codeArbiter
A governance layer for agentic coding: enforced gates for tests, reviews, security and decisions, shared across Claude Code, Codex CLI and Pi
Medium risk
We rate an entry medium when the tool runs code, makes network calls or reads project files. Check what exactly it does before installing.
Why this level
- Runs hooks and checks, reads and edits project files
- Manages branches, commits and PRs under user control
Install
Manual install
/plugin marketplace add arbiterForge/codeArbiter
/plugin install ca@codearbiterAfter install approve the trust prompt and run /ca:init in the repository.
This is third-party code. Review the repository files before installing.
What it does
The plugin adds a checkable order to agentic coding: every request enters a named lane that defines the evidence it must produce, the reviewers it must dispatch and the conditions that stop it. Rules live in the repository and are shared across three hosts: ca for Claude Code, ca-codex for Codex CLI and ca-pi for Pi, while project state sits in a .codearbiter directory. codeArbiter will not write feature code before a failing test exists, will not commit on a red suite, will not guess an answer to an open question and will not push to the default branch without the user. The depth of the gates scales to the work: a prose edit takes a short lane, a sensitive feature takes the full spec, test-first, review and PR path.
Who it is for. For teams and power users who let agents do consequential work and want a fixed process with evidence.
Good fit when
- You need enforced gates for tests, reviews and security in agent work
- You want decisions, specs and an audit trail kept next to the code
- You want one process across Claude Code, Codex CLI and Pi
Not a fit when
- You want a quick sketch without a formal process and evidence
- The environment lacks the Python and Git versions the plugin requires
Example request
Fix this regression on the failing-test-first lane, then assemble a PR with reviewLimitations
It needs Python 3 on PATH and a configured git name and email. Decision lifecycle checks require Git 2.45 or newer, and Pi additionally needs Node.js 22.19 or newer. The Pi host is marked as preview. The plugin stays dormant until a repository explicitly opts in, and it does not push to the default branch or merge without the user.
How to disable. Uninstall the ca plugin via /plugin in Claude Code, ca-codex via codex plugin in Codex CLI, and remove the .codearbiter directory from the project if it is no longer needed.
Security check
- Runs hooks and checks, reads and edits project files
- Manages branches, commits and PRs under user control
README in short
The README describes codeArbiter as a shared governance layer for agentic coding on top of Claude Code, Codex CLI and Pi. Every request runs a named lane with required evidence and review, and project state with decisions, plans and audit lives in a .codearbiter directory that moves between hosts. It lists the refusals: no code before a failing test, no commit on a red suite, no guessing an open question, no push to the default branch without the user. Install is through each host's plugin marketplace, ca for Claude Code, followed by /ca:init and /ca:doctor.
FAQ
How is this different from a plain linter or CI?
The gates are built into the agent's own turn: it cannot write code before a failing test, commit on a red suite or push to the default branch without the user, and it leaves an evidence trail.
Can I move a project between hosts?
Yes. Rules and state in the .codearbiter directory are shared, so the repository moves between Claude Code, Codex CLI and Pi without losing specs, plans or decisions.
Related
A skills library that gives coding agents a development process: brainstorming, planning, TDD, subagents and code review
Skills for real engineers by Matt Pocock
Skills For Real Engineers
Small composable skills for engineering with agents: plan grilling, TDD, bug diagnosis, code review and architecture
GitHub toolkit for spec-driven development: the specify CLI adds agent commands and skills to a project, from principles to implementation
Reference MCP servers
Model Context Protocol servers
Official reference MCP servers: Filesystem, Fetch, Git, Memory, Sequential Thinking, Time and Everything