Back to Details

create-brief

Compare original and translation side by side

🇺🇸

Original

English
🇨🇳

Translation

Chinese

Brief

简报

Turn conversation context into an executable 5-part brief.
The user has thought enough about a task — through 
/grill-me
, free-form conversation, or just typing a request — and is ready to hand it off. 
/create-brief
produces the structured handoff.
将对话上下文转换为可执行的5部分简报。
用户已对任务有足够思考——通过/grill-me、自由对话或直接输入请求——并准备好交付任务。
/create-brief
将生成结构化的交付文档。

Workflow

工作流程

  1. Read the conversation. Pull every signal about the task from the recent conversation, including any 
    /grill-me
    recap. If 
    $ARGUMENTS
    was provided (e.g., 
    /create-brief I want to draft a follow-up email...
    ), treat that as additional task input on top of the conversation.
  2. Fill what you can. Map signals to the 5 slots in the template below. Quote the user's own words where possible — don't paraphrase intent the user didn't express.
  3. Identify weak slots. A slot is weak if it's empty, vague, or relies on guessing the user's intent. The most commonly weak slots:
    • Verification — users rarely think in stopping conditions
    • Constraints — often unstated must-dos / must-not-dos
    • Output format — sometimes the user genuinely doesn't care
  4. Ask one focused question per weak slot. Use 
    AskUserQuestion
    . Each question must propose a recommended answer the user can accept with one tap. Skip the question entirely if you can derive a strong answer from the conversation.
  5. Respect "I don't have anything." If the user passes on a slot, record that as the slot value (e.g., "agent picks" for Output format) — don't force content. Absence is a real signal to the downstream agent.
  6. Output the brief in a single markdown code block so the user can copy it cleanly. Use the template below, replacing each bracketed placeholder with the user's content.
  7. Suggest the next step in one line. "Hand this to a fresh Claude session for cleanest execution, or paste it back here to run it now."
  1. 读取对话内容：从近期对话中提取所有与任务相关的信息，包括任何/grill-me的回顾内容。如果提供了
    $ARGUMENTS
    （例如：
    /create-brief I want to draft a follow-up email...
    ），则将其视为对话之外的额外任务输入。
  2. 填充已有信息：将提取到的信息映射到下方模板的5个模块中。尽可能引用用户的原话——不要转述用户未表达的意图。
  3. 识别待完善模块：如果某个模块为空、表述模糊或需要猜测用户意图，则该模块属于待完善状态。最常见的待完善模块包括：
    • 验证——用户很少会考虑终止条件
    • 约束——常存在未明确说明的必做/禁止事项
    • 输出格式——有时用户确实不在意格式
  4. 针对每个待完善模块提出一个聚焦的问题：使用
    AskUserQuestion
    。每个问题都应提供一个推荐答案，方便用户一键确认。如果能从对话中推导出明确答案，则跳过该问题。
  5. 尊重用户的“我没有相关内容”：如果用户跳过某个模块，将其记录为模块值（例如：输出格式填写“agent picks”）——不要强制填充内容。缺失信息对下游Agent来说也是真实的信号。
  6. 将简报输出在单个Markdown代码块中，方便用户干净地复制。使用下方模板，将每个带括号的占位符替换为用户提供的内容。
  7. 用一句话建议下一步操作：“将此简报交给新的Claude会话以获得最清晰的执行效果，或粘贴回此处立即执行。”

Output template

输出模板

markdown
undefined
markdown
undefined

Brief: [one-line task title]

Brief: [one-line task title]

Context [Background the agent needs: situation, goal, audience, files/URLs/docs to consult.]
Task [What you want done. A clear sentence for simple work; a paragraph or sub-bullets when the task needs detail to be accomplishable.]
Constraints [Rules, must-dos, must-not-dos. Bullet list.]
Verification — don't finish until: [Stopping conditions that tell the agent it's done. Bullet list.]
Output format [Shape of the deliverable, or "agent picks" if the user passed.]
undefined
Context [Background the agent needs: situation, goal, audience, files/URLs/docs to consult.]
Task [What you want done. A clear sentence for simple work; a paragraph or sub-bullets when the task needs detail to be accomplishable.]
Constraints [Rules, must-dos, must-not-dos. Bullet list.]
Verification — don't finish until: [Stopping conditions that tell the agent it's done. Bullet list.]
Output format [Shape of the deliverable, or "agent picks" if the user passed.]
undefined

Anti-patterns

反模式

  • Shipping 
    [TODO]
    placeholders.     If a slot is weak, ask. If the user passes, record their pass — don't leave a placeholder.
  • Paraphrasing intent. If the user said "be brief," don't translate to "concise and impactful." Keep their words.
  • Over-asking. If a slot is reasonably derivable from context, fill it. Save questions for what's genuinely missing.
  • Bloating the brief. Keep each slot as tight as the work demands. Task can grow when accomplishing the work needs detail, but don't pad it with rules (those go in Constraints) or shape (Output format). The brief is a handoff artifact, not documentation.
  • Forgetting Verification. This is the slot users most often skip and the one that most affects output quality. Always raise it, even if just to confirm the user has nothing specific.
  • 保留
    [TODO]
    占位符    ：如果模块待完善，就询问用户。如果用户跳过，记录其选择——不要留下占位符。
  • 转述用户意图：如果用户说“be brief”，不要翻译成“concise and impactful”。保留用户的原话。
  • 过度询问：如果能从上下文合理推导出模块内容，直接填充。仅对真正缺失的内容提问。
  • 冗余填充简报：每个模块的内容应尽可能精简，仅保留任务所需的信息。当任务需要细节才能完成时，任务模块可以适当扩展，但不要添加规则（规则应放在约束模块）或格式要求（格式要求放在输出格式模块）。简报是交付工件，不是文档。
  • 忽略验证模块：这是用户最常跳过的模块，也是对输出质量影响最大的模块。即使只是确认用户没有特定要求，也要始终提及该模块。