Claude-Cortex
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
| Path | Purpose |
|---|---|
agents/ | Agent definitions (specialized reviewers, implementers) |
skills/ | Reusable skill modules — workflow guidance, review prompts, quality standards |
rules/ | Behavioral guardrails and coding conventions |
hooks/ | Automation hooks (skill suggestions, validation gates) |
claude_ctx_py/ | Python CLI and TUI implementation |
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
# Recommended
pipx install claude-cortex
# Alternative
pip install claude-cortex
For 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]
| Command | What it does |
|---|---|
status | Show active agents, rules, hooks, skills |
agent list|status | Manage agent definitions |
skills list|info|recommend | Discover and inspect skills |
rules list|activate | Manage behavioral rules |
hooks list|install | Install and validate hooks |
mcp list|diagnose | MCP server discovery and diagnostics |
ai recommend|watch | AI-powered skill recommendations |
review | Run review workflows |
tui | Launch terminal UI |
docs | Browse bundled documentation |
install|uninstall | Manage Cortex installation |
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.
License
MIT. See LICENSE.