---
title: "correctionguy"
description: "Agents often drift and waste time on the wrong work. Correction Guy watches from the sidelines and steers them back on track."
canonical: https://agentpluginsdirectory.com/plugins/correctionguy
last-updated: 2026-09-21
---

# correctionguy
Agents often drift and waste time on the wrong work. Correction Guy watches from the sidelines and steers them back on track.
- Slug: correctionguy
- Publisher: correctionguy
- Repository: https://github.com/correctionguy/correctionguy
- Manifest: plugin.json
- Version: 3.26.0
- License: MIT
- Category (editorial): agent-tooling
- Skills: 3 (actually, correctionguy, setup)
- MCP servers: 0
- Stars: 5
- Repository created: 2026-05-25
- Repository last pushed: 2026-09-21
- Publisher type: Organization
- Listing: https://agentpluginsdirectory.com/plugins/correctionguy
- Schema: https://agent-plugins.org/schemas/1.0.0/plugin.schema.json

## What correctionguy does, in the publisher's words

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. Packaged as an Agent Plugins 1.0.0 plugin (portable skills in skills/ behind root plugin.json) with host adapters for 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.

From the project README, punctuation lightly normalized. Full text: https://raw.githubusercontent.com/correctionguy/correctionguy/HEAD/README.md

## Skills

- actually: User says Correction Guy rooted on old, outdated, or wrong conventions; record the override in.memory and apply it from now on.
- correctionguy: Work careful: understand, source claims, run it, ask candid review.
- setup: Set up Correction Guy in a repo: lay out.memory, merge the agent's native memory behind a symlink, mine every past conversation into memory with fanned-out subagents.

Descriptions come from the frontmatter of each SKILL.md, punctuation lightly normalized.