Continuous Claude v2 addresses the context window limitations inherent in large language models like Claude Code by implementing a robust state management system. It enables persistent, token-efficient coding sessions and sophisticated agentic workflows through ledgers, handoffs, and externalized script execution (MCP). This project is designed for developers and power users who require long-term continuity and reduced context degradation in AI-assisted development, aiming to prevent the loss of signal and prevent LLM hallucination due to compacted context.

How It Works

The core mechanism involves saving session state to a "continuity ledger" and using "handoffs" to transfer work between sessions, allowing Claude's context to be cleared and reloaded with full fidelity. It leverages Meta-Command Protocol (MCP) for executing scripts externally, preserving valuable tokens. Agent orchestration follows a Plan → Validate → Implement flow, with an artifact index (SQLite + FTS5) and Braintrust integration for session tracing, learning extraction, and correlating handoffs with detailed execution traces.

Quick Start & Requirements

• Installation: Options include cloning the repo and running uv sync for per-project use, or executing ./install-global.sh for system-wide availability. Project initialization is done via ~/.claude/scripts/init-project.sh.
• Prerequisites: Python, uv (Python package manager). Optional API keys are required for external services like Braintrust, Perplexity, Firecrawl, Morph, and Nia to unlock full functionality.
• Licensing: A RepoPrompt Pro license is required for advanced codebase exploration features (MCP tools, CodeMaps).
• Docs: Links to official quick-start, docs, demo, or other relevant pages are not explicitly provided in the README.

Highlighted Details

• State Management: Lossless ledgers and handoffs ensure session continuity across /clear commands and restarts.
• MCP Execution: External script execution via MCP significantly reduces token consumption.
• Agent Orchestration: A structured Plan-Validate-Implement flow powers complex agentic tasks.
• TDD Workflow: Enforces writing failing tests before production code.
• Code Quality: Integrates qlty for automated linting and code quality checks.
• Braintrust Integration: Provides hierarchical session tracing, automatic learning extraction, and artifact index correlation.

Maintenance & Community

No specific details regarding maintainers, community channels (like Discord/Slack), or roadmap are present in the provided README.

Licensing & Compatibility

The project is released under the MIT License. Compatibility for commercial use is generally permissive, but the use of advanced RepoPrompt features requires a paid Pro license.

Limitations & Caveats

Advanced codebase exploration features are gated behind the paid RepoPrompt Pro license. Some hooks may introduce minor latency during execution. Global MCP server configurations are discouraged in favor of per-project setup to avoid unexpected tool usage.