Why are reviewers isolated in fresh instances?

Isolation prevents bias and anchoring, ensuring each review cycle is a genuine, independent check of the plan's current state without influence from previous findings.

Why are there three different reviewers?

Each reviewer focuses on a specific risk: Feasibility (can it actually be executed?), Completeness (is every requirement covered?), and Scope & Alignment (is it exactly what the user asked for?).

What happens if a reviewer gives a 'FAIL' verdict?

The planner receives the consolidated feedback, revises the plan to address the issues, and then submits the new draft to a completely fresh set of isolated reviewers.

Does this work for simple, single-file bug fixes?

To maintain efficiency, the gate is designed to trigger only for more complex plans involving at least 2 work units or changes to 3 or more files.

What is the Plan Review Gate?

It is an automated skill for Claude Code that subjects implementation plans to three independent, adversarial reviews before they are presented to the user.

Plan Review Gate

Name: Plan Review Gate
Author: jpeggdev

byjpeggdev

0•

Security & Testing

Validates AI implementation plans through three independent adversarial reviewers to ensure technical feasibility, completeness, and scope alignment.

The Plan Review Gate serves as a rigorous quality assurance layer for Claude Code, preventing flawed or unverified implementation plans from reaching the user. It automatically triggers after a plan is drafted, spawning three isolated sub-agents—Feasibility, Completeness, and Scope & Alignment—to scrutinize the proposal in parallel. By requiring a unanimous 'PASS' from these adversarial reviewers, the skill ensures that proposed changes are technically sound, meet all user requirements, and avoid unnecessary scope creep. It supports iterative refinement, automatically incorporating feedback and re-reviewing until the plan reaches production-ready quality.

Key Features

01Read-only codebase verification to prevent hallucinations and path errors

02Isolated sub-agent execution to eliminate anchoring bias and context leakage

030 GitHub stars

04Automated iterative refinement that incorporates reviewer feedback into revised plans

05Triple-agent adversarial review architecture for maximum plan reliability

06Configurable threshold logic that activates for complex multi-file changes

Use Cases

01Preventing AI scope creep or over-engineering in response to simple requests

02Ensuring new feature implementations cover all edge cases and verification steps

03Validating complex multi-file refactors before any code is actually modified

What are Skills?·How to Install

Install with 🐟 Skill.Fish

npx skillfish add jpeggdev/buildwithjpegg plan-review-gate

For use in Claude.ai and ChatGPT

Key Features

01Read-only codebase verification to prevent hallucinations and path errors

02Isolated sub-agent execution to eliminate anchoring bias and context leakage

030 GitHub stars

04Automated iterative refinement that incorporates reviewer feedback into revised plans

05Triple-agent adversarial review architecture for maximum plan reliability

06Configurable threshold logic that activates for complex multi-file changes

Use Cases

01Preventing AI scope creep or over-engineering in response to simple requests

02Ensuring new feature implementations cover all edge cases and verification steps

03Validating complex multi-file refactors before any code is actually modified