CodeMoot

A second opinion for AI-generated code. Claude Code + Codex CLI collaboration platform that brings debate, review, autofix, and consensus patterns to your development workflow.

CodeMoot bridges Claude and GPT so they work as partners — one plans, the other reviews, and together they build better code than either could alone.

Quick Start

# Install globally
npm install -g @codemoot/cli

# One-command setup: verifies codex, creates config, runs first review
codemoot start

# Or step by step:
codemoot doctor           # check prerequisites
codemoot init             # create .cowork.yml
codemoot review src/      # review code with GPT
codemoot fix src/         # autofix loop: review → fix → re-review

# Debate architecture with GPT
codemoot debate start "Should we use REST or GraphQL?"

# Ship with confidence
codemoot shipit --profile safe

Prerequisites

• Node.js >= 20
• Codex CLI installed (npm install -g @openai/codex)
• ChatGPT subscription (Codex CLI uses your existing subscription — $0 API cost)

Commands

Getting Started

Command	Description
codemoot start	First-run concierge: verify codex, init config, run quick review
codemoot doctor	Preflight diagnostics: check codex, config, database, git, node
codemoot init	Initialize CodeMoot in current project

Core Workflows

Command	Description
codemoot review <file>	Code review via GPT with structured findings
codemoot review --prompt "..."	Freeform review — GPT explores codebase via tools
codemoot review --diff HEAD~3..HEAD	Review git changes
codemoot review --preset security-audit	Use named preset (5 built-in)
codemoot fix <file>	Autofix loop: review → apply fixes → re-review
codemoot cleanup [path]	Scan for unused deps, dead code, duplicates, hardcoded values
codemoot plan <task>	Generate plan via architect + reviewer loop
codemoot run <task>	Full plan-review-implement cycle

Multi-Model Debate

Command	Description
codemoot debate start <topic>	Start a Claude vs GPT debate (--timeout sets default)
codemoot debate turn <id> <prompt>	Send next prompt (--output, --force, --quiet, --response-cap)
codemoot debate next <id>	Auto-continue debate (--quiet for programmatic use)
codemoot debate status <id>	Show debate progress
codemoot debate list	List all debates
codemoot debate history <id>	Full message history (--output <file> for untruncated export)
codemoot debate complete <id>	Mark debate as done

Automation

Command	Description
codemoot shipit [--profile fast|safe|full]	Composite workflow: lint → test → review → cleanup → commit
codemoot watch	Watch files, auto-enqueue reviews on save
codemoot build start <task>	Automated build loop with GPT review

Background Jobs

Command	Description
codemoot review --background	Enqueue review, return immediately
codemoot jobs list	List background jobs
codemoot jobs status <id>	Job details with logs
codemoot jobs logs <id>	Full job log output
codemoot jobs cancel <id>	Cancel a job
codemoot jobs retry <id>	Retry a failed job

Session Management

Command	Description
codemoot session start	Start new persistent GPT session
codemoot session current	Show active session with token usage
codemoot session list	List all sessions
codemoot session close <id>	Close a session

Observability

Command	Description
codemoot cost	Token usage dashboard (by command, by day)
codemoot events --follow	Stream events as JSONL (for editors/CI)

Review Presets

Preset	Focus	Timeout	Use Case
security-audit	Injection, auth, secrets	1200s	Pre-deploy security scan
performance	N+1, memory, blocking	900s	Performance optimization
quick-scan	Top bugs only	240s	Quick sanity check
pre-commit	Changed code blockers	180s	Git pre-commit hook
api-review	Contracts, versioning	900s	API design review

Shipit Profiles

Profile	Steps	Use Case
fast	review	Quick check before push
safe	lint → test → review → cleanup	Default — catches most issues
full	lint → test → review → cleanup → commit	Full pipeline with auto-commit

Policy Engine

CodeMoot includes a built-in policy engine that gates actions:

• Block on CRITICAL: Any critical finding blocks the commit
• Warn on NEEDS_REVISION: Review verdict triggers a warning
• Custom rules via predicate-based engine (enforce/warn modes)

Architecture

TypeScript monorepo with 3 packages:

Package	Description	Status
@codemoot/core	Orchestration engine, memory, policy, caching	Stable
@codemoot/cli	Command-line interface (15 commands)	Stable
@codemoot/mcp-server	MCP server (5 tools for IDE integration)	Experimental

How It Works

1. Claude Code is your primary AI — it plans, writes code, manages your project
2. Codex CLI (GPT) acts as reviewer/critic — it reads your codebase, finds bugs, suggests fixes
3. CodeMoot bridges them — structured prompts, session persistence, token tracking, policy gates

All GPT calls happen via Codex CLI using your ChatGPT subscription — $0 API cost.

Configuration

.cowork.yml in your project root:

models:
  codex-architect:
    provider: openai
    model: gpt-5.3-codex
    providerMode: cli
  codex-reviewer:
    provider: openai
    model: gpt-5.3-codex
    providerMode: cli

roles:
  architect:
    model: codex-architect
  reviewer:
    model: codex-reviewer

workflow: plan-review-implement
mode: autonomous

.codemootignore

Exclude files from review/cleanup/watch (gitignore syntax):

node_modules
dist
*.db
.env

MCP Server (Experimental)

Add to your .mcp.json:

{
  "mcpServers": {
    "codemoot": {
      "command": "npx",
      "args": ["@codemoot/mcp-server"],
      "env": {
        "CODEMOOT_PROJECT_DIR": "/path/to/your/project"
      }
    }
  }
}

Tools: codemoot_review, codemoot_plan, codemoot_debate, codemoot_memory, codemoot_cost

Development

git clone https://github.com/katarmal-ram/codemoot.git
cd codemoot
pnpm install
pnpm build
pnpm test         # 624 tests across 47 files
pnpm lint         # Biome linter
pnpm typecheck    # TypeScript strict checks

Known Limitations

• Background job worker must be started manually (auto-spawn coming)
• Watch mode enqueues jobs but requires worker process
• MCP server is experimental — core + CLI are stable
• Autofix loop depends on GPT's ability to apply edits via Codex tools
• Windows path normalization may have edge cases

License

MIT