review-forge

Review Forge is a skill plugin for AI coding agents that brings structure and accountability to the code review process. Instead of getting a single AI pass on your code and hoping it catches everything, Review Forge runs multiple review passes from different perspectives, then compiles the findings into one checklist you can inspect and approve before any changes are made.

The core workflow has four steps. First, you run a review, which asks the AI to look at your local changes, a branch, a pull request, or a specific range of commits and produce a review file. Second, you synthesize, which merges multiple review files from different models or perspectives into a single summary. Third, you fix, which only touches the items you have checked off in that summary, then runs your tests. Fourth, you verify, which independently confirms that the fixes actually worked.

That human approval step is a deliberate design choice. The checklist acts as a boundary: the tool will not implement a fix unless you have explicitly ticked the box. This keeps you in control even when the AI is doing most of the analysis work.

All the files the workflow creates, including review notes, summaries, fix plans, and verification logs, land under a code_review/ folder inside your project. Each feature you review gets its own subfolder, so different code changes stay organized and traceable. If you do not want these files committed to version control, the README suggests adding code_review/ to your .gitignore.

You install Review Forge using the Skills CLI tool with a short npx command, then invoke it from any compatible agent by referencing $review-forge in your prompt. It works with agents like Codex, Cursor, and OpenCode. Generated reports can follow your prompt language automatically, while internal status values stay in English for consistency across tools.