Loop Engineering — Development skill for Claude Code

loop-engineering

A drop-in [Claude Code](https://github.com/anthropics/claude-code) skill for **loop engineering** — building a self-running agent loop that keeps working until a goal is met, instead of the manual write → wait → review → re-prompt cycle.

The skill walks you through the architecture and asks you the questions to set it up:

• The honesty test — four conditions that decide whether a loop is even worth building.
• Two axes — manual vs automatic (who closes the loop) and open vs closed (is there a gate).
• The three primitives — /goal vs /loop vs routines (/schedule), and which one to use.
• The gate — a Stop hook running a real test (un-foolable, zero tokens) for code, or a separate LLM-judge + rubric for non-testable work.
• The ceiling — max_turns / budget limits so the loop never spins forever.
• Model routing & cost — why a closed loop is cheaper, not a wallet-drainer.

The core rule: **never let the AI verify its own "done."** Maker and checker must be different things, or you get *false completeness* — code that claims to be finished but isn't.

Install

Drop the skill into your Claude Code skills directory:

git clone https://github.com/selmakcby/loop-engineering
cp -r loop-engineering ~/.claude/skills/loop-engineering

Then in Claude Code, the skill activates on prompts like "set up a loop", "run until tests pass", or "döngü mimarisi kur".

Read more

Full write-up (TR · EN · NL): [selma.codes/blog/loop-engineering](https://www.selma.codes/blog/loop-engineering)

MIT licensed.