HN Showcase
Back to browse
GitHub Repository

Automatically review Claude Code plans using external AI CLIs

57 starsJavaScript

Claude-plan-reviewer – Rival AI reviews Claude Code's plans

by yuu1ch13·Feb 28, 2026·2 points·0 comments
Visit ProjectView on HN

AI Analysis

●●SolidBig BrainNiche Gem

Adversarial plan review via hook injection—clever constraint on existing tools.

Strengths
  • •PreToolUse hook mechanism is elegant: non-invasive interception without patching Claude Code.
  • •Multi-model approach (Claude + Codex/Gemini) catches blind spots single-model approaches miss.
  • •Minimal footprint (400 lines, zero dependencies) with clean setup/teardown.
Weaknesses
  • •Narrow use case: only relevant to Claude Code plan mode; no broader applicability.
  • •Assumes Codex/Gemini CLI installed and configured—high dependency chain for small benefit.
Target Audience

Claude Code power users seeking second-opinion code review before implementation

Similar To

Claude Code · Codex CLI · Gemini CLI

Post Description

Claude Code has a plan mode where it writes implementation plans before coding. I built a tool that intercepts those plans and sends them to a competing AI (Codex CLI or Gemini CLI) for adversarial review before Claude is allowed to proceed.

When Claude calls `ExitPlanMode`, a `PreToolUse` hook fires, sends the plan to the reviewer, and returns `{ permissionDecision: "deny", permissionDecisionReason: "<feedback>" }`. Claude receives the denial with the feedback, revises the plan, and tries again. After a configurable number of rounds (default: 2), it passes through. At the end you get a diff of the original vs. final plan.

Different models have different blind spots. The value is in the second perspective, not in one model being "better."

``` npm install -g claude-plan-reviewer claude-plan-reviewer setup ```

Requires Node.js 18+ and Codex CLI or Gemini CLI. ~400 lines of JS, zero dependencies. MIT licensed.

https://github.com/yuuichieguchi/claude-plan-reviewer

Similar Projects

Developer Tools●●Solid

Claude-relais – A plan/build/judge loop mixing Claude with Cursor

Splitting thinking and typing by having Claude plan and judge while Cursor 'types' is a neat, pragmatic trade-off — you keep expensive reasoning where it matters and push bulk work to cheap, fast agents. The repo actually enforces safety: destructive command gates, scoped file access, install and preflight scripts, and a PLAN→BUILD→JUDGE contract rather than a toy demo. The downside is obvious: it's niche (requires Claude + Cursor CLIs and accounts) and adds orchestration complexity compared with single-model flows.

Big BrainNiche GemShip It
crog
104mo ago