16gagent/docs/reports/capability-boundary.md

Capability Boundary — v1.1.0-semantic

Frozen: 2026-06-05T19:40:00+08:00 This document defines what the system CAN and CANNOT do. Do not claim capabilities outside this boundary.

✅ CAN DO (Certified)

Web Fullstack Generation

• Input: Chinese natural language requirement (1 sentence)
• Output: Complete Next.js + Express monorepo
• Quality: Domain-specific tables, English identifiers, CRUD APIs, auth scaffolding
• Verified: 18 domains, 170 regression checks

Electron Desktop Packaging

• Input: Fullstack monorepo
• Output: Electron 35 app with tray, IPC, auto-updater skeleton
• Platforms: Windows, macOS, Linux

Release Packaging

• Input: Fullstack monorepo
• Output: Version manifest, SHA256 checksums, platform installers (placeholders)

Semantic Generation

• Chinese → English: API paths, table names, route names, service names
• Field specificity: 60+ domain-specific field templates
• Identifier safety: Singularize exceptions, longest-match slug, exclude rules

⚠️ CANNOT GUARANTEE (Not Certified)

Native Mobile

• iOS (SwiftUI) — Tech stack exists in template but no builder agent
• Android (Jetpack Compose) — Tech stack exists in template but no builder agent
• React Native / Flutter — Mentioned in tech stack, not generated

Mini Programs

• 微信小程序 — Platform detection works, no builder agent
• uni-app — Tech stack template only

Backend Alternatives

• NestJS — Architecture suggests NestJS, actual generation uses Express
• Prisma ORM — Architecture suggests Prisma, actual uses better-sqlite3
• PostgreSQL — Architecture suggests PostgreSQL, actual uses SQLite

Distributed Systems

• Microservices — Single-service architecture only
• Message queues — No Kafka/RabbitMQ integration
• Service mesh — No Istio/Linkerd support

AI Integration

• AI Agent clusters — No multi-agent orchestration
• LLM-powered features — No AI feature generation
• Smart recommendations — Placeholder only

Production Readiness

• Load testing — No performance benchmarks
• Security audit — No penetration testing
• CI/CD pipelines — GitHub Actions skeleton only
• Monitoring/observability — No APM integration

Boundary Rules

1. Do not add new agents — SF-01 through SF-07 only
2. Do not add new platforms — Web + Electron + Release only
3. Do not add new builders — Current 7 agents only
4. All changes must pass regression gate — node scripts/regression-gate.mjs
5. Capability expansion requires explicit unfreeze — Document in this file