Gauntlet Loop Skill — Development skill for Claude Code

Gauntlet Loop Skill

[](https://skills.sh/arjunkshah12345-hash/gauntlet-loop-skill/gauntlet-loop)

An agent skill that applies **Matt Shumer’s Gauntlet Loop** to ambitious software, game, product-design, writing, research, and creative work.

A lead agent sets a concrete, inspectable quality bar, decomposes the work, runs separate **builder** and **critic** agents, inspects the real artifact, and iterates until the work meets the bar (or you stop the run).

Independent adaptation of the method described by [Matt Shumer](https://x.com/mattshumer_). Not an official Matt Shumer skill.

The loop

build → inspect the real output → compare against the bar → fix the largest gap → repeat

The builder does not grade its own work. The critic gets the goal, constraints, quality bar, and artifact — not the builder’s rationale.

Install

Via [skills.sh](https://skills.sh/arjunkshah12345-hash/gauntlet-loop-skill/gauntlet-loop):

npx skills add arjunkshah12345-hash/gauntlet-loop-skill --skill gauntlet-loop

Or copy `SKILL.md` into your agent skills directory (Cursor: `~/.cursor/skills/gauntlet-loop/SKILL.md`, or `~/.agents/skills/gauntlet-loop/SKILL.md`).

When to use it

Ask for the Gauntlet Loop when you want unusually high-quality autonomous work judged against a real standard, for example:

• “Run this through the gauntlet”
• “Use Matt Shumer’s gauntlet loop”
• “Separate builder and critic until it wins against the reference”

Skip it for tiny edits, tasks with no inspectable artifact, or work where repeated evaluation costs more than it’s worth.

What’s in the skill

Piece	Role
Outcome brief	Goal, constraints, quality bar, verification surface
Quality bar	References, metrics, or acceptance tests — not vibes
Evaluation harness	Screenshots, tests, browser flows, benchmarks
Builder / critic split	Fresh-context critique