{
  "markdown": "# claude-review-loop\n\nAutomated 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.\n\n## How it works\n\n1. **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\n2. **Triage** — Presents each finding interactively with the reviewer's suggested fixes as selectable options. Pick a fix, defer, dismiss, or type your own approach\n3. **Plan** — Enters plan mode to build a remediation plan, explicitly dispatching to an agent team\n4. **Fix** — Creates a fixer agent team to remediate approved findings in parallel across files\n5. **Loop** — Automatically re-runs the review to verify fixes and catch regressions. Repeats until the review finds zero issues or you defer/dismiss everything\n\n## Installation\n\n```bash\nclaude /plugin install claude-review-loop\n```\n\nOr install from source:\n\n```bash\nclaude --plugin-dir /path/to/claude-review-loop\n```\n\n## Usage\n\nFrom any branch with changes relative to `main`/`master`:\n\n```\n/review-loop\n```\n\nThe plugin will detect your changes, run the review, and guide you through the rest interactively.\n\n## Prerequisites\n\n1. **[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.\n\n```bash\n/plugin install pr-review-toolkit\n```\n\n2. Agent teams must be enabled in your Claude Code settings (`~/.claude/settings.json`):\n\n```json\n{\n  \"env\": {\n    \"CLAUDE_CODE_EXPERIMENTAL_AGENT_TEAMS\": \"1\"\n  }\n}\n```\n\n## Required permissions\n\nThe plugin needs these tool permissions. Add to your `~/.claude/settings.json` or project `.claude/settings.json`:\n\n```json\n{\n  \"permissions\": {\n    \"allow\": [\n      \"Bash(git *)\",\n      \"Bash(date *)\",\n      \"Read(*)\",\n      \"Write(*)\",\n      \"Edit(*)\",\n      \"Glob(*)\"\n    ]\n  }\n}\n```\n\n## Plugin structure\n\n```\nclaude-review-loop/\n├── .claude-plugin/\n│   └── plugin.json          # Plugin manifest\n├── hooks/\n│   ├── hooks.json           # Stop hook registration\n│   └── stop-hook.sh         # Enforces loop after fixes\n├── skills/\n│   └── review-loop/\n│       └── SKILL.md         # Main orchestrator skill\n├── templates/\n│   └── review-report.md     # Markdown report template\n└── .gitignore\n```\n\nThe 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.\n\n## Output files\n\nThe plugin creates these files in your project root (all gitignored):\n\n- `.review-TIMESTAMP.md` — Human-readable review report with findings and action checkboxes\n- `.review-state.json` — Session state tracking cycles, counts, and progress\n\n## License\n\nMIT\n",
  "bytes": 3391,
  "sha": "6a0efdfa8e0fe2d8fb38d599582f18e05905da9b7d410faf7d8b26826d9a44e3",
  "repo_slug": "andytango/claude-review-loop",
  "fonte": "repo",
  "truncated": false,
  "api": "https://agentalog.com/api/listings/plg_andytango_claude_review_loop_claude_revi_07d60623/readme"
}