autonomous-agent-loops-plus

Built on affaan-m/ECC by @affaan-m (246,355 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 for safe agent loops, fixed code pipelines, and large plans with many agents.

Built for developers who want Claude Code to work through clear steps with checks, stop rules, and safe commits.

💡 Why

Long agent jobs can hang, retry forever, or edit the wrong files.

Two agents may change the same code at once.

A failed test may leave half-done work behind.

This skill gives you simple rules for small jobs and large plans. It plugs into your normal Claude Code flow.

⚡ Install

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

🛠️ Usage

Ask Claude Code to use the skill for a checked, step-by-step job:

Use the autonomous-loops skill.

Build the login form from docs/login.md.
Write tests first.
Run the related tests after each step.
Stop if a check fails twice.
Do not edit files outside src/login/ and tests/login/.
Commit only if every required check passes.

Expected output:

Step 1: Read the spec
Step 2: Add failing tests
Step 3: Build the login form
Step 4: Run related checks
Step 5: Review changed files
Step 6: Commit valid work or report why it stopped

The skill helps Claude pick the smallest useful loop. A small task can use a fixed pipeline. Separate work can use parallel agents. Large work can use an RFC with file owners, checks, and a merge owner.

🔄 What we changed vs upstream

• Rewrites the Japanese content into simpler English and adds explicit attribution to ECC.
• Adds a substantial safety section covering stop limits, retries, concurrent edits, secrets, deletion, commits, and escalation.
• Strengthens the pipeline example with set -euo pipefail, per-step testing, safer failure handling, and guarded commits.
• Clarifies execution behavior and adds practical edge-case guidance for hangs, partial work, flaky tests, and dirty worktrees.
• Introduces a likely typo in the compatibility note: autonomous-lo-loops instead of autonomous-loops; the improved file also appears truncated at the final bullet.

📜 License

This project is released under the MIT License. The upstream MIT license and credit are preserved in UPSTREAM_LICENSE.