🏗️ CXM (ContextMachine) - Precision Code Forge

The Architect's Command Center for Deterministic AI Orchestration & Vibecoding.

CXM is a high-performance local framework designed to transform vague user intent ("vibes") into production-grade, secure code. It features an autonomous Vibe Router, a Precision Protocol against prompt injections, and a Multi-Agent Audit system to ensure every line of generated code meets enterprise standards before automatically patching it into your project.

---

🚀 The Vibecoding Experience (Zero-Click Coding)

Stop writing boilerplate. Just provide the "vibe" or intent, and CXM handles the rest autonomously:

1. Vibe Routing: You ask for a "secure login route". CXM's semantic router automatically selects the appropriate api-security Blueprint from your Vault. No manual configuration needed.
2. Synthetic Blueprints: Ask for something completely new? CXM generates a temporary set of architectural constraints on-the-fly to ensure the LLM doesn't hallucinate bad practices.
3. Auto-Patching: Once the code is generated and audited, CXM's FilePatcher applies the code directly to your local files using XML-style <file_patch> blocks.

---

🔮 Telepathic Multi-Step Refactoring

For massive architectural changes, a single patch isn't enough. CXM introduces the Telepathic Architect Agent to manage complex, multi-file refactorings without becoming overeager and ruining your project:

1. Implicit Gap Inference: If you ask for "Rate Limiting", CXM scans your code, notices you don't have a caching layer, and identifies this as a missing architectural gap.
2. Predictive Shadow-Scaffolding: Instead of wildly implementing a full Redis client you didn't ask for, CXM generates an abstract CacheInterface (stub). This allows the rest of the generated code to compile safely against a contract.
3. The Alignment Vibe-Check: Before any files are touched, CXM presents a TaskGraph and asks for your permission: "I will scaffold a CacheInterface and then write the Middleware. Proceed?"
4. Contract-Driven Execution: CXM executes the approved steps sequentially, running the Multi-Agent Audit on each node.

---

🛡️ The Precision Security Stack & Guardrails

CXM is hardened for professional environments. You maintain absolute control via the .cxm.yaml Project Manifest:

• Strict Write Protection: The FilePatcher will only modify files within the allowed_write_paths defined in your .cxm.yaml. Malicious LLM outputs attempting to overwrite core files are hard-blocked.
• Scraping Boundaries: Define include_paths and exclude_paths in .cxm.yaml to ensure the RAG engine only indexes what is relevant, protecting isolated or sensitive modules.
• Injection-Proof Prompting: Uses strict [USER_DATA] delimiters to prevent prompt manipulation.
• Safe System Execution: Zero use of shell=True. All commands are executed via isolated argument lists.

---

✨ Key Features (Nuclear Arsenal)

🧠 1. Pattern Vault & Blueprints

Stop relying on generic "Senior Dev" personas. CXM uses Cognitive Blueprints (YAML patterns) to force LLMs into specific architectural rails.

🔍 2. Advanced Selection Layer (The Gatekeeper)

Retrieved context must justify its existence. Our Context Evaluator performs:

• Semantic Relevance Check: Only highly relevant snippets pass the gate.
• Trigram Deduplication: Advanced character-level filtering removes redundant code hits, saving massive token costs.

🧪 3. Multi-Agent Audit Loop

Before code reaches the FilePatcher, it undergoes a triple-agent review:

• EfficiencyAudit: Detects Big-O bottlenecks and scaling issues.
• ComplianceAudit: Verifies adherence to business logic and constraints.
• SecurityAudit: Hunts for injection vectors and malformed input handling.

⚡ 4. High-Performance RAG

• HNSW Vector Search: Graph-based search using True Cosine Similarity for millisecond response times on large codebases.
• Reciprocal Rank Fusion (RRF): Industry-standard algorithm for merging keyword and semantic searches robustly.

---

🚀 Quick Start

📋 Installation

# Clone and setup
git clone https://github.com/Joeavaib/partner.git
cd partner
python3 -m venv venv
source venv/bin/activate

# Install with pinned dependencies
pip install -r requirements.txt
pip install -e .

⚙️ Setup Guardrails

Initialize your .cxm.yaml in your project root to control CXM's boundaries:

scraping:
  include_paths: ["src/"]
  exclude_paths: ["tests"]
patching:
  mode: ask_first # Or 'true' for full autonomy
  allowed_write_paths: ["src/utils", "src/modules"]

🛠️ Basic Usage

Run the Autonomous Vibecoding Loop:

# Test the autonomous flow with a vibe (no specific pattern required)
python3 src/main.py

---

📂 Project Structure

partner/
├── docs/                  # Documentation
├── src/
│   ├── core/              # RAG, Audit, Patcher, Factory, ContextStore
│   ├── ml/                # Intent-Analyzer, Evaluator, Assembler
│   ├── resources/         # Coding-Patterns (Blueprints)
│   ├── engines/           # Model Geometry (JSON)
│   ├── utils/             # Paths, Logger
│   └── main.py            # Orchestration Loop
├── .cxm.yaml              # Security Guardrails & Manifest
├── tests/                 # Pytest Suite
└── pyproject.toml         # Package Configuration

🧪 Testing

export PYTHONPATH=$PYTHONPATH:$(pwd)/src
pytest tests/

📄 License

MIT