# Claude Code auto-continue: what's built in, and what it doesn't do

Source: https://ryansam.com/agent-auto-continue/claude-code/
Published: 2026-10-02 Updated: 2026-10-02

## You hit the limit mid-task

Claude Code is halfway through a refactor and prints `You've hit your session limit · resets 3:45pm`.
You don't need to wait up.
Since v2.1.234 (August 2026), Claude Code [waits in the open session and picks the task back up by itself](https://code.claude.com/docs/en/interactive-mode#wait-for-a-usage-limit-to-reset) after the reset.
It's on by default in interactive sessions signed in with a claude.ai subscription.
The switch is **Continue automatically at usage limit** in `/config`, or [`autoContinueAtUsageLimit`](https://code.claude.com/docs/en/settings-reference#autocontinueatusagelimit) in your user settings.

Leaving the terminal open and just want the task picked back up?
That's it.
Don't install anything.

![A rendered Claude Code session: a prompt to fix flaky auth tests and a one-line reply, then the session limit message with a 3:45pm reset and a line saying it will continue automatically at 3:45pm. Under the empty prompt, two lines repeat the reset time and the automatic continue, with esc to cancel.](https://ryansam.com/_astro/platform-limit-light.JMI9iQNd.png)

*What Claude Code shows at the session limit, rendered from its documented wording.*

## What the built-in one does

- At the reset, it prompts Claude to pick up where it stopped.
  It doesn't resend your last message.
- The session has to stay open.
  Exit and the wait is gone, even if you resume later.
- Permission prompts still appear, so the task can stall while you're away.
- A weekly limit over 24 hours out won't start a wait on its own; start one from `/rate-limit-options`.
- If the Mac slept over about 30 minutes through the reset, it waits for Enter.
- Claude Code Desktop has a separate [**Auto-continue when limits reset**](https://code.claude.com/docs/en/errors#youve-hit-your-session-limit) checkbox on the session-limit card (not the weekly one).

![The Config tab of Claude Code settings, opened with /config, listing Auto-compact, Continue automatically at usage limit, Switch models when a message is flagged and more. A blue box marks the Continue automatically at usage limit row, set to true.](https://ryansam.com/_astro/platform-config-light.B7dQSA2-.png)

*The switch lives in /config, and it's on by default.*

Limits refresher: a [rolling five-hour window and a weekly one](https://code.claude.com/docs/en/statusline#rate-limit-usage), and [usage counts against both](https://code.claude.com/docs/en/errors#youve-hit-your-session-limit).
`/usage` shows [when each resets](https://code.claude.com/docs/en/errors#youve-hit-your-session-limit).

## Side by side

[Agent Auto-Continue](https://ryansam.com/agent-auto-continue/) is a free macOS menu-bar app that sends a message into a Claude Code session when you choose ("run the tests and fix what fails", at 03:00).

| | Built-in | Agent Auto-Continue |
| --- | --- | --- |
| After a session limit | Picks the task up after the reset | Sends your message after the reset |
| Any message at a time you choose | No | Yes |
| Several turns | No | Up to a turn limit or until stopped, optional stop phrase |
| The session | Must stay open | Must be closed |
| Weekly limit | Waits itself only within 24 hours | Waits when it knows the reset time |
| Permission prompts | Asks, then waits for you | Denied by Claude Code; the turn carries on without it |
| Codex and the desktop apps | No (Desktop has its own checkbox) | Yes ([Codex guide](https://ryansam.com/agent-auto-continue/codex/)) |
| Keep the Mac awake | No | Optional, off by default |

![The Agent Auto-Continue menu-bar rail with the Claude Code session "Fix the flaky auth tests" chosen, When free selected because Claude Code is limited until 03:00, and Up to 6 turns with the stop phrase TASK COMPLETE. A blue box marks the line under Continue, which reads "When Claude Code is free, around 03:00 · up to 6 turns, or at TASK COMPLETE".](https://ryansam.com/_astro/when-free-light.CVd8Aboi.png)

*The line under Continue says exactly what will happen.*

## Setting it up

1. Install the app and check Claude Code's status under **Settings > Agents**.
2. Switch the session to a permission mode you're happy to leave unattended.
   The app reuses the session's last mode.
3. For several turns, ask Claude to end its last message with `TASK COMPLETE` when it's done.
4. Exit Claude Code.
5. In the menu-bar popover, pick the session, when (**When free**, or a time) and how far (**Once**, **Up to** some turns, or **Until done** with the stop phrase).
6. Press **Continue**.
   The line under it says exactly what will happen.

## The honest bits

- **The session must be closed.**
  One still open in a live `claude` fails with `conversation_busy` (a second writer would fork it).
  A one-off message then retries with backoff for up to six hours.
- **Limits are inferred, not reported.**
  They come from Claude Desktop's usage samples if one is under 20 minutes old, else the newest limit message in recent transcripts, with its reset time if stated.
  A limit with no known reset doesn't block a send.
- **It can't tell when the task is done.**
  The turn limit and Stop are the real guards.
  Three usage-limited turns in a row pause the chain, as do three sub-minute ones.
- **Quitting the app interrupts a running turn** with SIGINT, which Claude Code treats as a clean stop.
- **It checks before sending.**
  It cancels if the session's gone or you've replied, and never resends an unconfirmed send.

It runs Claude Code's own headless mode (`claude -p --resume`), verified with 2.1.286.
No fake keystrokes, nothing changed in Claude Code.
Claude Desktop and the ChatGPT app are supported through macOS Accessibility instead; they'll get their own page.