Loop-Me Skill: Turn Repetitive Work into Actionable Workflow Specifications Through Follow-Up Questions

Skill Overview

Loop-Me is a Claude skill that helps you clarify recurring cycles in your life through “round after round of follow-up questions” and directly distills them into workflow specification documents. Its sole output is specification files in the workflows/ directory, with the goal of writing them clearly enough that an implementation-oriented Agent can execute them without asking any further questions.

Use Cases

1. Before handing daily or weekly routine work over to AI: For example, processing emails every morning, writing a weekly report, or regularly organizing your to-do list. You have an intuitive sense that these tasks could be automated, but cannot clearly articulate exactly what you want. Loop-Me asks questions one round at a time until these vague ideas become one or more fully specified documents.
2. When the requirements are still in your head and you do not know where to start: You have an automation workflow in mind, but have not yet figured out the trigger conditions, whether human confirmation is needed, or what form the output should take. The skill does not pressure you to provide answers. Instead, it recommends an answer before asking each question; you only need to confirm or correct it.
3. When you need to align expectations with an implementer (human or Agent): Once the specification is complete, any implementer can pick up workflows/*.md and start work immediately, without having to repeatedly ask, “What exactly does this mean?”

Core Features

1. The Loop Lens for discovering worthwhile tasks to delegate: The skill views a person’s life as “cycles nested within cycles”—a career, a week, a morning, or a repeatedly performed action. From these cycles, it identifies patterns that are consistent enough to be worth delegating, and proactively suggests cycles you may not have noticed.

2. One round, one question, with a recommended answer for each question: Each time, it presents only one round of questions rather than dumping a long checklist on you. Every question is followed by “Here’s how I suggest answering,” which you can accept or revise. The questioning continues until there are no unresolved issues remaining.

3. Real-time creation, modification, and deletion of specification files as the discussion progresses: As soon as a point is settled, the skill immediately creates, modifies, or deletes the corresponding workflows/*.md, rather than generating everything at the end. It also maintains NOTES.md, recording your working context, commonly used tools, the channels through which you process information, and your own terminology. When these notes are still too sparse, it interviews you first, before writing any specifications.

4. An on-demand shared vocabulary: Trigger (how a workflow starts: event-driven, such as a new email or issue, or schedule-driven, such as every morning), Checkpoint (a step requiring human confirmation or decision), Push right (pushing confirmation points as far back as possible so you are asked once, later, when everything is ready), Brief (the content presented at a confirmation point—a compact, decision-ready summary with links pointing to the actual outputs rather than dumping raw output on the person). These are tools, not a checklist: the skill does not enforce any particular structure. A workflow does not need AI, a checkpoint, or a schedule unless the questioning process shows that it genuinely needs one.

Frequently Asked Questions

What is Loop-Me?

It is a conversational skill that runs under the /loop-me command. Through continuous follow-up questions, it helps you clarify the workflow you want to build and write it into specification files. It does not write code or set up services; its deliverables are documents. It is configured to be invoked manually by you only; the model will not trigger it automatically.

How is it different from the brainstorming skill?

Brainstorming is for open-ended exploration of “what to do.” Loop-Me serves one purpose only: producing workflow specifications. It has a fixed vocabulary (triggers, checkpoints, Push right, Brief), fixed storage locations (workflows/*.md and NOTES.md), and a strict completion criterion.

When is it considered “complete”?

There is only one standard: after receiving the specification, an implementation-oriented Agent can execute it directly without asking you a single additional question. As long as even one question remains unresolved, it is not complete—the skill will continue asking questions.

Must a workflow involve AI, human confirmation, or scheduled tasks?

No, and this is explicitly not assumed in advance. All three are optional and are included only when the questioning process reveals that the workflow genuinely needs them. Some workflows are fully automatic with no checkpoints; some do not use AI at all.

Does it support designing multiple workflows at once?

Yes. The session is stateful, so you can clarify multiple cycles one by one in the same conversation, with each cycle corresponding to an independent specification file under workflows/.

What should I prepare before using it?

It is best to first consider what recurring activities you perform in daily life, what tools you currently use, and through which channels you receive information. If NOTES.md is still empty, the skill will interview you to fill in this information before beginning to write specifications.