB, patterns)

- Uses sections: #current-stack, #frameworks, #constraints, #patterns

• $JAAN_CONTEXT_DIR/config.md - Project configuration
• $JAAN_TEMPLATES_DIR/jaan-to-backend-scaffold.template.md - Output template
• $JAAN_LEARN_DIR/jaan-to-backend-scaffold.learn.md - Past lessons (loaded in Pre-Execution)
• ${CLAUDE_PLUGIN_ROOT}/docs/extending/language-protocol.md - Language resolution protocol

Input

Upstream Artifacts: $ARGUMENTS

Accepts 1-3 file paths or descriptions:

• backend-api-contract — Path to OpenAPI YAML (from /jaan-to:backend-api-contract output: api.yaml)
• backend-task-breakdown — Path to BE task breakdown markdown (from /jaan-to:backend-task-breakdown output)
• backend-data-model — Path to data model markdown (from /jaan-to:backend-data-model output)
• Empty — Interactive wizard prompting for each

Pre-Execution Protocol

MANDATORY — Read and execute ALL steps in: ${CLAUDE_PLUGIN_ROOT}/docs/extending/pre-execution-protocol.md Skill name: backend-scaffold Execute: Step 0 (Init Guard) → A (Load Lessons) → B (Resolve Template) → C (Offer Template Seeding)

Also read context files if available:

• $JAAN_CONTEXT_DIR/tech.md — Know the tech stack for framework-specific code generation
• $JAAN_CONTEXT_DIR/config.md — Project configuration

Language Settings

Read and apply language protocol: ${CLAUDE_PLUGIN_ROOT}/docs/extending/language-protocol.md Override field for this skill: language_backend-scaffold

Language exception: Generated code output (variable names, code blocks, schemas, SQL, API specs) is NOT affected by this setting and remains in the project's programming language.

PHASE 1: Analysis (Read-Only)

Thinking Mode

ultrathink

Use extended reasoning for:

• Analyzing upstream artifacts to derive code structure
• Mapping API contract schemas to framework-native patterns
• Planning multi-stack generation strategy
• Identifying edge cases in input parsing

Step 1: Validate & Parse Inputs

For each provided path:

• backend-api-contract: Read api.yaml, extract paths, schemas, error responses, security schemes
• backend-task-breakdown: Read markdown, extract task list, entity names, reliability patterns
• backend-data-model: Read markdown, extract table definitions, constraints, indexes, relations
• Report which inputs found vs missing; suggest fallback for missing (e.g., CRUD from backend-data-model if no API contract)

Present input summary:

INPUT SUMMARY
─────────────
Sources Found:    {list}
Sources Missing:  {list with fallback suggestions}
Entities:         {extracted entity names}
Endpoints:        {count from API contract}
Tables:           {count from data model}
Tasks:            {count from task breakdown}

Step 2: Detect Tech Stack

Read $JAAN_CONTEXT_DIR/tech.md:

• Extract framework from #current-stack (default: Fastify v5+)
• Extract DB from #current-stack (default: PostgreSQL)
• Extract patterns from #patterns (auth, error handling, logging)
• If tech.md missing: ask framework/DB via AskUserQuestion

Step 3: Clarify Architecture

AskUserQuestion for items not in tech.md:

• Project structure (monolith / modular monolith / microservice)
• Auth middleware pattern (JWT / API key / session / none)
• Error handling depth (basic / full RFC 9457 with error taxonomy)
• Logging (structured JSON pino / winston / none)

Step 4: Plan Scaffold Structure

Present directory tree, file list, resource count:

SCAFFOLD PLAN
═════════════

STACK: {framework} + {database} + {orm}

PROJECT STRUCTURE
─────────────────
{directory tree showing all files to generate}

FILES ({count} total)
─────────────────────
{numbered list with file purpose}

RESOURCES ({count})
───────────────────
{resource list with operations}

HARD STOP — Review Scaffold Plan

Use AskUserQuestion:

• Question: "Proceed with generating the scaffold?"
• Header: "Generate"
• Options:

- "Yes" — Generate the scaffold code - "No" — Cancel - "Edit" — Let me revise the scope or architecture first

Do NOT proceed to Phase 2 without explicit approval.

