Master Senior Backend Engineer — AI agent skill with production patterns from 30+ top repos (200k+ GitHub stars). Trained from research of top 53k+ star skill repos + system prompts from 7 major AI tools.
Works with: Claude Code · Cline · Roo Code · Cursor · Windsurf · Copilot · Codex · Gemini · Kimi · Kilo Code · Kiro · Antigravity
Supports: NestJS · Next.js · Express · Fastify · Vue.js · Nuxt · Django · FastAPI · Flask · Spring Boot · Laravel · Go · Ruby on Rails · Rust
# Interactive (checkbox UI — pick your agents)
npx @buivietphi/skill-backend-mt
# Specific agent
npx @buivietphi/skill-backend-mt --claude
npx @buivietphi/skill-backend-mt --codex
npx @buivietphi/skill-backend-mt --gemini
npx @buivietphi/skill-backend-mt --all
# Project-level rules (Cursor, Windsurf, Cline, etc.)
npx @buivietphi/skill-backend-mt --init # Interactive
npx @buivietphi/skill-backend-mt --init cursor # Just .cursorrules
npx @buivietphi/skill-backend-mt --init all # All agents~/.claude/skills/
└── skill-backend-mt/
├── SKILL.md ← Entry point (auto-loaded)
├── AGENTS.md
├── nodejs/
│ ├── nestjs.md
│ ├── nextjs.md
│ ├── express.md
│ └── vuejs.md
├── python/
│ ├── fastapi.md
│ └── django.md
├── java/
│ └── spring-boot.md
├── php/
│ └── laravel.md
├── others/
│ └── go-ruby-rust.md
└── shared/ ← On-demand (28 files)
@skill-backend-mt
> "Create a user CRUD with JWT auth"
Uses production patterns from 30+ repos. Best for learning, new projects, generic advice.
@skill-backend-mt project
> "Add an orders module"
Reads YOUR project first, clones existing patterns. Best for existing codebases.
- Clean Architecture, Hexagonal, DDD, CQRS
- Module-based vs Layer-based structure
- Monolith vs Microservices decision matrix
- Reference repos: NestJS (70k+), Next.js (130k+), FastAPI (80k+), Spring Boot (76k+), Laravel (79k+), Django (82k+)
- REST conventions (resources, verbs, status codes)
- GraphQL schema design + N+1 prevention
- gRPC proto patterns
- WebSocket real-time patterns
- Pagination (cursor-based vs offset)
- JWT + refresh token rotation
- OAuth 2.0 / OpenID Connect
- RBAC / ABAC authorization
- SQL injection, XSS, CSRF prevention
- Rate limiting, CORS, secret management
- ORM patterns (Prisma, TypeORM, SQLAlchemy, Eloquent, Spring Data, Mongoose)
- Migration best practices
- N+1 query detection and fix
- Multi-level caching (Redis, in-memory, CDN)
- Connection pooling, query optimization
- CRUD service generation (validate → authorize → execute → side effects → return DTO)
- Error handling implementation (custom exceptions, global handler, retry + timeout)
- Caching implementation (cache-aside, write-through, invalidation, stampede prevention)
- Pagination & filtering (offset-based, cursor-based, sort security, filter whitelist)
- Testing fixtures (factories, builders, mocks, integration test setup)
- File upload & storage (S3, presigned URL, validation, image processing, cleanup)
- Background jobs (BullMQ, Celery, retries, dead letter queue, CRON)
- Webhook patterns (signature verification, deduplication, idempotent handlers)
- Concurrency patterns (optimistic/pessimistic locking, distributed lock, idempotency keys)
- Structured logging (Pino/structlog, correlation IDs, PII redaction)
- Testing pyramid (70% unit / 20% integration / 10% E2E)
- Per-framework testing (Jest, pytest, JUnit, PHPUnit)
- Docker multi-stage builds
- GitHub Actions CI templates
- Deployment strategies (blue-green, canary, rolling)
- Cardinal Rules (10 inviolable rules for code quality)
- Self-Critique Loop (re-read and verify before presenting)
- Context Staleness Rule (re-read files after 5 messages)
- Verification-First Pattern (read before suggesting)
- Assumption-Driven Progress (don't block on non-critical choices)
- Negative Space Pattern (check what you didn't write)
- Batched Operations (group related file creation)
- Error Recovery with Escalation (4-level escalation)
- Leverage Pyramid (40% research, 30% planning, 20% implement, 10% verify)
- Session State Tracking (never lose track of multi-file tasks)
- Grounding Protocol (anti-hallucination with source hierarchy)
- AI-DLC Workflow (4 Hats for complex features)
- Multi-Part Execution Protocol (discover all locations → work plan → checkpoint)
- Discovery-Execute Workflow (for "fix all X" / broad requests)
- Intent Analysis Engine (AI self-parses vague/circular input without asking user)
- Loop Detection & Self-Correction (detect repeated failures → change strategy automatically)
- Auto-think templates (Fix/Debug, Build/Create, Review, Refactor, Multi-Part)
- Source verification checklist
- Context-needed declarations
- Patterns learned from: Anthropic, Cursor, Lovable, Manus, Windsurf, Kiro, Replit system prompts
| Scenario | Tokens | % of 128K |
|---|---|---|
| SKILL.md only | ~14,140 | 11.0% |
| Core auto-load | ~29,740 | 23.2% |
| Smart load (core + 1 framework) | ~31,740 | 24.8% |
| Multi-framework (3 frameworks) | ~35,740 | 27.9% |
| Full load (all files) | ~106,340 | 83.1% |
Smart loading uses only 24.8% of context — 75% free for actual code work.
