[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"project-94398":3},{"id":4,"name":5,"fullName":6,"owner":7,"repo":5,"description":8,"homepage":9,"htmlUrl":9,"language":9,"languages":9,"totalLinesOfCode":9,"stars":10,"forks":11,"watchers":12,"openIssues":13,"contributorsCount":13,"subscribersCount":13,"size":13,"stars1d":13,"stars7d":13,"stars30d":14,"stars90d":13,"forks30d":13,"starsTrendScore":13,"compositeScore":15,"rankGlobal":9,"rankLanguage":9,"license":16,"archived":17,"fork":17,"defaultBranch":18,"hasWiki":19,"hasPages":17,"topics":20,"createdAt":9,"pushedAt":9,"updatedAt":21,"readmeContent":22,"aiSummary":23,"trendingCount":13,"starSnapshotCount":13,"syncStatus":12,"lastSyncTime":24,"discoverSource":25},94398,"gauntlet-loop","robonuggets\u002Fgauntlet-loop","robonuggets","Turn any goal into a short prompt that makes your agent set a real quality bar, run builder and critic pairs, compare blind, and loop until it wins.",null,273,32,2,0,121,54.56,"Creative Commons Attribution 4.0 International",false,"main",true,[],"2026-08-24 04:01:22","\u003Cp align=\"center\">\n  \u003Cimg src=\"assets\u002Fbanner.png\" alt=\"gauntlet loop\" width=\"100%\">\n\u003C\u002Fp>\n\n# Gauntlet Loop\n\nA skill that turns any goal into one short, paste-ready prompt. That prompt makes your agent pick a real quality bar, split the work into small pieces, run a builder and a separate harsh critic on each one, compare blind against the bar, and keep looping until it wins.\n\nMost agent output stops at \"good enough\" because nothing is holding it to a standard. This gives it a standard it cannot argue with.\n\n> The gauntlet loop is [Matt Shumer's](https:\u002F\u002Fgithub.com\u002Fmshumer) idea. He wrote the original prompt and named the technique while building [Claude of Duty](https:\u002F\u002Fgithub.com\u002Fmshumer\u002FClaude-of-Duty). This repo packages that pattern as a reusable skill.\n\n## Quick start\n\n```\ngit clone https:\u002F\u002Fgithub.com\u002Frobonuggets\u002Fgauntlet-loop\n```\n\nCopy the skill folder into your project:\n\n```\ncp -r gauntlet-loop\u002F.claude\u002Fskills\u002Fgauntlet-loop your-project\u002F.claude\u002Fskills\u002F\n```\n\nThen in your agent:\n\n```\n\u002Fgauntlet-loop build me a pricing page for my SaaS\n```\n\nIt offers you 2 or 3 quality bars to aim at, you pick one, and it hands back a single prompt you paste into a fresh session.\n\n## What's included\n\n```\n.claude\u002Fskills\u002Fgauntlet-loop\u002F\n└── SKILL.md      # the whole skill, one file\nREADME.md\nLICENSE           # CC BY 4.0\n```\n\n## How it works\n\n1. **You give a goal.** Anything. A site, an essay, a CLI tool, a research brief.\n2. **It offers 2 or 3 bars.** Each one is a specific, real thing your agent can actually fetch and compare against. Not \"award-winning design\", but a named page, a named post, a named repo.\n3. **You pick one.** It writes one short prompt, around 150 words, and stops.\n4. **You paste it into a fresh session.** That agent splits the work, runs builder and critic pairs, and loops.\n\nThe critic is the part that matters. It is a separate agent with fresh context, it opens the actual output, it puts your work next to the bar with the labels stripped, and it says which one is better. Not a score out of 10, which drifts upward every round. A pick.\n\nThe loop exits when your work wins the blind comparison, or when you stop the run. Never after a fixed number of rounds.\n\n## Why a bar and not a rubric\n\nA rubric asks the agent to grade itself against words it wrote. A bar makes it compare against something that already exists and is undeniably good.\n\nThe skill will not accept a vague bar. It checks three things before it writes anything:\n\n- **Named.** A specific thing, not a category.\n- **Fetchable.** The critic can screenshot it, read it, run it, or open it. If the agent cannot get the reference, it hallucinates the comparison and approves everything.\n- **Comparable.** Both can sit side by side and a judge can pick one.\n\n## Examples\n\n```\n\u002Fgauntlet-loop a landing page for my running brand, dark and green, has to feel alive\n```\nBar becomes a specific brand's live campaign page, screenshotted at desktop and mobile.\n\n```\n\u002Fgauntlet-loop a 2000 word explainer on vector databases for non-engineers\n```\nBar becomes a named writer's actual published posts, judged on which one a non-engineer understands faster.\n\n```\n\u002Fgauntlet-loop a CLI that formats JSON logs\n```\nBar becomes a named tool's implementation plus its benchmark, so taste and a number both have to win.\n\n## Works with any agent\n\n`\u002Floop` and `ultracode` are Claude Code features. `\u002Floop` reruns a prompt until you stop it, and `ultracode` opts a turn into multi-agent orchestration.\n\nFor any other agent, the skill swaps those two lines for plain instructions: keep looping until the critic picks ours, and run the builders and critics as parallel subagents. The structure is identical.\n\n## What breaks it\n\n- A vague bar. The critic invents a comparison and approves everything. By far the most common failure.\n- The builder judging its own work. The critic needs fresh context and no knowledge of how hard the builder tried.\n- A soft critic. Give it a binary job, not a score.\n- A fixed round count. The exit is winning, or you calling it.\n\n## Credit\n\nThe gauntlet loop technique is **[Matt Shumer's](https:\u002F\u002Fgithub.com\u002Fmshumer)**. He built [Claude of Duty](https:\u002F\u002Fgithub.com\u002Fmshumer\u002FClaude-of-Duty), wrote the [original prompt](https:\u002F\u002Fgithub.com\u002Fmshumer\u002FClaude-of-Duty\u002Fblob\u002Fmain\u002Fprompt.md), and named the loop. Every idea underneath this skill - the harsh critic, the blind comparison, the refusal to stop until the work wins - comes from that prompt.\n\nThis repo is not the technique. It is a skill that writes a gauntlet loop prompt for you, for any goal, so you do not have to hand-write one each time.\n\nRelated reading: [Anthropic on building effective agents](https:\u002F\u002Fwww.anthropic.com\u002Fengineering\u002Fbuilding-effective-agents), which covers the evaluator pattern the loop is built on.\n\n## License\n\nCC BY 4.0. Free to use with attribution.\n\nSkill by Jay E at [RoboNuggets](https:\u002F\u002Frobonuggets.com). Technique by Matt Shumer.\n","Gauntlet Loop 是一个面向 AI 代理（agent）的提示工程技能，用于将任意目标转化为具备严格质量约束的可执行提示。其核心是引入不可辩驳的‘质量标杆’（如真实存在的网页、文章或代码库），驱动代理拆分任务、并行运行构建者与独立批评者、进行盲对比评估，并持续迭代直至输出胜出标杆。技术上依赖外部可获取、具名且结构可比的参照物，规避主观评分漂移。适用于需要高可靠性交付的场景，如营销页面生成、技术文档撰写、产品文案打磨等对输出质量有明确对标要求的 AI 应用。","2026-08-08 02:30:03","CREATED_QUERY"]