Backing up Agent session logs made recurring instruction failures visible. Revisions came first; stable task practices later became skills.
从 Agent 会话日志到可复用的 skills
发表于
分类于
工程实践
最初备份 Agent 会话日志，是因为本地记录有保留期限，也为了分析指令在实际任务中反复失效的地方。通过回看纠偏过程，逐步修订指令；当一类任务的做法稳定下来，再提炼成 skill，并结合后续日志持续调整。
IAM User Model: Resource Ownership Trade-offs Across Multiple Identities
发表于
分类于
Security
Trade-offs in separating acting identities from resource owners, aligning consistency boundaries, and using the login user as the ownership subject.
IAM 用户模型：多身份下资源归属的取舍
发表于
分类于
Security
登录用户与业务用户拆分之后，讨论操作身份、资源归属与一致性边界如何分别确定，以及复用登录主体的前提和代价。
IAM User Model: Trade-offs in Separating Login Users and Business Users
发表于
分类于
Security
Design trade-offs in separating authentication identity from business identity, their relationship to multitenancy, and continuity for existing integrations.
IAM 用户模型：登录用户与业务用户的拆分取舍
发表于
分类于
Security
从认证主体与业务主体的职责出发，讨论为什么拆分登录用户与业务用户，以及这种拆分与多租户模型的关系和取舍。
Pulling Permissions Out of the Menu: Design Trade-offs in an Authorization Kernel
发表于
分类于
Architecture
The design decisions behind extracting authorization out of the menu model and converging it into a standalone component, and why each alternative was rejected.
把权限从菜单里拆出来：授权内核的几个设计取舍
发表于
分类于
架构设计
记录把授权部分从菜单模型中拆出来、收敛成独立组件的几个关键设计决策，以及当时为什么不选另一条路。
从 Figma 到编码 Agent：设计上下文的结构化、缓存与校验
发表于
分类于
架构设计
从成本约束出发，讨论设计 API 接入、凭据池调度、本地上下文缓存与校验，以及编码 Agent 的职责和验收边界。
PostgreSQL Index Bloat: Why Small Tables Can Have Huge Indexes
发表于
分类于
Database
A PostgreSQL index bloat remediation note: why tables with little data can still carry huge B-tree indexes, why ordinary VACUUM does not shrink them, and how to combine REINDEX, redundant-index cleanup, and write-model changes.