用 Grok Bot 做外层循环，把调研与编码执行分开
作者主张不要让一个 agent 在同一上下文里调研、规划和执行，而应让 Grok Bot 收集 GitHub、Slack、Notion 和文档信息并生成结构化提示词，交给专用编码 agent 执行。
One of the biggest mistakes in AI-assisted development is making one agent research, plan and execute inside the same context Grok Bot offers a cleaner architecture: use it as the outer loop. It gathers information from GitHub, Slack, Notion and documentation, identifies the relevant files and constraints, then builds a structured prompt for a specialized coding agent to execute. Research and implementation have different context requirements. Mixing them creates unnecessary context, distracts the executor and makes handoffs harder to control. The better setup separates discovery from execution, keeps the coding agent focused, and leaves implementation to the tool designed for it. This is already documented by a SpaceXSI developer working with Grok Bot.
这条是引用着另一条写的：「用 Operator Bot 与 Reviewer Bot 把重复工作固化成技能」