ENHANCED-CLAUDE

Enhanced Claude

Self-Improving AI with Infinite Context and Auto-Learning Skills - Powered by Automatic Hooks

Overview

This repository transforms Claude Code into Enhanced Claude - a self-improving AI that:

1. Never forgets - RLM-based session persistence with intelligent segment recovery
2. No limits - Large inputs automatically detected, RLM workflow suggested
3. Self-improves - Skills automatically matched, tracked, and learning moments detected
4. Searchable history - Find past solutions without filling context

Everything is automatic via 9 Claude Code hooks.

Quick Start

Install the systems you need:

# Clone the repo
git clone https://github.com/RohanRamanna/ENHANCED-CLAUDE.git
cd ENHANCED-CLAUDE/installers

# Install all systems (recommended)
./system-a-session-persistence/install.sh
./system-b-rlm/install.sh
./system-c-auto-skills/install.sh

# Or install just what you need
./system-a-session-persistence/install.sh  # Session persistence only

After installation:

1. Run /hooks in Claude Code to reload hooks
2. Read the INSTRUCTIONS.md in each installer for CLAUDE.md configuration

The Five Systems (All Automatic)

The Hooks System

All automation is powered by 9 Python hooks in ~/.claude/hooks/:

~/.claude/hooks/
├── skill-matcher.py          # Every message: suggests matching skills
├── large-input-detector.py   # Every message: detects large inputs
├── history-search.py         # Every message: suggests relevant past work
├── learning-moment-pickup.py # Every message: picks up pending learning moments
├── skill-tracker.py          # After Read: tracks skill usage
├── detect-learning.py        # Before stop: detects learning moments, saves to file
├── history-indexer.py        # Before stop: indexes conversation history
├── live-session-indexer.py   # Before stop: chunks session into segments
└── session-recovery.py       # After compact: RLM-based intelligent recovery

Hook Events

What Happens Automatically

Skills Library (18 Skills)

Usage: Skills are auto-suggested, or invoke manually with /skill-name

RLM: Infinite Context

Based on MIT CSAIL's paper arXiv:2512.24601v1.

When large-input-detector.py suggests RLM:

# 1. Probe structure
python rlm_tools/probe.py input.txt

# 2. Chunk
python rlm_tools/chunk.py input.txt --output rlm_context/chunks/

# 3. Process with Task subagents (Claude does this)

# 4. Aggregate
python rlm_tools/aggregate.py rlm_context/results/

Verified Results

Session Persistence (RLM-Based)

Three files that persist across context compaction, plus intelligent segment recovery:

After compaction:

1. live-session-indexer.py has already chunked conversation into semantic segments
2. session-recovery.py loads persistence files + scores and retrieves the most relevant segments
3. Claude sees actual conversation excerpts, not just summaries - zero data loss

Repository Structure

PERSISTANT MEMORY/
├── CLAUDE.md              # Main guidance (hooks, systems, reference)
├── context.md             # Session persistence: current goal
├── todos.md               # Session persistence: task tracking
├── insights.md            # Session persistence: learnings
├── skills/                # Skills library (18 skills)
│   └── */SKILL.md
├── rlm_tools/             # RLM processing tools
│   ├── probe.py           # Analyze structure
│   ├── chunk.py           # Split files (semantic code chunking, progress tracking)
│   ├── aggregate.py       # Combine results
│   ├── parallel_process.py # Parallel processing coordination
│   └── sandbox.py         # Safe execution
├── rlm_context/           # RLM working directory
├── docs/
│   ├── HOW_TO_USE.md      # Complete guide
│   └── VERIFIED_TEST_RESULTS.md
└── requirements.txt

~/.claude/                 # User-level Claude Code config
├── settings.json          # Hook configuration (9 hooks)
├── hooks/                 # The 8 automation hooks + shared utilities
│   ├── hook_logger.py     # Shared logging utility
│   ├── skill-matcher.py
│   ├── large-input-detector.py
│   ├── history-search.py
│   ├── skill-tracker.py
│   ├── detect-learning.py
│   ├── history-indexer.py
│   ├── live-session-indexer.py
│   └── session-recovery.py
├── sessions/              # Live session segment indexes
│   └── <session-id>/segments.json
├── history/               # Searchable history index
│   └── index.json
└── skills/                # Skills library

Installation

Modular Installers

Install systems independently:

installers/
├── system-a-session-persistence/   # Session Persistence & Searchable History
│   ├── install.sh / install.bat    # Installers
│   ├── uninstall.sh / uninstall.bat
│   └── INSTRUCTIONS.md             # Claude instructions for setup
├── system-b-rlm/                   # RLM Detection & Processing
│   └── ...
└── system-c-auto-skills/           # Auto Skills & Skills Library
    └── ...

macOS/Linux:

cd /path/to/your/project
/path/to/installers/system-a-session-persistence/install.sh

Windows:

cd \path\to\your\project
\path\to\installers\system-a-session-persistence\install.bat

What Each System Installs

Each installer includes an INSTRUCTIONS.md with:

• Installation commands for macOS/Linux and Windows
• CLAUDE.md configuration (copy-paste ready)
• Verification steps

Install All Systems

# macOS/Linux
./installers/system-a-session-persistence/install.sh
./installers/system-b-rlm/install.sh
./installers/system-c-auto-skills/install.sh

:: Windows
installers\system-a-session-persistence\install.bat
installers\system-b-rlm\install.bat
installers\system-c-auto-skills\install.bat

After Installation

1. Run /hooks in Claude Code to reload hooks
2. Read each system's INSTRUCTIONS.md for CLAUDE.md configuration
3. See installers/TESTING.md for verification steps

Inspiration & References

Original Idea: Dylan Davis

The 3-file persistence system (context.md, todos.md, insights.md) was inspired by Dylan Davis's video on Claude Code persistent memory.

We extended this idea by applying RLM (Recursive Language Model) principles to conversation history, enabling intelligent segment-based recovery after context compaction.

RLM Research Paper

@article{zhang2025rlm,
  title={Recursive Language Models},
  author={Zhang, Alex L. and Kraska, Tim and Khattab, Omar},
  journal={arXiv preprint arXiv:2512.24601},
  year={2025}
}

License

MIT