Skip to content

Latest commit

 

History

1 Commit

Folders and files

NameName
Last commit message
Last commit date
 
 
 
 
 
 
 
 
 
 

Repository files navigation

continuous-agent-loop-plus

hero

License: MIT Works with Claude Code Skill version 1.0 Zero dependencies

Built on affaan-m/ECC by @affaan-m (249,486 stars, MIT). All credit for the original idea to them. This fork improves and repackages it; upstream license preserved in UPSTREAM_LICENSE.

A Claude Code skill that keeps long tasks moving with small steps, clear checks, safe retries, and firm stop rules.

💡 Why

Long tasks can drift, repeat bad fixes, or stop without a clear result.

This skill gives Claude Code a steady work loop. Each pass does one small task, checks the result, and records what happened.

It is for developers who want safer work across many agent passes. It plugs into your current flow and has no extra packages.

📦 Install

Replace OWNER with the GitHub account that hosts this repo, then run one command:

mkdir -p ~/.claude/skills/continuous-agent-loop && curl -fsSL https://raw.githubusercontent.com/OWNER/continuous-agent-loop-plus/main/skill/SKILL.md -o ~/.claude/skills/continuous-agent-loop/SKILL.md

🚀 Usage

Ask Claude Code:

Use continuous-agent-loop to add a CSV export button.

Done rules:
- Show the button on the report page.
- Download a valid CSV file.
- Keep headers when the report is empty.
- Escape commas, quotes, and new lines.
- Pass all current tests.
- Add and pass one export test.
- Stop after two failed tries.

Expected progress:

Pass 1:
  Added the export code.
  Ran the export test.
  Result: Failed because quotes were not escaped.

Pass 2:
  Fixed CSV escaping.
  Ran the export test and all current tests.
  Result: All checks passed.

Stop:
  Saved the result.
  Marked the goal done.

If the same quote bug fails again, the loop stops, saves the error, and asks for help.

🔁 How the loop works

Each pass follows four rules:

  1. Pick the smallest useful task.
  2. Make the change and run the needed checks.
  3. Save what changed, passed, and failed.
  4. Continue only when there is clear progress.

Before work starts, set the goal, scope, tests, done rules, retry limit, time or cost limit, and points where a person must step in.

The loop stops when all checks pass, a limit is hit, the same cause fails twice, access is missing, or the next step may cause harm.

🛠️ What we changed vs upstream

  • 日本語の簡潔な概要から、平易な英語による具体的な運用手順へ全面的に拡張された。
  • 各パスの小タスク化、検証、進捗記録、継続・停止判断を含む明確なループ手順が追加された。
  • 開始条件、完了基準、再試行回数、時間・コスト上限、人による介入条件などの事前設定と停止ルールが追加された。
  • 失敗モードと復旧手順が大幅に具体化され、同じ根本原因で再失敗した場合は停止して支援を求める方針になった。
  • CSVエクスポート機能を題材に、完了条件、2回の実行、成功・失敗時の停止判断を示す実例が追加された。

📄 License

This project is licensed under MIT. The upstream MIT license and credit are preserved in UPSTREAM_LICENSE.

About

Free Claude Code skill for steady agent loops with tests, quality checks, stop rules, and safe recovery. Built on affaan-m/ECC.

Topics

Resources

Stars

0 stars

Watchers

0 watching

Forks

Releases

Packages

Contributors