quentinlemarie
Community@quentinlemarie
Cross-model code review loops where Claude, Codex, and Gemini share one on-disk plan, with implementer and reviewer roles running autonomously to commit.
Agent Skills by quentinlemarie
Showing 3 vetted skills indexed across 1 GitHub repositories.
coreview
Orchestrates cross-model plan and code review loops between CLI coding agents.
coreviewlight
Runs a two-agent review loop where the reviewer steers and the implementer writes all code.
revshell
Gates code changes through a blocking subprocess reviewer before they land.
Frequently Asked Questions About quentinlemarie
FAQPage SchemaWhat tasks can I run with quentinlemarie's coreview skills?▼
You can run a shared-plan implement-and-review loop across Claude, Codex, and Gemini. Modes include coreview (both sides may write), coreviewlight (reviewer only steers), and shellrev/revshell (single-session in-shell review of a diff before it lands).
Who are these skills designed for?▼
Developers who already work with multiple coding models and want a second-model pass on plans or committed code. It suits users asking to 'have Codex review this', iterate with another model, or monitor committed code without manual handoffs.
How does a coreview session run end-to-end?▼
The user picks the mode, the implementer, and the reviewer upfront. The implementer writes the plan then code; the reviewer audits; the loop finishes autonomously with a commit on dev/test, no push, and saved learnings, stopping only for user arbitration.
How does revshell differ from the full coreview loop?▼
Revshell is a single-session review gate: the host implements a change, then drives a reviewer (Codex via scripts/codex-review.sh or a Claude review subagent) as a blocking subprocess. Approval lands curated commits merged to dev, deletes the temp branch, and ends with one memory commit.
What are the prerequisites for using these skills?▼
You need access to the participating models (Claude, Codex, and optionally Gemini) and a git repository with a dev branch. Revshell additionally relies on scripts/codex-review.sh for the Codex backend; no separate watcher or second human session is required.