SuperAPI AI Directory
Partner with us中文AI newsFavorites
Skillsloop-build-path

loop-build-path

General & other

View repository

Updated 10-15 23:44

About this skill

Source Metadata

Original cangjie-skill frontmatter from the distillation run:

name: loop-build-path
description: |
  从"手动做事"到"系统自动做事"的四步渐进构建路径。
  当用户已经决定要做 loop、但不知道从何入手时;或团队正在将手工流程自动化时使用。
  不适用于: 尚未决定要不要做 loop 的任务、或已经稳定运行的 loop 优化。
  关键 trigger: "怎么开始做 loop"、"自动化第一步"、"从手动到自动"。
source_book: "Loop Engineering 视频合集"
source_chapter: 视频4 (Idoos Money) / 视频2 (Boris Cherny)
tags: [build-path, incremental, automation, getting-started]
related_skills: [loop-worthiness-test, loop-three-elements, goal-verification]
The Progressive Loop Build Path — Four Steps from Manual to Automatic

R — Reading (original text)

"The very first thing... do it manually, right? So, confirm the AI can even do the task at all by hand. Next, everyone is turn it into a skill... Next, everyone is the trigger. Now, it runs on a schedule or an event. It is still not a loop yet. It is just an automation. The loop starts, everyone, when you are adding the verification plus state." — Idoos Money (Video 4)

I — Interpretation (in my own words)

Do not build a loop system all at once; instead, proceed progressively through four stages:

  1. Manual validation: First have the AI complete the task manually to confirm that itcan do it. Do not build automation on a task you have not yet validated.
  2. Package into a Skill: Solidify the instructions into SKILL.md so that each execution is consistent. At this point it is "repeatable manual."
  3. Add a trigger: Add a schedule (cron) or event trigger. At this point it is "automation," but still not a loop — because there is no verification or iteration.
  4. Add verification + state: Add an automated verification step and persistent state records.Only now does it become a true Loop。

Key insight: Automation ≠ Loop. Automation is "executing on a schedule"; a Loop is the closed cycle of "execute → verify → adjust → execute again."

A1 — Past Application (case in the book)

Case 1: Topic inbox (Video 3)

  • Step 1: Manually use the research tool to pull information → confirm it is feasible
  • Step 2: Write the instructions as a skill (research + topic-score)
  • Step 3: Add a cron trigger (every morning at 8)
  • Step 4: Add inbox.md as persistent state + rating as verification

Case 2: Boris's multi-Loop system (Video 2)

  • First manually run support to confirm the AI can handle it → package into a skill → add a 30-minute trigger → add a signals folder as state

A2 — Future Trigger

  1. When starting your first loop: "I want to automate X, where do I start?"
  2. When loop building fails: "I built a loop but the output is very poor" → check whether Step 1 was skipped
  3. When a team is rolling out loops: Use this path as the "getting-started guide"
  4. When upgrading from automation to a loop: "My cron job has been running for a long time, how do I upgrade it to a loop?"

Language signals: "how to start building a loop," "first step to automation," "from manual to automatic," "how do I upgrade my cron"

Differences from adjacent skills:

  • loop-worthiness-test: deciding whether to do it (this skill is about how to do it after deciding)
  • loop-three-elements: static structure analysis (this skill is a dynamic build process)
  • goal-verification: focuses on verification design (this skill focuses on the overall path)

E — Execution (executable steps)

Step 1: Manual Proof

  • Manually have the AI complete the task 3 times
  • Record: success rate? output quality? time taken?
  • If it can't be done manually, don't build a loop

Step 2: Package into a Skill

  • Write the instructions as SKILL.md
  • Include: input, output, steps, boundaries
  • Ensure consistent results on every execution

Step 3: Add a Trigger

  • Choose: scheduled (cron) / event (webhook) / state change
  • At this point it is "automation" — it runs on a schedule, but does not self-adjust

Step 4: Add Verification + State

  • Define: how do you judge whether the output is good or bad? (see goal-verification)
  • Add: a persistent state file (recording historical operations)
  • Only now does it become a Loop — the system can adjust its behavior based on verification results

B — Boundary

Scenarios where you should not use this skill:

  1. You have already decided not to build a loop: you should first go through loop-worthiness-test
  2. A loop that is already running stably: what is needed here is optimization, not building from scratch
  3. Purely exploratory tasks: "let's see if the AI can do X" — still in the experimental stage, no formal build needed

The author's blind spots and limitations:

  • The four-step path assumes the task is already clear and does not address "how to discover tasks worth automating"
  • Step 1 "manual proof" may take a very long time for complex tasks, and the author does not give a criterion for "when to give up"
  • The video cases are all success paths, lacking failure handling for "what to do if you get stuck at some step"

Adjacent but easily confused methodologies:

  • Agile development: "small steps, fast iteration"; this skill is "small-step automation"
  • CI/CD pipeline: "code→test→deploy"; similar in structure, but this skill is aimed at AI agent tasks