coreviewlight

Runs a two-agent review loop where the reviewer steers and the implementer writes all code.

Updated Jul 18, 2026
One-click install
npx skills add https://github.com/quentinlemarie/revshell-coreview --skill coreviewlight-quentinlemarie
Or copy as Structured Prompt for Agent
Please help me install this Agent Skill.
Skill: coreviewlight
Source: https://github.com/quentinlemarie/revshell-coreview/tree/main/coreviewlight
Command: npx skills add https://github.com/quentinlemarie/revshell-coreview --skill coreviewlight-quentinlemarie

SYSTEM DOCUMENTATION & REQUIREMENTS

What problem does it solve? Coordinating a second AI agent to review code often produces redundant edits or conflicting changes. This Skill defines a lightweight cross-model review loop where the REVIEWER only sends findings and instructions while the IMPLEMENTER writes every line of code, keeping responsibilities cleanly separated. ## Core Features & Use Cases - Steer-only reviewer mode: The reviewer sends findings and instructions but never edits code; the implementer performs the initial build and every fix. - Shared coreview mechanics: Marker files, background watchers, and the autonomous finish sequence (curated commits, one no-ff merge, one docs(memory) commit) are inherited from the consolidated coreview skill. - Use Case: You want Claude to implement a feature while Codex reviews each round of the working tree without committing, looping until the reviewer says it is ready to ship. ## Quick Start Ask the agent to run coreviewlight so it implements the change while a second agent reviews the working tree and steers with findings until ready to ship.

Frequently Asked Questions about coreviewlight

High-intent search queries and answers about installing and using this skill.

FAQPage Schema
How do I run a lightweight two-agent code review loop?

Load the coreview skill and follow its coreviewlight mode section. Do the upfront setup by picking the implementer, reviewer, commit target, and temp branch, then let the reviewer steer with findings while the implementer writes all code until ready to ship.

What is the difference between coreview and coreviewlight?

In coreview mode both agents may correct each other's work, while in coreviewlight the reviewer only steers by sending findings and instructions and the implementer writes all code. coreviewlight is a mode of the consolidated coreview skill, not a separate implementation.

Which CLI coding agents support this review workflow?

The skill is shared by Claude Code, Codex CLI, and Gemini CLI through symlinks to the coreview directory in each runtime's skills folder. Each agent follows the same SKILL.md contract, so behavior is identical across runtimes.

Do review rounds create git commits?

No, review rounds never create commits; the working tree is the review artifact with the diff base at git merge-base. The finish sequence creates only curated approved commits, one no-ff merge, and a single docs(memory) commit.

When should I use coreviewlight instead of full coreview?

Use coreviewlight when you want one agent to fix while the other reviews only, reducing redundant edits and handoff text. Choose full coreview when both sides should be allowed to correct each other's work directly.