parcadei/continuous-claude-v3

Continuous Claude is a configuration framework that sits on top of Claude Code, the AI coding assistant's command-line tool. It exists to solve a specific problem: when you use Claude Code long enough in one session, the conversation history gets compressed and Claude loses the context of earlier decisions. Continuous Claude works around this by writing structured handoff notes at the end of each session, which the next session reads to pick up where things left off. The project's motto is "compound, don't compact." The framework installs 109 skills, 32 specialized sub-agents, and 30 hooks into your Claude Code configuration. Skills are reusable commands for common tasks like committing code, running quality checks, or kicking off a debugging workflow. Hooks are scripts that run automatically at certain moments, such as after you edit a file, to catch errors early without waiting for tests. Sub-agents handle specific kinds of work in isolated context windows so their activity does not crowd out your main conversation. A key feature is what the project calls TLDR code analysis: instead of reading entire source files into context, the system builds a semantic index that lets Claude understand a codebase without burning tokens on every file. There is also a memory system backed by a PostgreSQL database that extracts lessons from each session and surfaces them at the start of the next one. You do not need to memorize slash commands to use it. When you type a natural-language request, a hook injects a reminder about which skills and agents are relevant, and Claude decides which to invoke. Direct slash commands like /fix or /build still work for users who prefer them. Setup requires Python 3.11 or later, the uv package manager, Docker for the database, and the Claude Code CLI. A wizard script guides you through a 12-step installation. The project is licensed under MIT. The full README is longer than what was shown.