PHASE 2: Generation (Write Phase)

Phase 2 Output — Flat folder (no nested subfolder)

All files in $JAAN_OUTPUTS_DIR/backend/scaffold/{id}-{slug}/:

{id}-{slug}/
├── {id}-{slug}.md                    # Main doc (setup guide + architecture)
├── {id}-{slug}-routes.ts              # Route handlers (all resources)
├── {id}-{slug}-services.ts            # Service layer (business logic)
├── {id}-{slug}-schemas.ts             # Validation schemas
├── {id}-{slug}-middleware.ts           # Auth + error handling middleware
├── {id}-{slug}-prisma.prisma          # ORM data model
├── {id}-{slug}-config.ts              # Package.json + tsconfig content
└── {id}-{slug}-readme.md              # Setup + run instructions

File extensions adapt to detected stack (.ts for Node.js,.php for PHP,.go for Go).

Step 6: Generate Content

Read $JAAN_TEMPLATES_DIR/jaan-to-backend-scaffold.template.md and populate all sections based on Phase 1 analysis.

If tech stack needed, extract sections from tech.md:

• Current Stack: #current-stack
• Frameworks: #frameworks
• Constraints: #constraints
• Patterns: #patterns

Step 7: Quality Check

Validate generated output against checklist:

• All API endpoints from contract have route handlers
• All entities from data model have ORM models
• Validation schemas generated for all request bodies
• Error handler covers validation, ORM, and generic errors
• Service layer stubs exist for all business logic
• DB singleton + graceful disconnect configured
• No anti-patterns present in generated code

If any check fails, fix before preview.

Step 8: Preview & Approval

Present generated output summary. Use AskUserQuestion:

• Question: "Write scaffold files to output?"
• Header: "Write Files"
• Options:

- "Yes" — Write the files - "No" — Cancel - "Refine" — Make adjustments first

Step 9: Generate ID and Folder Structure

source "${CLAUDE_PLUGIN_ROOT}/scripts/lib/id-generator.sh"
SUBDOMAIN_DIR="$JAAN_OUTPUTS_DIR/backend/scaffold"
mkdir -p "$SUBDOMAIN_DIR"
NEXT_ID=$(generate_next_id "$SUBDOMAIN_DIR")
slug="{project-name-slug}"
OUTPUT_FOLDER="${SUBDOMAIN_DIR}/${NEXT_ID}-${slug}"

Preview output configuration:

Output Configuration - ID: {NEXT_ID} - Folder: $JAAN_OUTPUTS_DIR/backend/scaffold/{NEXT_ID}-{slug}/ - Main file: {NEXT_ID}-{slug}.md

Step 10: Write Output

1. Create output folder: mkdir -p "$OUTPUT_FOLDER"
2. Write all scaffold files to $OUTPUT_FOLDER
3. Update subdomain index:

source "${CLAUDE_PLUGIN_ROOT}/scripts/lib/index-updater.sh"
add_to_index \
  "$SUBDOMAIN_DIR/README.md" \
  "$NEXT_ID" \
  "${NEXT_ID}-${slug}" \
  "{Project Title}" \
  "{Executive summary — 1-2 sentences}"

1. Confirm completion:

Scaffold written to: $JAAN_OUTPUTS_DIR/backend/scaffold/{NEXT_ID}-{slug}/ Index updated: $JAAN_OUTPUTS_DIR/backend/scaffold/README.md

Step 11: Suggest Next Actions

Scaffold generated successfully! Next Steps: - Copy scaffold files to your project directory - Run npm install (or equivalent) to install dependencies - Run /jaan-to:dev-integration-plan to plan integration with existing code - Run /jaan-to:dev-test-plan to generate test plan

Step 12: Capture Feedback

Use AskUserQuestion:

• Question: "How did the scaffold turn out?"
• Header: "Feedback"
• Options:

- "Perfect!" — Done - "Needs fixes" — What should I improve? - "Learn from this" — Capture a lesson for future runs

If "Learn from this": Run /jaan-to:learn-add backend-scaffold "{feedback}"

Key Generation Rules — Node.js/TypeScript (Research-Informed)

