jorinyang Cross-Project Adaptation
触发条件
通用领域触发矩阵
跨项目架构概念迁移覆盖7大领域，21个子场景。
| 领域 | 场景 | 触发信号 | 示例 |
|---|---|---|---|
| AI/ML | 模型服务架构 | 用户想从项目A的模型部署方案迁移到B | "把vLLM的continuous batching搬到我们自己的server" |
| AI/ML | Agent编排 | 用户想从开源Agent框架迁移编排模式 | "把LangGraph的checkpoint机制搬到我们的Agent系统" |
| AI/ML | RAG Pipeline | 用户想从论文实现迁移检索架构 | "把ColBERT的late interaction改到我们已有的pipeline" |
| Web/后端 | API设计模式 | 用户想从项目A的API设计迁移到B | "Stripe的API版本策略搬到我们的项目" |
| Web/后端 | 数据库设计 | 用户想从Postgres方案适配到MongoDB | "把SQL schema的partition策略迁移到Mongo sharding" |
| Web/后端 | 缓存架构 | 用户想从Redis方案迁移到替代方案 | "把Redis pub/sub的event模式适配到Kafka" |
| 前端 | 组件库迁移 | 用户想从React组件库迁移到Vue | "把Radix UI的accessibility模式搬到Vue组件" |
| 前端 | 状态管理 | 用户想从Redux迁移到Zustand | "把Redux中间件的副作用管理搬到Zustand" |
| 前端 | 路由架构 | 用户想从Next.js迁移到其他框架 | "把Next.js的SSR策略适配到Nuxt" |
| 数据工程 | 数据处理管道 | 用户想从Spark方案适配到Python | "把Spark的window function逻辑用Pandas重写" |
| 数据工程 | 流处理 | 用户想从Flink迁移到RisingWave | "把Flink的时间窗口语义适配到RisingWave" |
| 数据工程 | 数据湖/仓库 | 用户想从Delta Lake迁移到Iceberg | "把Delta Lake的ACID事务模式适配到Iceberg" |
| 基础设施 | CI/CD | 用户想从GitHub Actions迁移到GitLab | "把GHA的matrix build搬到GitLab CI" |
| 基础设施 | 容器编排 | 用户想从Docker Compose迁移到K8s | "把compose的网络拓扑适配到K8s Service" |
| 基础设施 | 监控告警 | 用户想从Datadog迁移到Prometheus | "把Datadog的复合告警逻辑适配到PromQL" |
| 安全 | 认证授权 | 用户想从OAuth迁移到SAML | "把OAuth2的token刷新机制适配到SAML" |
| 安全 | 加密方案 | 用户想从AES迁移到ChaCha20 | "把AES-GCM的认证加密模式适配为ChaCha20-Poly1305" |
| 安全 | 审计日志 | 用户想从ELK迁移到Loki | "把ELK的结构化审计日志适配到Loki label" |
| 移动/IoT | 推送通知 | 用户想从FCM迁移到APNs | "把FCM的topic订阅模式适配为APNs的channel" |
| 移动/IoT | 离线存储 | 用户想从SQLite迁移到Realm | "把SQLite的migration策略适配到Realm schema" |
| 移动/IoT | 蓝牙/WiFi | 用户想从BLE迁移到WiFi Direct | "把BLE的GATT profile映射到WiFi Direct服务发现" |
手动触发
- "把这个项目里的X搬到我们的项目"
- "adapt X from Y to Z"
- "把A的架构模式迁到B"
- "cross-project adaptation"
- "从XX项目学它的YY实现"
When to Use
Borrowing capabilities from project A (source) to enhance project B (target), especially when:
- Different languages (TypeScript → Python)
- Different architectures (plugin system → monolith)
- Different protocols (MCP → REST)
Methodology
Phase 1: Deep Source Analysis
Don't just read READMEs. Examine:
- Core interfaces/types — What contracts does the module define?
- Algorithms — What's the actual computation? (formulas, data structures)
- Integration points — How does it connect to the rest of the system?
- Dependencies — What does it need from the host environment?
Phase 2: Target Architecture Mapping
Map source concepts to target equivalents:
| Source Concept | Target Equivalent | Adaptation Needed |
|---|---|---|
| TypeScript interface | Python Protocol/dataclass | Type system translation |
| npm package | pip package or stdlib | Dependency evaluation |
| MCP tool | REST endpoint or MCP tool | Protocol adaptation |
| EventEmitter | threading + callback | Concurrency model |
| async/await | threading or asyncio | Execution model |
Phase 3: Selective Import
Don't transplant entire frameworks. Extract:
- Algorithms (formulas, scoring, search) — usually portable
- Data models (types, enums, interfaces) — translate to target idioms
- Patterns (lifecycle, state machines) — reimplement natively
- Integrations — rewrite for target's plugin/extension system
Phase 4: Integration Verification
Critical: After implementing modules, verify they're wired into the actual system.
See references/integration-gap-audit.md in clawshell-development skill for methodology.
Example: Ruflo → ClawShell
Ruflo (TypeScript) concepts adapted to ClawShell (Python):
| Ruflo Module | ClawShell Adaptation | Key Decision |
|---|---|---|
| HNSW index (custom TS) | hnswlib library |
Use battle-tested C++ binding |
| Trust evaluator formula | Direct port | Formula is language-agnostic |
| Swarm topology types | Python Enum + manager | Native dataclasses |
| Hook system (EventEmitter) | threading + priority chain | Sync, not async |
| Plugin lifecycle (IPlugin) | Protocol + state machine | Python typing.Protocol |
| SONA learning (MicroLoRA) | Not ported | Too complex, defer |
Pitfalls
- Don't copy code verbatim — different language idioms, error handling, concurrency models
- Don't port everything — YAGNI applies; port what you need now
- Don't assume API compatibility — source's API may not fit target's patterns
- Always verify integration — unit tests passing ≠ feature working; check call chains
- Check for existing equivalents — target may already have something similar under a different name
Reference Files
references/ruflo-adaptation-results.md— Actual results of Ruflo (TypeScript) → ClawShell (Python) adaptation: 6 modules ported, 398 tests, integration gap findings