| File | Bytes | Tokens |
|---|---|---|
| SKILL.md | 56,573 | ~14,140 |
| shared/code-review.md | 26,528 | ~6,630 |
| shared/bug-detection.md | 22,209 | ~5,550 |
| shared/architecture-intelligence.md | 18,494 | ~4,620 |
| shared/intent-analysis.md | 17,270 | ~4,320 |
| shared/crud-patterns.md | 15,445 | ~3,860 |
| shared/prompt-engineering.md | 13,689 | ~3,420 |
| shared/error-handling-impl.md | 11,554 | ~2,890 |
| shared/api-design.md | 11,386 | ~2,850 |
| shared/error-recovery.md | 11,096 | ~2,770 |
| shared/testing-fixtures.md | 10,888 | ~2,720 |
| shared/database-patterns.md | 10,789 | ~2,700 |
| shared/ai-dlc-workflow.md | 9,371 | ~2,340 |
| shared/pagination-patterns.md | 8,975 | ~2,240 |
| shared/concurrency-patterns.md | 8,802 | ~2,200 |
| shared/webhook-patterns.md | 8,496 | ~2,120 |
| shared/logging-impl.md | 8,320 | ~2,080 |
| shared/caching-implementation.md | 8,078 | ~2,020 |
| shared/background-jobs.md | 7,732 | ~1,930 |
| shared/file-handling.md | 7,646 | ~1,910 |
| php/laravel.md | 9,031 | ~2,260 |
| java/spring-boot.md | 8,977 | ~2,240 |
| python/fastapi.md | 8,616 | ~2,150 |
| python/django.md | 8,143 | ~2,040 |
| nodejs/nestjs.md | 7,922 | ~1,980 |
| nodejs/nextjs.md | 7,720 | ~1,930 |
| shared/auth-security.md | 7,469 | ~1,870 |
| shared/performance-optimization.md | 7,267 | ~1,820 |
| others/go-ruby-rust.md | 7,195 | ~1,800 |
| shared/microservices.md | 7,179 | ~1,800 |
| nodejs/vuejs.md | 7,139 | ~1,790 |
| shared/ci-cd.md | 7,051 | ~1,760 |
| nodejs/express.md | 6,257 | ~1,560 |
| shared/observability.md | 6,100 | ~1,530 |
| shared/testing-strategy.md | 5,629 | ~1,410 |
| shared/version-management.md | 4,923 | ~1,230 |
| shared/common-pitfalls.md | 4,443 | ~1,110 |
| humanizer/humanizer-backend.md | 3,923 | ~980 |
| shared/agent-rules-template.md | 2,916 | ~730 |
| shared/document-analysis.md | 2,236 | ~560 |
| shared/claude-md-template.md | 1,878 | ~470 |
| Agent | Type | Install |
|---|---|---|
| Claude Code | Skills directory | --claude |
| Codex | Skills directory | --codex |
| Gemini CLI | Skills directory | --gemini |
| Kimi | Skills directory | --kimi |
| Antigravity | Skills directory | --antigravity |
| Cursor | Project-level | --init cursor |
| Windsurf | Project-level | --init windsurf |
| Cline | Project-level | --init cline |
| Roo Code | Project-level | --init roocode |
| Copilot | Project-level | --init copilot |
| Kilo Code | Project-level | --init kilocode |
| Kiro | Project-level | --init kiro |
humanizer-backend is auto-installed alongside the main skill. It humanizes AI-generated code comments, API docs, and error messages.
@humanizer-backend
> "Humanize this API documentation"
MIT — by buivietphi