• Routing: Use @fastify/autoload v6 for file-based route loading — register twice (plugins with encapsulate: false, routes encapsulated per resource); add ignorePattern: /.*\.(?:schema|service)\.ts/ to prevent non-plugin files from being auto-loaded as routes
• Type Provider: Use fastify-type-provider-zod v6.1+ with validatorCompiler/serializerCompiler set once at app level; must call withTypeProvider<ZodTypeProvider>() on each encapsulated context (type providers don't propagate across encapsulation boundaries)
• Prisma Singleton: Use globalThis pattern to prevent connection pool exhaustion during hot-reload; conditional assignment based on NODE_ENV
• Zod Schemas: Define schemas in .schema.ts files, export z.infer<> types; derive from OpenAPI contract component schemas
• Error Handler: Use Fastify's setErrorHandler (NOT Express-style middleware) — use hasZodFastifySchemaValidationErrors(error) for 400 (NOT instanceof ZodError which fails across module boundaries), use isResponseSerializationError(error) for 500 serialization errors; map PrismaClientKnownRequestError P2002 → 409 (unique constraint), P2003 → 409 (foreign key), P2025 → 404 (not found), all others → 500; always set Content-Type: application/problem+json
• RFC 9457 Fields: type (URI), title, status, detail, instance; extension errors[] for validation details
• Service Layer: Plain exported functions importing the Prisma singleton — module caching acts as built-in singleton, making DI containers (tsyringe, inversify) unnecessary; testable via vi.mock(); callable from CRON jobs or queue consumers outside HTTP context; use Prisma $transaction for cross-service operations
• Route Structure: Collocated index.ts (routes) + {resource}.schema.ts (Zod) + {resource}.service.ts (logic) per resource
• TypeScript: Extend fastify-tsconfig v2 with target: "ES2023", module: "NodeNext", strict: true
• Import Extensions: With "type": "module" and moduleResolution: "NodeNext", all imports MUST include .js extensions — NodeNext mirrors Node.js runtime behavior; never use moduleResolution: "bundler" for backends (allows vague imports that fail at runtime)
• Env Vars: Parameterize DATABASE_URL, PORT, HOST, NODE_ENV, LOG_LEVEL, CORS_ORIGIN
• Env Validation: Validate environment variables with Zod at startup — crash immediately on missing/invalid variables; use Node.js 20.6+ --env-file=.env flag for loading
• Scripts: dev (tsx watch), build (tsc), start, lint, test, db:generate, db:migrate:dev, db:migrate:deploy, db:push, db:seed, db:studio, postinstall (prisma generate)

Multi-Stack Support (Research-Informed)

The skill reads tech.md #current-stack to determine which stack to generate:

PHP Stack (Laravel) — Key Patterns:

• PSR-4 autoloading, single public/index.php entry point
• Route model binding + Form Requests for validation ($request->validated(), never $request->all())
• Eloquent Active Record with utf8mb4, BIGINT PKs, JSON columns
• Strictness in AppServiceProvider::boot(): preventLazyLoading() (catches N+1), preventSilentlyDiscardingAttributes() (catches mass assignment typos), preventAccessingMissingAttributes(); in production, lazy loading violations log instead of throwing
• API Resources for response shaping (never expose raw models); use whenLoaded(), whenCounted(), conditional when() helpers
• Sanctum for auth (SPA cookies + API tokens); cookie auth requires SANCTUM_STATEFUL_DOMAINS and supports_credentials: true
• Pest 3/4 for testing with architecture presets (arch()->preset()->laravel()) and mutation testing (--mutate)
• RFC 9457 via crell/api-problem v3.8.0 (PHP ^8.3)
• Zero-downtime MySQL migrations: expand-contract pattern (add nullable → backfill → deploy → drop old); use daursu/laravel-zero-downtime-migration for large tables

PHP Stack (Symfony) — Key Patterns:

• API Platform v4.x: #[ApiResource] annotations for automatic CRUD REST APIs with OpenAPI documentation
• Doctrine Data Mapper ORM: entities are POPOs, persistence via EntityManager (better separation than Active Record)
• DTOs with #[MapRequestPayload] and Symfony Validator constraint attributes (#[Assert\NotBlank], #[Assert\Positive])
• JWT via lexik/jwt-authentication-bundle v3.2.0 with RS256 signing + gesdinet/jwt-refresh-token-bundle for refresh tokens

Go Stack — Generation Rules:

