Learning

海外用户增长与 AI 工作流的实践笔记。写的是方法和机制，不含任何真实账户与数据。Working notes on global user acquisition and AI workflows. Methods and mechanics only; no real accounts or data.

• 2026-08-28
  用户增长 × Claude Code 工作流全景（v2）User growth × Claude Code: the full workflow map (v2)

  底座（目录、规则分层、数据链路、判据、护栏、长上下文管理）→ 投放线五条 → 数据线七条 → 方法八条 → 待办。内部手册的对外脱敏版。Foundations (directory, rule layers, data links, metrics, guardrails, long-context management) → five paid-media lines → seven data lines → eight method notes → open items. De-identified public version of an internal manual.

  AI workflowoverview
• 2026-08-28
  长期任务的上下文结构：为什么这样设计Context structure for long-running work: why it's built this way

  问题诊断、几个渠道其实是同一个循环、分层原则、生命周期、参考了哪些方案。Problem diagnosis, one loop across channels, layering principles, lifecycle, and what was borrowed from where.

  AI workflowcontext engineering
• 2026-08-28
  投放优化运行机制：AI 怎么长期管多个渠道而不失忆How the ad-ops loop runs: keeping an AI agent on top of paid media for months

  会话时序、信息三层、动作→效果闭环、四条保证、上下文负担前后对比。Session sequence, three information layers, the action→outcome loop, four guarantees, and the before/after context cost.

  AI workflow