Continuous Claude

Stale
GitHub Python MIT

Description

Context management for Claude Code with hooks for state maintenance via ledgers and handoffs. Enables MCP execution without context pollution and agent orchestration with isolated context windows for long-running conversations.

Key Features

  • Persistent learning system with YAML handoffs that preserve context across sessions, solving Claude Code's compaction problem
  • 109 natural-language-triggered skills with automatic activation via rule-based context injection hooks
  • 32 specialized agents including scout (exploration), oracle (research), debug-agent, and kraken for parallelized task execution
  • TLDR 5-layer code analysis with semantic index that reduces token burn by analyzing only what's needed
  • Memory daemon that auto-extracts learnings from each session and surfaces them in future contexts
  • 30 hooks for shift-left validation running pyright/ruff after edits, catching errors before tests

Use Cases

💡 Long-running development projects where context must persist across multiple Claude Code sessions
💡 Complex debugging workflows that benefit from specialized scout and debug agents working in parallel
💡 Codebases where understanding architecture requires multi-layered code analysis to save token costs
💡 Teams wanting Claude Code to learn and improve from past decisions without manual knowledge management
💡 Workflows that need automatic validation hooks to catch type errors and linting issues immediately

Tags

mcp agent python tools memory workflow

Categories

🔌 MCP Tools ⚡ Agent Tools

Quick Start

Clone the repo, cd into `opc/`, and run `uv run python -m scripts.setup.wizard` for the 12-step setup wizard. It handles Docker, database, API keys, and installs 32 agents, 109 skills, and 30 hooks into your Claude Code environment.
Visit GitHub

Related Projects