• Routing: Go 1.22+ net/http.ServeMux with method+wildcard patterns (GET /users/{id}, r.PathValue("id")); use Chi v5.2.x only for middleware grouping/subrouters; avoid gorilla/mux (archived 2023), Gin/Fiber (diverge from net/http idioms)
• Structure: Feature-based internal/ packages (internal/user/handler.go, service.go, repository.go); avoid layer-based internal/handlers/ anti-pattern (excessive cross-package imports); shallow hierarchies (1-2 levels)
• DI: Constructor injection with small interfaces (1-3 methods) defined at consumer site; accept interfaces, return structs; wire manually in main.go; manual DI preferred over Wire/Dig except for very large projects
• Database: sqlc generates type-safe Go code from annotated SQL queries (-- name: GetUser:one); golang-migrate for sequential numbered up/down migration files
• Validation: go-playground/validator v10 (v10.27.0) with struct tags (validate:"required,email"); single instance (caches struct info); WithRequiredStructEnabled() for v11 compatibility; RegisterTagNameFunc() for JSON field names
• OpenAPI: oapi-codegen v2 generates Go types, server interfaces, and request validation middleware; developers implement ServerInterface; YAML config with Chi/stdlib backend support
• Error Handling: RFC 9457 via custom ProblemDetail struct; Content-Type: application/problem+json
• Testing: Table-driven tests with httptest.NewRecorder() + httptest.NewRequest(); t.Run() subtests; t.Parallel() for concurrent execution
• Docker: Multi-stage builds → 10-20MB images using distroless/static-debian12; CGO_ENABLED=0 for static binaries; -ldflags="-s -w" to strip debug info
• Graceful Shutdown: signal.NotifyContext with 10-second timeout, closing HTTP server and database connections

WebSocket Support (Optional — all stacks):

• Go: coder/websocket, Hub pattern for connection management
• Node.js: ws / Socket.IO
• PHP: Ratchet / Swoole
• Auth: ephemeral single-use token via query parameter (ws://host/ws?ticket=abc123); 30-second TTL, consumed on first use to prevent log-exposure attacks
• SSE handles 95% of real-time use cases — suggest SSE first; SSE works over standard HTTP, supports auto-reconnection, multiplexed over HTTP/2

Test Framework & Mutation Tool Recommendations

When generating scaffold, include test framework and mutation tool recommendations based on detected stack:

Add to generated README: "Run /jaan-to:qa-test-mutate to validate test suite effectiveness."

Anti-Patterns to NEVER Generate

All Stacks: Business logic in route handlers, hardcoded secrets, missing .gitignore, no error handling

Node.js: Direct Prisma calls in handlers, multiple PrismaClient instances, any types, Express-style error middleware, missing response serialization schemas, instanceof ZodError (use v6 helpers), missing .js extensions in ESM imports, moduleResolution: "bundler" for backends

PHP: Fat controllers, N+1 queries, exposing raw Eloquent models, env() outside config files, utf8 instead of utf8mb4, missing Eloquent strictness modes

Go: Generic package names (utils/), global database connections, ignoring errors, unlimited connection pool, goroutine leaks, layer-based internal/handlers/ structure

Package Dependencies (Research-Validated)

Node.js/TypeScript:

• Production: fastify ^5.7, @fastify/autoload ^6, @fastify/cors ^10, @fastify/sensible ^6, @fastify/swagger ^9, @fastify/swagger-ui ^5, @prisma/client ^6, fastify-plugin ^5, fastify-type-provider-zod ^6.1, zod ^3.24
• Dev: typescript ^5.6, @types/node ^22, fastify-tsconfig ^2, prisma ^6, tsx ^4, vitest ^2, eslint ^9

Go: chi v5.2.x (optional), go-playground/validator v10, golang-migrate, sqlc, oapi-codegen v2

PHP (Laravel): laravel/sanctum, crell/api-problem ^3.8, pestphp/pest ^3

PHP (Symfony): api-platform/core ^4, lexik/jwt-authentication-bundle ^3.2, gesdinet/jwt-refresh-token-bundle

Skill Alignment

• Two-phase workflow with HARD STOP for human approval
• Multi-stack support via tech.md detection
• Template-driven output structure
• Output to standardized $JAAN_OUTPUTS_DIR path

Definition of Done

• All API endpoints from contract have route handlers
• All entities from data model have ORM models (Prisma/Eloquent/Doctrine/sqlc)
• Validation schemas generated for all request bodies
• Error handler covers validation errors, ORM errors, and generic errors
• Service layer stubs exist for all business logic
• DB singleton + graceful disconnect configured
• Setup README is complete and actionable
• Output follows v3.0.0 structure (ID, folder, index)
• Index updated with executive summary
• User approved final result