Cortex

This repository packages the Cortex (cortex) context management toolkit as a Claude Code plugin. It bundles the curated agents, commands, modes, rules, and supporting Python CLI + TUI so teams can install the complete experience through the plugin system or keep using the standalone cortex / cortex-ui scripts.

Note: The claude-ctx command has been deprecated but will remain available for a while.

📚 Docs: https://cortex.atlascrew.dev/

🎬 Presentations:

• Intro Overview
• Technical Deep Dive
• Executive Roadmap
• Feature Catalog

What's Inside

Core Assets (Loaded by Claude Code)

| Directory | Purpose | |-----------|---------| | agents/ | Claude subagents with dependency metadata. Move to inactive/agents/ to park. | | commands/ | Slash command definitions (50+ curated commands) | | skills/ | Reusable skill modules (80+ skills) with activation triggers and dependencies | | modes/ | Behavioral context modules that toggle workflow defaults | | rules/ | Rule sets symlinked to ~/.claude/rules/cortex/ on launch | | flags/ | Modular token-efficient flag packs (22 categories, 3,380 tokens total) | | hooks/ | Automation hooks for pre/post tool execution and session events |

Orchestration & Templates

| Directory | Purpose | |-----------|---------| | profiles/ | Project type presets (frontend, backend, devops, etc.) | | scenarios/ | Multi-step workflow scenarios | | workflows/ | Complex workstream orchestration templates | | prompts/ | Prompt templates: guidelines, personas, and reusable snippets | | templates/ | Initialization templates for new projects | | plugins/ | Additional plugin modules (accessibility, frontend, UX) |

Python Package (claude_ctx_py/)

| Module | Purpose | |--------|---------| | cli.py | Main CLI entrypoint with all subcommands | | launcher.py | cortex start implementation (plugin resolution, rules symlinking) | | core/ | Core domain logic (agents, skills, modes, rules, hooks, backups) | | tui/ | Textual-based terminal UI with 15+ specialized views | | intelligence/ | AI recommendation engine and pattern learning | | memory/ | Session memory and knowledge persistence | | analytics.py | Usage analytics and skill metrics |

Supporting Files

| Path | Purpose | |------|---------| | .claude-plugin/plugin.json | Plugin manifest for Claude Code marketplace | | docs/ | Documentation site (Jekyll-based) | | tests/ | Test suite with pytest | | schema/ | JSON schemas for validation | | scripts/ | Helper scripts for installation and maintenance |

🆕 Latest Updates

• Hook logging – Hook failures now write to ~/.cortex/logs/hooks.log (override via CORTEX_HOOK_LOG_PATH).
• Extra plugin dirs – cortex-config.json supports extra_plugin_dirs to pass multiple --plugin-dir entries.
• Watch daemon mode – cortex ai watch --daemon runs watch mode in the background with status/stop controls.
• Template guardrails – The TUI detects missing templates/ files in the active CLAUDE_PLUGIN_ROOT and offers to initialize them or run the setup wizard.
• Multi-LLM consult skill – Ask Gemini, OpenAI (Codex), or Qwen for a second opinion; configure provider API keys via the TUI Command Palette -> "Configure LLM Providers".
• Asset Manager reliability – “Update All” and “Install All in Category” now behave consistently with clearer prompts.
• Flag toggles restored – Spacebar toggling works again in the Flag Explorer and Flag Manager, updating FLAGS.md immediately.

✅ Stability Update: AI + Context Management

We’ve fixed major issues across AI recommendations and context state tracking. Auto-activation and watch mode are more reliable, and context activation now uses .active-* state files with cortex doctor and cortex setup migrate to keep everything consistent.

🔥 New: Super Saiyan Mode

Universal visual excellence framework with platform detection:

• modes/Super_Saiyan.md – Core generic mode with auto-detection
• modes/supersaiyan/ – Platform-specific implementations (Web, TUI, CLI, Docs)
• claude_ctx_py/tui_supersaiyan.py – Enhanced Textual components
• examples/supersaiyan_demo.py – Interactive demo
• Three power levels: ⭐ Super Saiyan → ⚡ Kamehameha → 💥 Over 9000

Quick start:

python examples/supersaiyan_demo.py  # See it in action!

See Super Saiyan Integration Guide for details.

📦 New: Asset Manager

Install, diff, and update plugin assets directly from the TUI:

• Discover hooks, commands, agents, skills, modes, workflows, and flags
• Install/uninstall to any detected .claude directory
• Diff installed vs source, bulk install by category, update outdated assets

Quick start:

cortex tui
# Press 'A' for Asset Manager
# i=install, u=uninstall, d=diff, U=update all, I=bulk install, T=target dir

🌿 New: Worktree Manager

Manage git worktrees from the CLI or TUI.

