Loop Maker

loop-maker

A portable agent skill — works in Claude Code, Codex, Hermes, and OpenClaw — that interviews you and scaffolds a self-running autonomous loop, complete with verifier, state file, and human gate, so you don't build one wrong.

Why most loops go wrong

• Forgot to write a separate verifier (the agent judged its own work)
• Leaked changing state into the skill file (the loop lost memory between runs)
• Shipped a loop with no stop condition (it ran forever and burned budget)

`loop-maker` catches all three before you write a line.

Here's what it'll ask you

🛠 To build a loop, it asks you 7 things — one at a time:
 1. Goal        — what checkable condition means "done for now"? (a true/false test, not a vibe)
 2. Trigger     — what starts each run: a schedule, an event, or run-until-done?
 3. Discovery   — how does it find the work to do each round?
 4. Action      — what's it allowed to do, and through which tools?
 5. Verification— who checks the result, and against what? (a separate judge)
 6. State       — where does "what's done / what's left" live, outside the chat?
 7. Human gate  — which actions are irreversible and must ask you first?

At a glance

What it does	Runs a 7-question wizard; catches missing verifiers, state leaks, and runaway loops before scaffold
What you get	A ready-to-run loop folder: SKILL.md + STATE.md + verifier script + stop condition wired in
How you start	/loop-maker, or just describe an automate/schedule/monitor task — fires without the word "loop"
Install	git clone … ~/.claude/skills/loop-maker — or see Install below

Wizard UX

Q3/7  ▓▓▓░░░░  43%
---
🎙 elicit  →  🔎 survey  →  🎯 select  →  🏗 scaffold
(you are here: 🎯 select)
---
┌─ LOOP BLUEPRINT ───────────────────────────────────┐
│ GOAL      ✓ verifiable                             │
│ TRIGGER   schedule · 08:00                         │
│ VERIFY    ✓ separate script