cli-continues

I need you to build a CLI tool that lets developers seamlessly handoff AI coding sessions between different tools. The idea is simple but powerful — when you hit a rate limit or want to switch tools mid-session in Claude Code, Cursor, Copilot, or any of 16 other AI coding agents, you run `continues` and it extracts your entire session (conversation history, file changes, working state, reasoning) and injects it into whichever tool you want to continue in next.
The tool needs to support 16 different AI coding agents: Claude Code, Codex, GitHub Copilot CLI, Gemini CLI, Cursor, Amp, Cline, Roo Code, Kilo Code, Kiro, Crush, OpenCode, Factory Droid, Antigravity, Kimi CLI, and Qwen Code. Each tool stores sessions in a completely different format — some use JSONL, some JSON, some SQLite, some YAML — so you'll need flexible parsers for each.
The main workflow is discovery, parsing, extraction, and handoff. First, scan the user's system for session directories from all 16 tools. Then parse whatever format each tool uses. Extract the recent conversation messages, file changes that were made, any commands that ran, and the AI's reasoning. Finally, generate a clean handoff document that explains what was being worked on and inject it into the target tool.
Build an interactive CLI using commander and clack prompts where users can either run `continues` to get an interactive picker showing all their sessions, or use quick shortcuts like `continues claude` for the latest Claude session or `continues codex 3` for the 3rd most recent Codex session. When they pick a source session and target tool, generate the context document and write it to the right location so the new tool can pick it up.
Use TypeScript for type safety since you're dealing with lots of different data structures. Use zod for validation since you're parsing potentially messy session data from multiple sources. Use YAML for config files so users can customize behavior. The interactive prompts should be smooth with ora for spinners and chalk for colors. Keep the whole thing lightweight — this needs to run fast since users will run it constantly between sessions.
The key insight is that this isn't about replicating entire conversations — it's about extracting the essential context: what files changed, what's the current problem being solved, what's been tried, what's next. Make the handoff document human-readable and well-structured so any AI agent can immediately understand the state and continue productively.