- Published on
- 1,587 字 · 8 min read
单靠 Prompt 加限制词防不住注入，正确做法是搭建「输入侧 → 模型侧 → 执行侧 → 输出侧」四层纵深防御体系，其中执行侧权限管控是最硬的底线。
Security
All Posts
- agent (15)
- engineering (13)
- ai (12)
- architecture (7)
- opinion (7)
- open-source (4)
- methodology (3)
- llm (3)
- product (3)
- python (3)
- knowledge-management (2)
security (2)
- skill (2)
- startup (2)
- api (2)
- frontend (2)
- mindset (2)
- workflow (2)
- rag (2)
- context-engineering (1)
- reliability (1)
- system-design (1)
- dev-lessons (1)
- prompt-injection (1)
- governance (1)
- industry (1)
- anthropic (1)
- cost (1)
- retry (1)
- billing (1)
- devops (1)
- cloudflare (1)
- design-system (1)
- tailwind (1)
- css (1)
- career (1)
- github (1)
- trending (1)
- ide (1)
- devtools (1)
- developer-experience (1)
- pr (1)
- nextjs (1)
- performance (1)
- react (1)
- onboarding (1)
- fastapi (1)
- streamlit (1)
- embedding (1)
- chromadb (1)
- nodejs (1)
- research (1)
- webpack (1)
- turbopack (1)
- build-tools (1)
- Published on
- 924 字 · 5 min read
从独立开发者视角，系统梳理建站全链路数据安全实践：Cloudflare 隧道架构、开发阶段单测与静态扫描、API 六维防护、渗透测试验证、灰度发布与回滚、二阶安全兜底。