skills.rest
quentinlemarie
quentinlemarieCOMMUNITY
Explore all skills by this author →
GitHub

coreview

Orchestrates cross-model plan and code review loops between CLI coding agents.

Updated Jul 18, 2026
Search Tags:#code review#git workflow#claude code#multi-agent#codex cli#plan review#review loop
One-click install
npx skills add https://github.com/quentinlemarie/revshell-coreview --skill coreview-quentinlemarie
Or copy as Structured Prompt for Agent▼
Please help me install this Agent Skill.
Skill: coreview
Source: https://github.com/quentinlemarie/revshell-coreview/tree/main/coreview
Command: npx skills add https://github.com/quentinlemarie/revshell-coreview --skill coreview-quentinlemarie

SYSTEM DOCUMENTATION & REQUIREMENTS

💡 This Skill includes scripts (resource) components.

What problem does it solve? Getting a second AI model to rigorously review a plan or implementation usually means ad-hoc copy-paste between sessions with no shared protocol, no convergence criteria, and no clean commit discipline. This Skill gives Claude Code, Codex CLI, and Gemini CLI one identical review contract so two agents iterate over a single on-disk plan file until the reviewer declares it ready to ship. ## Core Features & Use Cases - Two-session review loop: An IMPLEMENTER and a REVIEWER exchange turns via footer markers in one plan file, with background watcher scripts detecting each handoff and terminator phrases ("ready for implementation", "ready to ship"). - Three modes: regular coreview (both sides may correct each other), coreviewlight (reviewer only steers), and shellrev (single session delegating to the revshell skill via blocking subprocess calls). - Guard-railed finish sequence: finish.sh merges only curated commits into dev/test (refusing wip/checkpoint/review-round subjects, never pushing), and save-learnings.sh writes exactly one docs(memory) commit with the session's goal, resolution, and learnings. - Use Case: After drafting a refactor plan, ask for a cross-model review; the reviewer agent edits the plan directly, then reviews the implemented diff round by round until it converges, and the host lands clean commits without any manual git choreography. ## Quick Start Run a coreview loop where Claude implements and Codex reviews my current plan, landing the approved work on the dev branch.

Frequently Asked Questions about coreview

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

FAQPage Schema
How do I run a cross-model code review between Claude and Codex?▼

Invoke the coreview skill and state the mode, implementer, reviewer, and commit target upfront. The implementer writes the plan and code while the reviewer appends verdict markers to the shared plan file; watcher scripts detect each handoff until the reviewer writes ready to ship.

What is the difference between coreview, coreviewlight, and shellrev modes?▼

In coreview both agents may edit the plan and code to correct each other. In coreviewlight the reviewer only steers with findings and the implementer writes all code. In shellrev a single session drives the reviewer as a blocking subprocess via the revshell skill, with no watcher or markers.

Does coreview work with Gemini CLI and Codex CLI?▼

Yes, the same SKILL.md is symlinked into each runtime's skills directory so the protocol is identical everywhere. Only the watcher mechanism and agent dispatch differ per runtime, mapped in scripts/TOOLS.md.

Why does the coreview watcher stall or miss the counterpart's turn?▼

Stalls usually come from editing an earlier marker block in place instead of appending at the footer, splitting markers across multiple files the watcher is not armed on, or a hand-typed timestamp older than the counterpart's last verdict. Always use scripts/append-marker.sh and keep markers in one canonical plan file.

Can coreview push or merge to main automatically?▼

No. The finish.sh script only merges the temp branch into dev or test, refuses review-process commit subjects, and never pushes. Push, promotion, and further merges are always left for the user to decide from the printed commit list.

When should I not use the coreview review loop?▼

Skip it for trivial one-file changes where a full two-agent loop adds overhead without benefit. It is designed for plan-shaped tasks where a second-model pass on the plan and implementation materially reduces risk.

Related Skills

Explore More Software Engineering Skills→
githubnextgithubnext
5.1k

git-workflow

Implements a dev-first branching model with worktree-based parallel issue workflows.

Official
Intermediate
MINT-IAMINT-IA

mint-commit

Enforce standardized git commit and promotion workflows with branch and merge rules.

Official
Advanced
human-centric-engineeringhuman-centric-engineering

pr-gates

Runs pre-merge validation, coverage, security, and code review gates on a branch and fixes findings until clean.

Official
Advanced
Disaster-TerminatorDisaster-Terminator
5

pr-review

Automate pull request reviews with AI bots and archive learnings.

Community
Advanced