adamsreview – better multi-agent PR reviews for Claude Code

adamsreview is a Claude Code plugin that enhances PR reviews by using multi-stage, parallel sub-agents, persistent JSON state, and optional ensemble review. It aims to catch more bugs with fewer false positives than existing AI review tools. The tool includes commands for review, fix dispatch, and an interactive walkthrough.

How you'd build it

1. 1Develop a core agent orchestration engine in Python, capable of managing sequential and parallel sub-agent calls for code review tasks, saving intermediate states to a local file system.
2. 2Implement a plugin architecture for Claude Code, defining specific slash commands (e.g., /review, /fix, /walkthrough) that interface with the core engine.
3. 3Integrate with Anthropic's Claude API, specifically leveraging 'AskUserQuestion' for interactive review and Opus for re-reviewing fixes. Ensure efficient token usage and context management.
4. 4Build a 'fix' command that dispatches dedicated agents for bug fixes, re-reviews the changes, and manages reverts for regressions. This requires robust diffing and git integration.
5. 5Design and implement a state management system using local JSON files to maintain review context across different stages and commands, with clear update scripts.
6. 6Create a robust testing suite for various code scenarios, focusing on false positive reduction and ensuring actual bug detection.

Risks & moats

• The effectiveness is highly dependent on the underlying LLM's (Claude Opus) capabilities, which can change and may introduce regressions. Constant fine-tuning of prompts will be needed.
• Managing persistent state with local JSON files can become complex and prone to conflicts or inconsistencies in a multi-user or concurrent review environment.
• It requires users to be heavily invested in the Claude Code ecosystem, limiting its market to a specific developer segment.
• Competing with built-in features from Claude Code, CodeRabbit, and GitHub Copilot could be challenging as these platforms rapidly evolve their offerings.

Reach out to Claude Code users and AI engineers discussing adamsreview to offer AI usage monitoring, agent evaluation tools, or frameworks for building similar AI-powered developer tools.

Original context

I built adamsreview, a Claude Code plugin that runs deeper, multi-stage PR reviews using parallel sub-agents, validation passes, persistent JSON state, and optional ensemble review via Codex CLI and PR bot comments. On my own PRs, it has been catching dramatically more real bugs than Claude’s built-in /review, /ultrareview, CodeRabbit, Greptile, and Codex’s built-in review, while producing fewer false positives. adamsreview is six Claude Code slash commands packaged as a plugin: review, codex-review, add, promote, walkthrough, and fix. I modeled it after the built-in /review command and extended it meaningfully. You can clear context between review stages because state is stored in JSON artifacts on disk, with built-in scripts for keeping it updated. The walkthrough command uses Claude’s AskUserQuestion feature to walk you through uncertain findings or items needing human review one by one. Then, the fix command dispatches per-fix-group agents and re-reviews the work with Opus, reverting any regressions before committing survivors. It runs against your regular Claude Code subscription (Max plan recommended), unlike /ultrareview, which charges against your Extra

You may also want to look at

• 82
  GPT-4-powered web searches for developers

  🛠 buildHacker News· Data & AI
  →
• 80
  cmux - Ghostty-based terminal with vertical tabs and notifications

  🛠 buildHacker News· Data & AI
  →
• 78
  Ts-SSH – SSH over Tailscale without running the daemon

  🛠 buildHacker News· Data & AI
  →
• 75