@yaohui12138: I've finished reading it. Here are some key takeaways I've compiled for everyone: In this session, he primarily broke down a core mechanism overlooked by 90% of users: the CLAUDE.md context injection system. This system is divided into three levels: Enterprise-level: Organization-wide mandatory rules that cannot be overridden by individual settings. Project-level: Team-shared code standards and workflows. Loc...

I've finished reading it and compiled some key takeaways to share: In this session, he primarily broke down a core mechanism that 90% of users overlook: the **CLAUDE.md context injection system**. This system operates on three levels: - **Enterprise-level**: Organization-wide mandatory rules that cannot be overridden by individual settings. - **Project-level**: Team-shared code standards and workflows. - **Local-level**: Personal habits and shortcuts. Most people only @ files in conversations, but Boris's approach is different: he writes the team's CI tools, logging systems, and database connection rules into these three configuration layers, so that every time Claude starts up, it comes pre-loaded with your team's institutional memory. In addition to CLAUDE.md, he demonstrated three immediately applicable techniques: 1. **The `/loop` command (scheduled execution)** Allows Claude to automatically run tasks at set time intervals. - `/loop 30m /review` # Runs code review every 30 minutes - `/loop 1h "run unit tests"` # Checks tests every hour Boris said this is the command he recommends most often, as it solves the pain point of "repetitive checking tasks." 2. **Plan Mode (Plan first, execute later)** Before starting a large task, have one Claude instance write a plan, then spawn another Claude instance to review that plan from the perspective of a senior engineer. Only start coding once the plan is approved. Boris's exact words: "Spend 10 minutes on planning to save 2 hours of rework later." 3. **Keep CLAUDE.md under 200 lines** If it's too long, Claude may ignore your actual instructions. Boris suggests using `/init` to generate the initial version, then keeping only the rules that, if removed, would cause Claude to make mistakes. This workshop was covered by several tech communities, including Computer King Ada (Dian Na Wang A Da) and Industrial Intelligent Computing Network. Boris has also open-sourced 69 best practices on GitHub (the repository has already garnered 32,000 stars). Here are my thoughts: Tool usage can be divided into three levels of mastery: - **Level 1**: Using the tool like ChatGPT, starting from scratch with every conversation. - **Level 2**: Knowing how to @ files and pass context, but still doing it manually. - **Level 3**: Solidifying team knowledge into configurations, allowing the tool to carry its own memory. Boris's session provides the shortest path from Level 1 to Level 3.