CLI:

cortex worktree list
cortex worktree add my-branch --path ../worktrees/my-branch
cortex worktree remove my-branch
cortex worktree prune --dry-run
cortex worktree dir ../worktrees
cortex worktree dir --clear

TUI:

cortex tui
# Press 'C' for Worktrees
# Ctrl+N add, Ctrl+O open, Ctrl+W remove, Ctrl+K prune, Ctrl+B set base dir

🧭 New: Setup, Init & Migration

The installer and setup tooling have been overhauled to keep projects consistent across upgrades.

# Detect your project and recommend a profile
cortex init detect

# Apply a profile directly
cortex init profile backend

# Check init status
cortex init status

# Migrate CLAUDE.md comment activation → .active-* files
cortex setup migrate

🤖 New: AI Intelligence & Automation

Stay in Claude Code flow - Let AI manage the framework for you with context-aware intelligence, pattern learning, and auto-activation:

• Context Detection – Automatically analyzes changed files, detects auth/API/tests/frontend/backend
• Pattern Learning – Learns from successful sessions, recommends optimal agent combinations
• Workflow Prediction – Predicts agent sequences based on similar past work
• Auto-Activation – High-confidence agents activate automatically (≥80%)
• Watch Mode – Real-time monitoring with instant recommendations (no daemon required)
• TUI AI Assistant – Interactive view with recommendations and predictions (press 0)
• Skill Palette Shortcuts – Ctrl+P → type “Skill…” to run info, versions, deps, analytics, trending, or community install/validate/rate/search commands without leaving the TUI
• Multi-Reviewer Auto-Activation – Code changes can auto-activate multiple reviewers (quality, code, TS/React/UX, DB/SQL, performance, architecture)

Quick start:

# Get AI recommendations for current context
cortex ai recommend

# Auto-activate high-confidence agents
cortex ai auto-activate

# Start watch mode (real-time monitoring)
cortex ai watch

# Interactive TUI with AI assistant
cortex tui
# Press '0' for AI Assistant view
# Press 'A' to auto-activate recommendations

# Record successful sessions for learning
cortex ai record-success --outcome "feature complete"

Multi-review output example:

1. 🔵 quality-engineer [AUTO]
   Confidence: 85%
   Reason: Changes detected - quality review recommended

2. 🔵 code-reviewer [AUTO]
   Confidence: 75%
   Reason: Changes detected - code review recommended

3. 🔵 typescript-pro [AUTO]
   Confidence: 85%
   Reason: TypeScript changes detected - review recommended

4. 🔵 react-specialist [AUTO]
   Confidence: 80%
   Reason: React/UI component changes detected - review recommended

5. 🔵 ui-ux-designer [AUTO]
   Confidence: 80%
   Reason: User-facing UI changes detected - UX review recommended

Watch Mode Example:

══════════════════════════════════════════════════════════════════════
🤖 AI WATCH MODE - Real-time Intelligence
══════════════════════════════════════════════════════════════════════

[10:33:12] 🔍 Context detected: Backend, Auth
  3 files changed

  💡 Recommendations:
     🔵 quality-engineer [AUTO]
        85% - Changes detected - quality review recommended
     🔵 code-reviewer [AUTO]
        75% - Changes detected - code review recommended
     🔴 security-auditor [AUTO]
        95% - Auth code detected

[10:33:12] ⚡ Auto-activating 3 agents...
     ✓ quality-engineer
     ✓ code-reviewer
     ✓ security-auditor

See AI Intelligence Guide and Watch Mode Guide for complete documentation.

⭐ New: Skill Ratings & Auto-Feedback Loops

Phase 5 introduces a first-class feedback engine so skills can improve themselves:

• Ratings & Reviews – cortex skills rate <skill> stores star ratings, helpful/not-helpful votes, and optional text feedback in ~/.cortex/data/skill-ratings.db.
• Quality Metrics – cortex skills ratings <skill> shows averages, distributions, success correlation, and token efficiency; skills top-rated, skills export-ratings, and skills analytics expose the aggregate view.
• TUI Surfacing – The Skills table now includes a Rating column (press 5). Select a skill and press Ctrl+R to launch an inline rating dialog without leaving the terminal.
• Auto Prompts – Recent skill activations trigger modal prompts shortly after the TUI launches. The prompt explains why the skill was selected (usage count, task types, success rate) and offers to collect feedback on the spot. Dismiss once to snooze for 24 h; rating it clears future prompts until another burst of usage.
• Recommendation Feedback Loop – Ratings