correctionguy

A second pair of eyes for your coding agent. Correction Guy watches from the sidelines and steers the work back on track — it never touches the work itself. Works in Claude Code, Cursor, and Pi.

An agent grading its own work is the weakest check there is. It is biased toward declaring victory: it papers over stubs, asserts things it never verified, asks permission instead of delivering, and forgets what it just learned. A sharper self-prompt doesn't fix this — the agent is still inside its own story.

So Correction Guy hands the review to someone else. A different model (Codex, through the Codex SDK — no other plugin required) reads the session as an outsider, with no stake in the agent's narrative, and calls out what the agent talked itself past. It is deliberately narrow: it flags real failures — wrong scope, stubs, unsourced claims, ignored test failures, abandoned work, drift, a to-do list or memory gone stale — and stays quiet on taste like naming, formatting, and structure. It steps in at the three moments that matter: as work begins, while it is underway, and the instant the agent tries to call it done.

What it does

• Session prelude — sets expectations up front: restate the task, check memory, keep the to-do list and session title honest, verify third-party behavior, finish the work, run the code.
• Live monitor — every few tool batches, an outside pass catches drift, bad assumptions, stale to-dos, and contradicted memory before they compound.
• Stop check — when the agent tries to stop, the reviewer can block a premature "done" — a stub, abandoned work, ignored feedback, an unrun build — and feed the correction back.
• On-demand — /correctionguy restates the discipline whenever you want it.

The review prompts, and the corrections that come back, are written in compressed "caveman" style to save tokens.

Memory

Memory belongs with the code it describes. Correction Guy keeps the agent's memory in a project-local .memory folder, so it lives alongside the work instead of in a global store, and it treats facts you state about the project as sourced — recorded, trusted, and never second-guessed for lacking a citation. The reviewer flags when a session contradicts what is written there, or learns something worth keeping and never writes it down.

Requirements

• Bun — for the Claude Code and Cursor hooks
• Codex authenticated — run codex login once, or set an API key.
• Claude Code, Cursor, or Pi

Install

Claude Code

/plugin marketplace add correctionguy/correctionguy

/plugin install correctionguy@correctionguy

Cursor — install from the Cursor Marketplace, or test locally from a checkout.

Pi

pi install git:github.com/correctionguy/correctionguy

Or try it for a single run with pi -e git:github.com/correctionguy/correctionguy.

Configuration

Everything is tuned through CORRECTIONGUY_* environment variables — model, reasoning effort, review cadence, and a CORRECTIONGUY_YOLO switch that drops the read-only sandbox for the reviews. Defaults live in scripts/codex.ts. You're going to ask an agent to do it, so I'm not going to write it out here 😏

Licensed under MIT.