loop-me
在这个工作区中,就我想构建的工作流规格访谈我。
Security Assessment
About loop-me
Loop-Me is a Chinese-language (zh-CN) workflow-design skill that interviews the user to produce executable workflow specifications. It runs a stateful grilling session whose only output is workflow specs, using a disciplined interviewing style: asking one question at a time, continuously probing, and attaching a recommended answer to every question. It solves the problem of turning vague, repetitive parts of a user's work life into precise, implementable specifications.
The skill frames the user's life as loops within loops, repeating patterns such as a career, a week, a morning, or a single recurring activity, and uses that lens to find which loops are predictable enough to be worth delegating and writing a spec for. A workflow is the spec for a loop that makes it executable. It offers a lightweight shared vocabulary used only when a workflow needs it, deliberately avoiding forced structure: Trigger (an event or a schedule that starts each run), Checkpoint (a human-in-the-loop verification or decision point), Push right (deferring checkpoints so the most work is done before a person is involved), and Brief (the compact, decision-ready summary shown at a checkpoint, never raw output). Specs live as files in workflows/*.md as the source of truth, with a NOTES.md capturing raw notes about the user's tools, channels, and their own terminology. A spec is done only when an implementer agent could build it without asking any questions.
It targets users who want to design and delegate recurring workflows and is invoked deliberately (model invocation is disabled), taking an optional argument naming a workflow to design. It creates, edits, and deletes markdown spec files within its own workspace, which is benign local file authoring with no external calls or risky operations.
FAQ
What language is this skill written in?
It is written in Simplified Chinese (zh-CN), a translation of Matt Pocock's loop-me skill. It conducts the interview and vocabulary in Chinese while retaining some English technical terms like Trigger, Checkpoint, and Brief.
What is the output of a session?
The sole output is workflow specifications stored as markdown files under workflows/*.md, treated as the source of truth. A NOTES.md file additionally captures raw notes about the user's tools, channels, and terminology.
How does it decide a spec is finished?
A workflow spec is done only when an implementer agent could build it without asking any questions. The grilling continues one question at a time until that bar is met; as long as open questions remain, it is not complete.
Does it impose a required structure on every workflow?
No. The shared vocabulary is used only when a workflow needs it and is not a checklist. A workflow does not require AI, a checkpoint, or a schedule unless the interview shows it is needed.
How is the skill invoked?
It has model invocation disabled and is triggered deliberately by the user, with an optional argument naming a workflow to design; it can also be left blank so the skill helps find a loop worth specifying.
Install loop-me
Quick Setup:
- Copy the skill folder to
.claude/skills/ - Claude will automatically detect and use the skill
Repository
vinvcn/mattpocock-skills-zh-cn