Claude-Cortex

Multi-model development orchestration for Claude Code, Codex, and Gemini.

Documentation · Install · Quick Start

Cortex is a development orchestration framework that coordinates AI agents across model providers. It enforces quality gates — independent code review, test coverage audits, and lint checks — so that no agent grades its own homework. The result is a structured, auditable development workflow where Claude, Codex, and Gemini collaborate with built-in verification at every step.

How It Works

Multi-model review with no self-review

Cortex's core principle: the agent that writes the code never reviews it. When Codex implements a feature, the review is routed to Claude first, then to a different model family, with same-model review as a last resort. Every review produces a structured artifact with severity levels and a pass/fail verdict.

Codex implements → Claude reviews → Codex remediates → Claude re-reviews
                   ↓ (unavailable)
                   Gemini reviews (fallback)
                   ↓ (unavailable)
                   Fresh-context Codex reviews (last resort)

Agent-loops: progressive quality gates

Every code change flows through three sequential loops, each with circuit breakers and escalation rules:

Code Change Loop    →  Implement → Independent review → Remediate P0/P1 → Re-review (max 3 cycles)
Test Writing Loop   →  Audit gaps → Write tests → Verify → Re-audit (max 3 cycles)
Lint Gate           →  Discover linter → Auto-fix → Check → Remediate (max 2 cycles)

P0/P1 findings must be resolved before the loop exits. P2/P3 findings are filed as issues automatically. If circuit breakers trigger, the agent stops and escalates to a human — no infinite remediation loops.

Skill recommendations

Skills are suggested automatically as you work via a two-layer pipeline: fast keyword matching on every prompt (~50ms), with optional semantic matching for deeper recommendations. The TUI runs a background watch daemon for continuous suggestions.

What's Inside

Key skills

• agent-loops — The core workflow: structured implementation with independent review, test audit, and lint gates. Includes provider-aware review scripts with fallback chains.
• test-review — Test quality and coverage auditing across modules.
• doc-claim-validator — Validates documentation claims against actual code.
• doc-maintenance — Systematic documentation audit and lifecycle management.

Install

macOS (Homebrew)

brew tap NickCrew/cortex
brew install cortex

post_install symlinks bundled agents, skills, rules, and schemas into ~/.claude/ automatically. Opt out with CORTEX_SKIP_LINK=1 brew install cortex.

Python (any platform)

# Recommended
pipx install claude-cortex

# Alternative
pip install claude-cortex

The Python package also installs cortex-minimal, a smaller command surface for agent runtime environments:

uvx --from claude-cortex cortex-minimal --help
uvx --from claude-cortex cortex-minimal git --help

It exposes only git, tmux, statusline, hooks, and completions while sharing the same tested implementations as the full cortex CLI.

If you install via Homebrew and pip, whichever binary comes first in your $PATH wins. Uninstall one to avoid ambiguity.

Development

git clone https://github.com/NickCrew/claude-cortex.git
cd claude-cortex
pip install -e ".[dev]"

Quick Start

cortex install link          # Symlink bundled assets into ~/.claude
cortex install post          # Shell completions + man pages (optional)
cortex status                # Check what's active
cortex tui                   # Launch the terminal UI

CLI Overview

cortex <command> [options]

Run cortex --help or cortex <command> --help for details.

Development

just test                    # Run test suite
just lint                    # Check formatting (Black)
just type-check              # Strict mypy
just docs                    # Serve docs locally

See CONTRIBUTING.md for guidelines.

Documentation

Full documentation at cortex.atlascrew.dev.

• Getting Started
• Commands Reference
• Skills Guide
• Architecture
• Skill Recommendation Engine
• CHANGELOG

License

MIT. See LICENSE.