Turn a written feature backlog into a set of small, independently implemented stories.
Run long, multi story coding sessions without losing earlier context or quality.
Keep a durable log of learnings and conventions that future coding sessions can read.
Let Claude Code manage a git branch and commit history for a batch of related changes.
| aronprins/claude-loop | 00kaku/wp-rest-playground | 1ncendium/aibuster | |
|---|---|---|---|
| Stars | 5 | 5 | 5 |
| Language | — | JavaScript | Python |
| Setup difficulty | easy | hard | moderate |
| Complexity | 2/5 | 3/5 | 3/5 |
| Audience | developer | developer | ops devops |
Figures from each repo's GitHub metadata at analysis time.
Requires Claude Code already installed and authenticated, plus a git repository to work in.
Claude Loop is a skill for Claude Code, Anthropic's AI coding assistant, that automates working through a list of development tasks one at a time without quality dropping over long sessions. The core problem it addresses: in long coding sessions, AI assistants gradually lose quality as their context window fills up with prior decisions, code, and conversation history. The longer the run, the more the assistant forgets earlier choices or makes inconsistent changes. Claude Loop sidesteps this. You write a prd.json file listing the user stories you want implemented, each with acceptance criteria and a priority. For each story, the tool spawns a fresh subagent, a clean Claude Code session with no accumulated context, to implement just that one story. The subagent writes the code, runs tests, commits to git, and exits. Then the next story gets its own fresh subagent. State carries forward through three durable channels: git history, since each story produces a focused commit, a progress.txt append only log for recording learnings, and AGENTS.md files placed in source directories that Claude automatically reads when working in that folder. To use it, you install the skill by cloning it into .claude/skills/claude-loop/, create a prd.json at your project root, and tell Claude to run the loop. An orchestrator session reads the task list, manages the git branch, spawns subagents one after another, verifies their output, and archives the run once all stories pass. Parallel execution is not supported by design, since stories often touch overlapping files. The full README is longer than what was shown.
A Claude Code skill that breaks a feature list into small stories, each implemented by its own fresh AI subagent, so long coding runs do not lose quality over time.
No license information is given in the explanation, so it is not clear what uses are permitted.
Setup difficulty is rated easy, with roughly 30min to a first successful run.
Mainly developer.
This repo across BitVibe Labs
Verify against the repo before relying on details.