Looping Skills — 40 Loops de Agentes em Coreano
NewTurn2017/looping-skills
40 agent loops coreanos para teste, CI, manutenção e automação
Ver no GitHubPlugin Claude Code que localiza 40 loops de agentes (do loops.elorm.xyz) para coreano, oferecendo skills prontas para automação de testes, CI/CD, manutenção e debugging. Cada loop é um sistema com trigger, feedback gate e condição de saída — em vez de prompts únicos.
Instalação
npx convez-ai NewTurn2017/looping-skills1 comando no terminal: detecta o formato do repositório (marketplace ou plugin avulso) e instala via Claude Code. Requer Node 18+ e o Claude Code instalado.
Métricas
Casos de uso
- Executar testes repetidamente até passar (test-until-green)
- Monitorar falhas de CI e corrigir automaticamente (ci-failure-watcher)
- Atualizar dependências uma por uma com validação (dependency-upgrade-one-by-one)
- Refinar código até atingir métrica (coverage-until-threshold)
- Loops de debugging reflexivo e investigação
Veredito

Recurso técnico robusto para equipes coreanas usando Claude Code, focado em loops agentas (não prompts únicos) com feedback gates e guardrails. Bem estruturado, mas requer entender o conceito de loops para aproveitar — não é para iniciantes em automação.
README
A tradução para o português deste README ainda está na fila.
Looping — Korean agent-loop skills for Claude Code
A Claude Code plugin that localizes the 40 agent loops from loops.elorm.xyz into Korean.
Installing it adds 40 skills named looping:<loop>, each with a Korean explanation of what / when / how to use it, plus a ready-to-run guide for Claude Code (/loop, /schedule, hooks).
The skill bodies are written in Korean — this plugin is aimed at Korean-speaking Claude Code users.
Quick start
Add this repository as a marketplace in Claude Code, then install the plugin:
/plugin marketplace add NewTurn2017/looping-skills
/plugin install looping@looping-marketplace
Reopen the session and the looping:<loop> skills become active. Say something like "keep running tests until they pass" and the right loop triggers automatically.
What is a loop? (prompt vs loop)
- A prompt is a one-shot instruction. Run once, done.
- A loop is a system that keeps prompting the agent until a goal is met, with a feedback gate (a check) between each pass.
In one line: prompts are ingredients, loops are recipes. Trigger → act → check → repeat.
Five core concepts: trigger (manual / interval / event) · self-pacing (the agent stops itself) · feedback gate (a check each pass) · exit condition / max iterations · guardrails (anti-gaming rules so the exit condition can't be faked).
Usage
Each loop is a skill. Two ways to use it:
- Auto-trigger — say "keep fixing tests until they all pass" and Claude consults the matching loop (
looping:test-until-green) on its own. - Explicit — call a skill by name (
looping:<loop>) when you want to pick it yourself.
How it runs depends on the trigger type (each skill explains this in Korean):
- manual → paste the kickoff prompt; the agent self-paces.
- interval →
/loop <interval>(this session) or/schedule(cron-style, survives session close). - event (hook) → wire it into
.claude/settings.jsonhooks to fire on edit / commit / merge.
The 40 loops (by category)
| Category | Count | Examples |
|---|---|---|
| Testing | 13 | test-until-green, flaky-test-triage, coverage-until-threshold, e2e-until-green |
| CI | 5 | ci-failure-watcher, ship-pr-until-green, pr-babysitter |
| Maintenance | 5 | knip-until-clean, dependency-upgrade-one-by-one, docs-sync-after-edits |
| Review | 2 | pr-self-review, de-sloppify-pass |
| Automation | 2 | ralph-story-executor, guardrails-learning-loop |
| Debugging | 2 | reflexion-debug-loop, investigation-script-loop |
| API | 2 | api-contract-until-match, openapi-sync-until-valid |
| DevOps | 2 | deploy-verification-loop, staging-smoke-test |
| Quality / Planning / Docs / Security / Performance / Git / DB | 1 each | a11y-audit-until-clean, spec-first-ship, migration-until-applied, … |
→ Full catalog (one-line descriptions for all 40, in Korean): looping/README.md.
Repository layout
looping-skills/
├── .claude-plugin/marketplace.json # marketplace definition
├── looping/
│ ├── .claude-plugin/plugin.json # plugin manifest
│ ├── README.md # concepts, install, usage + full 40-loop catalog (Korean)
│ └── skills/<40>/SKILL.md # the 40 looping:<loop> skills
├── README.md / README.ko.md # this document
└── LICENSE
Credits
- Original: loops.elorm.xyz by elorm — the source of the loop concept, the 40 workflows, the kickoff prompts, and the guardrail rules.
- This repository is an unofficial Korean localization that re-frames that content for Claude Code's execution model. Each skill preserves the original English kickoff and a source link in a
<details>block. - If you like the original, you can support it via Buy me a coffee.
License
MIT — covers the Korean localization and packaging work. The loop concept and original prompts originate from loops.elorm.xyz, as noted in Credits above.