loop-me

Compare original and translation side by side

🇺🇸

Original

English

🇨🇳

Translation

Chinese

运行一个 stateful

/grilling

session，唯一输出是 workflow specs。使用 grilling discipline：持续追问、一次一个问题、每个问题都附推荐答案。围绕下面的 vocabulary 和 goal 来访谈。随着 grilling 解决问题，创建、编辑、删除 specs。

Run a stateful

/grilling

session, with the only output being workflow specs. Use the grilling discipline: keep asking questions, one at a time, and attach recommended answers to each question. Conduct the interview around the following vocabulary and goals. As grilling resolves issues, create, edit, and delete specs.

The loop lens

Loop 是用户生活中的重复模式：career、week、morning，或单个重复活动。把生活看成 loops within loops，会揭示活动到底有多 predictable，也因此揭示哪些值得 delegating。用这个 lens 找到值得写 spec 的 loops，并提出用户尚未注意到的 loops。

Workflow 是一个 loop 的 spec，并让它变得可执行。你在 loop 上运行 workflow；loop 是它的运行实例。Workflows 存在

workflows/*.md

中，并作为 source of truth。

Loop is a repeating pattern in the user's life: career, week, morning, or a single repeated activity. Viewing life as loops within loops reveals how predictable activities are, and thus which are worth delegating. Use this lens to find loops worthy of writing specs for, and propose loops the user hasn't noticed yet.

Workflow is the spec for a loop that makes it executable. You run a workflow on a loop; the loop is its running instance. Workflows reside in

workflows/*.md

and serve as the source of truth.

Vocabulary

只有当 workflow 需要时才使用 shared language；它不是 checklist。不要强制任何结构：除非 grilling 表明需要，否则 workflow 不需要 AI、不需要 checkpoint，也不需要 schedule。

Trigger - 每次运行的触发物：一个 event（新 email、新 issue）或一个 schedule（每天早晨）。Event-triggering 通常更高效。
Checkpoint - human-in-the-loop 点，用户在这里验证或决策。有些 workflows 没有 checkpoint，可自主运行；有些完全不使用 AI。
Push right - 尽可能延后 checkpoint。在涉及人之前先完成最大量工作，让用户只被晚些时候问一次，并且一切都已准备好。
Brief - checkpoint 展示的内容：紧凑、decision-ready 的 summary，说明产出了什么、为什么、并链接到底层 asset；永远不是 raw output。用户读 brief，不读 draft。Review 速度至关重要。

Only use shared language when the workflow requires it; it's not a checklist. Do not force any structure: unless grilling indicates a need, a workflow doesn't require AI, checkpoints, or schedules.

Trigger - The initiator of each run: an event (new email, new issue) or a schedule (every morning). Event-triggering is usually more efficient.
Checkpoint - A human-in-the-loop point where the user verifies or makes a decision. Some workflows have no checkpoints and can run autonomously; some don't use AI at all.
Push right - Delay checkpoints as much as possible. Complete the maximum amount of work before involving humans, so the user is only asked once later, and everything is ready.
Brief - The content presented at a checkpoint: a compact, decision-ready summary that explains what was produced, why, and links to underlying assets; never raw output. Users read the brief, not the draft. Review speed is critical.

Definition of done

当 implementer agent 能在不问任何问题的情况下构建它时，workflow spec 才算 done。持续 grilling 到那时；只要还有问题，就不算完成。

A workflow spec is done when the implementer agent can build it without asking any questions. Keep grilling until that point; as long as there are still questions, it's not complete.

The workspace

```
workflows/*.md
```
- 每个 workflow 一个 spec。
```
NOTES.md
```
- 关于用户世界的 raw notes：他们使用的 tools、处理的 channels，以及他们对两者的自有术语。当它为空或很薄时，先访谈用户的世界，再写任何 spec。随着 fuzzy terms 浮现，把它们打磨成 canonical terms，并记录在这里。

```
workflows/*.md
```
- One spec per workflow.
```
NOTES.md
```
- Raw notes about the user's world: the tools they use, the channels they handle, and their own terminology for both. When it's empty or thin, interview the user about their world first before writing any specs. As fuzzy terms emerge, refine them into canonical terms and record them here.