continues CLI
continues carries a session from Claude Code, Codex, Cursor and 13 other agents into any other one, with messages, file edits and commands in tow
- 1,584 ★ on the repository
- Last commit
- MIT
Extended access
Why this level
- Reads the session folders of 16 agents, which hold the full text of your conversations and commands
- Writes `.continues-handoff.md` into the project folder, and a file with your work history can end up in a commit
- Launches the chosen agent with a prompt and passes it flags such as `--yolo` that remove confirmations
Install
- Works in
In your terminal, with SkillFoxx CLI
npx skillfoxx add cli/continuesDetects the agents on your machine, shows the access level and pins the version.
Other ways to install
Access and checks
How we reviewExtended accessThe tool runs code or makes network calls and only reads external data.
- Reads the session folders of 16 agents, which hold the full text of your conversations and commands
- Writes `.continues-handoff.md` into the project folder, and a file with your work history can end up in a commit
- Launches the chosen agent with a prompt and passes it flags such as `--yolo` that remove confirmations
This is third-party code. Review the repository files before installing.
What it does
continues is a Node.js CLI that takes one coding agent's session and opens its continuation in another. Running `npx continues` scans the session folders of 16 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. Sessions are listed with those from the current folder first, and you pick a session and the agent to continue in. The tool then reads the native storage format (JSONL, JSON, SQLite or YAML), pulls out the recent messages, file edits, commands and the agent's reasoning, and assembles a handoff document from them. The document is written to `.continues-handoff.md` in the project folder, and continues launches the target agent with a prompt that points to that file. Common flags (model, sandbox, auto-approve, extra folders) are translated into the target agent's equivalents, and unknown ones pass through unchanged. The `--debug-prompt` flag writes the file, prints the prompt and does not launch the agent. When a session should continue in the same agent, `continues claude` or `continues codex 3` calls its native resume without any context transfer. The other commands are `list`, `resume`, `inspect` for previewing what goes into the document, `dump` for exporting sessions to Markdown or JSON, and `scan` and `rebuild` for the index.
Who it is for. For developers who work with several coding agents, hit a usage limit in the middle of a task, or want a local export of their sessions.
Good fit when
- The usage limit in Claude Code ran out mid-debug and the work has to continue in Codex or Gemini CLI with the history and the list of edits
- A task started in Cursor but is easier to finish in a terminal agent, and the latest messages need to come along
- You want one table of sessions from every installed agent, or its JSON form: `continues list --json`
- For an archive or a review you need sessions exported to Markdown with `continues dump all ./sessions`
- Before a transfer you want to see what the receiving agent will get: `continues inspect <id> --write-md handoff.md`
Not a fit when
- You need to search the text of old sessions: continues has no such command, and Coding Agent Session Search (cass) covers that task
- The machine runs Node.js older than 22.5 and cannot be upgraded: the tool uses the built-in `node:sqlite` module
- Your agent is not among the 16 supported ones, for example Windsurf
- The receiving agent needs the whole conversation: the document is built from recent messages, not from the full history
Example request
The Claude Code limit ran out mid-debug. Move the latest session of this project to Codex so it knows which files we editedLimitations
- The receiving agent gets a text document, not the state of the earlier session. Four presets set its size: `minimal` takes the 3 latest messages, `standard` 10, `verbose` 20, `full` 50.
- The `.continues-handoff.md` file stays in the project folder. The developer's AGENTS.md warns that it may be committed or read by other AI tools, so check that it did not land in the git index.
- It needs Node.js 22.5 or newer and at least one of the 16 supported agents. The author wrote the parsers for nine of them (Amp to Qwen Code) by studying the Go adapters of another project, mnemo, and adapting the schemas to TypeScript, so storage formats in newer agent versions may drift from the parsing.
- The tool only reads sessions, and it keeps its index in `~/.continues/sessions.jsonl`, refreshed every 5 minutes. Defaults are looked up in the project's `.continues.yml`, then in `~/.continues/config.yml`.
- Version 4.1.1, last commit on 7 May 2026. MIT license.
How to disable. Running through `npx continues` leaves nothing behind except the index cache. A global install is removed with `npm uninstall -g continues`. The cache and config live in `~/.continues`, and `.continues-handoff.md` files in project folders are deleted by hand.
FAQ
Between which agents does continues move sessions?
Between any two of 16: 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. The author counts 240 transfer paths.
How is a transfer different from an ordinary session resume?
`continues claude` or `continues codex 3` calls the same agent's native resume and the history stays whole. A transfer to another agent goes through a handoff document that continues assembles from the parsed session.
What goes into the handoff document?
Recent messages, file edits and reads, shell commands with exit codes, code searches, MCP calls, reasoning blocks, subagent runs, token usage and the model. At the end come the path to the original session file, the session id and the project directory.
How do I set transfer defaults for one project?
Put a `.continues.yml` file in the project root, for example with `preset: verbose` and `recentMessages: 15`. The lookup order is the `--config` option, `.continues.yml` in the current folder, `~/.continues/config.yml`, then the `standard` preset. A full sample sits in `.continues.example.yml`.
Which flags does continues pass to the receiving agent?
It maps model, sandbox, auto-approve and extra directories to the target agent's equivalents, for example `--yolo`, `--search` and `--add-dir /tmp` for Codex. Other flags go through unchanged.
Discussion
No questions or reviews yet.
Related
CLI toolsAn environment for Claude Code that keeps context across sessions via hooks, ledgers and handoffs, orchestrates agents and saves tokens
A tmux-based TUI that runs Claude Code, Codex and other agents in one list: status, quick prompts, worktrees and diff review
A skill for messaging between Claude Code, Codex, Gemini CLI and other CLI agents through a shared local SQLite database, written in bash
A CLI and skill that gather coding-agent session history into one SQLite file the agent can search for past decisions
A memory and coordination layer for coding agents: specs and issues live in git while agents work in parallel
A CLI that strips ballast from Claude Code sessions, watches context with a background guard and preserves Agent Teams state
Comparison with similar tools
Entries that share the main search query, and how they differ.
| Entry | Official | Repository stars | Main difference |
|---|---|---|---|
| continues (this entry) | no | 1,584 | |
| ctx | no | 1,168 | searches the history of agent sessions on your machine and does not move a session to another agent |
| Coding Agent Session Search (cass) | no | 1,196 | gathers the session history of many agents into one search index and does not hand a session to another agent |
README in short
The repository's front page opens with `npx continues` and the scenario of hitting a usage limit, followed by tables of supported agents, launch modes, detail presets and where each agent stores its sessions. A separate part describes the tool activity block in the document and the steps for adding a new parser. The repository holds a changelog, parser tests for every agent, end-to-end transfer checks, instructions for agents in AGENTS.md and CLAUDE.md, and a sample config.