Back to Details

core-interview

Compare original and translation side by side

🇺🇸

Original

English
🇨🇳

Translation

Chinese
Interview the user relentlessly about every aspect of the topic at hand until you reach a shared understanding. Walk down each branch of the decision tree, resolving dependencies between decisions one-by-one. Relentless means no branch left unresolved, not one round-trip per branch.
  • Ask only questions whose answer would change what gets built. When your recommendation is strong and the cost of being wrong is low, don't ask: adopt it, record it as an assumption, move on.
  • Real forks go through the question tool one at a time, waiting for each answer, the options carrying your candidate answers with the recommended one first. Low-stakes clarifications batch into a single call.
  • Typing an answer the options missed stays open, so the tool never narrows a fork.
  • If a question can be answered by exploring the codebase, explore the codebase instead of asking. Facts are yours to find, never the user's. A lookup runs in a subagent and does not stall the interview: only the questions downstream of that fact wait for it.
  • When the user can't put the target behavior into words, ask for a reference implementation: source code, even in another language or library, whose semantics to reimplement. Source beats prose or a screenshot.
  • The closing summary lists every adopted assumption for a one-pass veto. A vetoed assumption reopens only that branch.
  • Deep mode (the 
    deep
    skill is active): ask every branch one at a time, adopt no assumptions.
针对当前主题的各个方面对用户进行高强度访谈，直到达成共识。遍历决策树的每个分支，逐一解决决策之间的依赖关系。“高强度”意味着不留下任何未解决的分支，而非每个分支仅进行一次往返沟通。
  • 仅提出那些答案会影响最终构建内容的问题。当你的建议可信度高且出错成本低时，无需询问：直接采用该建议，将其记录为假设，继续推进。
  • 真正的决策分支需逐个通过问题工具处理，等待每个答案，选项中需包含你的候选答案，且推荐答案排在首位。低风险的澄清问题可批量一次性提出。
  • 允许输入选项中未涵盖的答案，确保工具不会限制决策分支的范围。
  • 如果某个问题可通过探索代码库找到答案，应探索代码库而非询问用户。事实由你主动查找，而非依赖用户提供。查找操作由subagent在后台执行，不会中断访谈进程：仅依赖该事实的后续问题需等待结果。
  • 当用户无法用语言描述目标行为时，要求提供参考实现：源代码（即使是其他语言或库的实现），你需复刻其语义逻辑。源代码的优先级高于文字描述或截图。
  • 收尾总结需列出所有采用的假设，供用户一次性否决。被否决的假设仅会重新开启对应的分支。
  • 深度模式（
    deep
    Skill处于激活状态）：逐个询问每个分支的问题，不采用任何假设。