/loop skill

SkillProductivity

Run the self-improving DeepSeek/Luna/Sol loop. Use when the user says "/loop", "$loop start", or asks to start a self-improving loop with a task, including a repeating 10-minute cadence.

Available today. Use it from your connected AI after setup.

Add ahel to your AI once: Claude, ChatGPT, Cursor, Claude Code or Codex. Then ask it to use this.

Then ask your AI: use the /loop skill skill

What this skill tells your AI

The instructions your AI receives, as published by codejunkie99/codex-self-improving-loop in skills/loop/SKILL.md and read by ahel’s review.

Run the self-improving loop: DeepSeek runs the task, Luna grades each run, Sol rewrites the prompt. The user says "$loop start " or "/loop ".

Start

  1. If the project has no prompt.md, copy the template from loop/prompt.md into the project root.

  2. Replace the task text in prompt.md with the user's task. Keep the DONE: convention from the template.

  3. Run from the project directory:

    bash ./loop.sh
    

    Add --every 600 when the user asks for a 10-minute cadence.

Utility menu

When the user mentions a utility, map it to a model:

UtilityModel
fast / flashopencode-go/deepseek-v4-flash
deep / deepseekopencode-go/deepseek-v4-pro
luna / grade / watchgpt-5.6-luna
sol / review / rewritegpt-5.6-sol
build / implementopencode-go/deepseek-v4-pro

The same mapping is available as slash commands: /loop, /flash, /deepseek, /luna, /sol, and /utility for the menu.

Stop conditions

The driver stops when the runner reports DONE:, when 6 cycles run, or when two consecutive cycles leave prompt.md unchanged. The user can also say stop, which means interrupt the run and keep the current prompt.md and log.md.

Signals

GitHub stars
48
Forks
8
Last commit
Sep 2026
Advanced
Item type
skill
Key
loop-codejunkie99
Source
github.com/codejunkie99/codex-self-improving-loop