claude-review-loop
Automated code review and swarm remediation for large changesets. Spawns adversarial reviewer agents, presents findings interactively, then
Open source Repository Open in the app JSON README (API)
About
Automated code review and swarm remediation for large changesets. Spawns adversarial reviewer agents, presents findings interactively, then dispatches fixer agents to remediate approved issues.
Details
- Kind
- Plugins
- Topic
- Developer tools
- Publisher
- andytango
- Origin
- marketplace
- Category
- ferramentas
- Last push
- 2026-03-14T07:59:40Z
- Repository state
- ativo
- Language
- Shell
- Added
- 2026-08-30 01:48:58
- Updated
- 2026-08-30 01:48:58
- Origin id
andytango/claude-review-loop/claude-review-loop
README
# claude-review-loop
Automated code review and swarm remediation plugin for Claude Code. Spawns adversarial reviewer subagents, presents findings interactively with selectable fix options, then dispatches an [agent team](https://docs.anthropic.com/en/docs/claude-code) to remediate approved issues. Loops until clean.
## How it works
1. **Review** — Spawns 4 specialist agents from the [PR Review Toolkit](https://github.com/anthropics/claude-code-plugins) in parallel (code-reviewer, silent-failure-hunter, pr-test-analyzer, type-design-analyzer) to examine your branch diff with fresh eyes
2. **Triage** — Presents each finding interactively with the reviewer's suggested fixes as selectable options. Pick a fix, defer, dismiss, or type your own approach
3. **Plan** — Enters plan mode to build a remediation plan, explicitly dispatching to an agent team
4. **Fix** — Creates a fixer agent team to remediate approved findings in parallel across files
5. **Loop** — Automatically re-runs the review to verify fixes and catch regressions. Repeats until the review finds zero issues or you defer/dismiss everything
## Installation
```bash
claude /plugin install claude-review-loop
```
Or install from source:
```bash
claude --plugin-dir /path/to/claude-review-loop
```
## Usage
From any branch with changes relative to `main`/`master`:
```
/review-loop
```
The plugin will detect your changes, run the review, and guide you through the rest interactively.
## Prerequisites
1. **[PR Review Toolkit](https://github.com/anthropics/claude-code-plugins)** plugin must be installed — provides the 4 specialist review agents (code-reviewer, silent-failure-hunter, pr-test-analyzer, type-design-analyzer) with confidence scoring and structured output. The plugin will check for this on startup and tell you how to install it if missing.
```bash
/plugin install pr-review-toolkit
```
2. Agent teams must be enabled in your Claude Code settings (`~/.claude/settings.json`):
```json
{
"env": {
"CLAUDE_CODE_EXPERIMENTAL_AGENT_TEAMS": "1"
}
}
```
## Required permissions
The plugin needs these tool permissions. Add to your `~/.claude/settings.json` or project `.claude/settings.json`:
```json
{
"permissions": {
"allow": [
"Bash(git *)",
"Bash(date *)",
"Read(*)",
"Write(*)",
"Edit(*)",
"Glob(*)"
]
}
}
```
## Plugin structure
```
claude-review-loop/
├── .claude-plugin/
│ └── plugin.json # Plugin manifest
├── hooks/
│ ├── hooks.json # Stop hook registration
│ └── stop-hook.sh # Enforces loop after fixes
├── skills/
│ └── review-loop/
│ └── SKILL.md # Main orchestrator skill
├── templates/
│ └── review-report.md # Markdown report template
└── .gitignore
```
The orchestrator uses a hybrid approach: **PR Review Toolkit agents** for review (with confidence scoring and structured output) and **agent teams** for fixes (coordinated parallel writes with task tracking). A stop hook enforces the loop — if the agent tries to exit after fixes, the hook blocks and re-injects the review prompt.
## Output files
The plugin creates these files in your project root (all gitignored):
- `.review-TIMESTAMP.md` — Human-readable review report with findings and action checkboxes
- `.review-state.json` — Session state tracking cycles, counts, and progress
## License
MIT