Verify Loop Skill — Productivity skill for Claude Code

Verify Loop

A Claude skill that runs a **supervised, self-verifying loop** on a coding task — instead of you hand-prompting every step.

Works in **Claude Code** and **Claude on the web** (claude.ai).

Everyone's saying "stop prompting, start writing loops." They're right — but the loop is the easy part. A loop with nothing checking its work is just the agent confidently agreeing with itself twenty times in a row. This skill is the loop done properly: the part that actually matters is **verification**.

---

What's a skill?

A skill is a small set of instructions you hand to Claude — a job description for one specific task. You install it once, trigger it with a phrase, and Claude knows how to handle that job from then on.

This skill teaches Claude how to run a verification loop: you give it a goal, and it works one task at a time, checking each change for real before moving to the next.

---

What it does

You give it a goal. It then:

• Pins down "done" — turns your request into a concrete, checkable outcome, and finds the strongest way to verify it.
• Works one task per lap — smallest next step from a progress checklist it keeps on disk (GOAL.md + PROGRESS.md), so its memory survives a fresh context and the git history stays clean.
• Verifies for real — after each change it actually runs the thing (starts the app and drives the flow, or runs the tests), with the check done by a separate perspective so the agent isn't grading its own homework. This is the step that 2–3×'s the quality.
• Commits on green — one task, one commit, only when it's observed working.
• Stops itself — max iterations, a no-progress detector, and green-commits-only. The same error three times means stop and get a human.

It runs **supervised by default**: it does a task, verifies, commits, then checks in — so you can't accidentally let it grind the wrong direction for an hour. When you trust the goal and the verification is solid, you can tell it to ru