← ClaudeAtlas

gauntlet-looplisted

Domain-agnostic method for producing top-tier work with the Gauntlet Loop — split → build → blind-critic → repeat — against a hard "bar" the agent cannot argue its way around. A lead splits a goal into independently gradeable parts; each part gets a specialist builder plus a ruthless blind critic (clean context) who inspects the REAL artifact and only passes it when it beats the bar. Use for ANY kind of work: coding, writing, design, data analysis, research, prompt/agent evaluation, blue-team detection engineering. On invocation, identify the work type and read the matching file in references/domains/ for how to pick the bar plus a checklist and worked example. Trigger keywords: "gauntlet loop", "iterate to a high bar", "build-critic loop", "builder critic", "make this world-class", "beat a reference", "test-as-bar", "eval-as-bar", "blind A/B", "write a detection rule", "keep looping until it's great". For security vulnerability hunting use the separate security-vuln-gauntlet skill.
trilwu/gauntlet-loop-skills · ★ 3 · AI & Automation · score 66
Install: claude install-skill trilwu/gauntlet-loop-skills
# Gauntlet Loop — core skill (domain-agnostic) A disciplined loop for producing high-quality work in **any** domain: **build → blind critique → revise → repeat**, against a bar the agent cannot talk its way past. Coding, writing, design, data, research, prompt engineering and detection all use the *same* loop — they differ only in **how you pick the bar**, **what the critic inspects**, and **the checklist**. Those live in `references/domains/`. > **Origin:** the Gauntlet Loop method is **Matt Shumer's**, from *"How to Run a Gauntlet Loop"* (https://somethingbig.ai/gauntlet-loop), with the companion repo `mshumer/Claude-of-Duty`. This skill is an independent, attributed adaptation — see `../../CREDITS.md`. ## Prerequisite: a real agent harness This loop **requires an agentic harness** that can open files, run code, render output, take screenshots, call tools, and **spawn sub-agents with their own clean context** (Claude Code, Cowork, Codex, or equivalent). A single-turn chat cannot produce an independent blind critic — if you can only reply in one context, say so and fall back to a single self-review pass instead of pretending the loop ran. For large multi-agent runs, use the harness's parallel-agent facilities (in Claude Code: sub-agents, or `/ultracode` when available). ## The four pillars 1. **A bar the agent cannot argue around.** The strongest form: the artifact must **match or beat something real** — a reference implementation, a test suite, a top product's screen