graph-engineer skill
A skill that runs an eight-step loop: Claude orchestrates and arbitrates while Codex writes, critiques and fixes the code.
- 106 ★ on the repository
- Last commit
- MIT
Extended access
Why this level
- Codex write mode edits project files directly.
- Code and context go to a second model through an OpenAI account.
Install
- Works in
In your terminal, with SkillFoxx CLI
npx skillfoxx add skills/graph-engineerDetects the agents on your machine, shows the access level and pins the version.
Other ways to install
Access and checks
How we reviewExtended accessThe tool runs code or makes network calls and only reads external data.
- Codex write mode edits project files directly.
- Code and context go to a second model through an OpenAI account.
This is third-party code. Review the repository files before installing.
What it does
The `graph-engineer` skill splits roles between two models. Claude Code writes the contract, drives the loop and rules on disputed findings, while Codex, through the `codex-plugin-cc` plugin, implements the code, critiques it read-only and applies fixes. The loop has eight nodes: pre-flight, spec, implementation, a mechanical gate (lint, formatting, types, build), critique, debate, refactor and test verification. Return edges rerun the critique until the stop condition holds. State is saved in `PROJECT_CONTEXT.md`, so the loop can resume after a session restart.
Who it is for. Developers with both Claude Code and Codex subscriptions who want a second model to independently review their code.
Good fit when
- An existing module should get an adversarial review without write access.
- A new feature needs a contract before the first commit.
- The author worries Codex would grade its own fixes without an arbiter.
- The work spans several sessions and loop state has to be saved.
Not a fit when
- The edit is cosmetic and touches no other file's contract.
- The working tree is dirty or it is `main` with uncommitted changes.
- There is no Codex subscription and you do not want to switch roles to Claude.
Example request
Run a graph-engineer review-only cycle on `src/billing` with no writes and show accepted and rejected findings.Limitations
The skill needs the `openai/codex-plugin-cc` plugin, a Codex subscription and the built-in `/goal` command in Claude Code. It was verified on plugin version 1.0.6 and in a single repository; other versions are unconfirmed. Write mode edits files directly, so run it on a clean branch. Codex calls are billed separately, and the author did not benchmark Claude token savings. For short edits the loop is excessive: the author suggests review-only mode or reading the code yourself.
How to disable. Remove `~/.claude/skills/graph-engineer` and, if needed, the Codex plugin via `/plugin`. Delete `PROJECT_CONTEXT.md` in your projects by hand.
FAQ
How does review-only differ from the full loop?
Review-only makes one Codex call for critique and writes no files. The full loop adds spec, implementation and verification.
Is Codex required?
By default yes, through the `codex-plugin-cc` plugin. `backend: claude` modes run without it but lose model diversity.
What is the mechanical gate?
Lint, formatting, types and build. A failure returns work to the writer, while tests run at a separate step.
Where is state kept?
In `PROJECT_CONTEXT.md` inside the project. The loop resumes from there after a restart.
Is it safe to run on `main`?
No. Pre-flight requires a clean tree and a separate branch, and Codex write mode edits files directly.
Discussion
No questions or reviews yet.
Related
Skillslencx skills
lencx-skills
Two skills for Claude Code and Codex: keel works through architecture questions, coding-protocol keeps code edits inside the task
A Claude Code plugin: Claude does the task, then a stop hook starts an independent Codex review and Claude fixes the findings in the same session
A Claude Code skill that delegates tasks to OpenAI's Codex CLI for code analysis, review and edits
A ready harness for Claude Code and Codex: spec-driven development, TDD, memory between sessions, quality gates and human checkpoints
A Claude Code skill that drives Codex over JSON-RPC: tasks, code reviews, parallel research and messaging between two agents.
A dev process built from composable skills for Claude Code and Codex: plan, implement, review, test, commit and PR
Comparison with similar tools
Cards that share the main search query, and how they differ.
| Tool | Official | Repository stars | Main difference |
|---|---|---|---|
| graph-engineer (this card) | no | 106 | |
| cross-review | no | 167 | one model checks another; graph-engineer adds a contract, a mechanical gate and loop return edges |
| CodexPotter | no | 636 | another Codex-centred skill; graph-engineer splits Claude and Codex roles across eight nodes |
README in short
Implements two Anthropic patterns: orchestrator-workers and evaluator-optimizer. The Limitations / Risks section lists claims that are not verified.