How to continue a Claude Code task after a usage limit

What Claude Code does at the limit

Subscription plans include a rolling usage allowance. When it runs out, Claude Code shows one of these and blocks requests until the time shown:

You've hit your session limit · resets 3:45pm
You've hit your weekly limit · resets Mon 12:00am
You've hit your Opus limit · resets 3:45pm
You've hit your Sonnet limit · resets 3:45pm

• The 5-hour and weekly limits are shared across models, so switching models doesn't help. The Opus and Sonnet limits apply only to that model family: /model to a model outside it keeps you working, at the cost of re-reading the conversation without the prompt cache.
• Claude Code warns you before a window runs out, for example You've used 85% of your session limit. /usage shows your limits and when they reset.
• /usage-credits buys extra usage on Pro, Max and Ultra, or requests it from your admin on Team and Enterprise.

Automatic continue (v2.1.234 and later)

In an interactive session signed in with a claude.ai subscription, Claude Code waits in the open session and continues the task after the reset. The setting is on by default. While it waits, the bottom of the session reads:

Usage limit reached · limit resets 3:45pm
Continuing automatically at 3:45pm · Esc to cancel

• Keep the session open. At the reset, Claude Code prompts Claude to pick the task up where it stopped. It doesn't resend your last message.
• If your computer slept for more than about 30 minutes across the reset, the session shows Press enter to continue, and waits for you to press Enter.
• Adding usage credits with /usage-credits, signing back in after /upgrade, or switching models with /model during the wait makes Claude Code check right away and continue if usage is available.
• The continued turn still asks for permissions as usual, so it can stop again on a prompt while you're away.
• If it hits the limit again, Claude Code waits again at most twice in a row, then stops with Automatic continue stopped after repeated usage-limit hits.

To cancel one wait, press Esc at an empty prompt or Ctrl+C, or run /rate-limit-options and pick Don't continue automatically. To turn the feature off, switch off Continue automatically at usage limit in /config, or set autoContinueAtUsageLimit to false in your user settings. Dynamic workflows pause the same way (v2.1.271 and later) and resume their waiting agents after the reset.

In the Desktop app, the Code tab's session-limit card has its own Auto-continue when limits reset checkbox. The weekly-limit card doesn't. The checkbox and the CLI setting are separate.

When it won't continue by itself

Claude Code doesn't start the wait on its own in these cases, though you can start one from /rate-limit-options (Wait here, then continue automatically):

• Remote Control and agent team teammate sessions.
• A reset more than 24 hours away, which is common for a weekly limit.
• An Opus or Sonnet limit while you're running a model outside that family.

It doesn't offer the wait at all for background sessions and claude -p runs, for API keys, cloud providers and usage-based billing (metered per request, so there's no reset), or behind an LLM gateway without a saved claude.ai login. A wait also ends without continuing if you exit Claude Code, switch accounts with /login, clear or rewind the conversation, /resume another session, or move the session to Desktop, the background or the cloud. After an exit, the wait doesn't come back when you resume the session.

Resume it yourself

Once the limit has reset, reopen the conversation and tell Claude to carry on:

claude --continue        # the most recent session in this folder
claude --resume          # pick from the session list
claude --resume <name>   # a named session, or pass a session ID

Then send a short prompt such as "The usage limit has reset. Check git status and the plan, then carry on from the step you were on." On Pro and Max, Claude Code warns you when resuming an inactive conversation would use a noticeable share of your 5-hour limit, and lets you resume anyway or start a new conversation that can look up the old one when you refer to it.

For scripted claude -p runs, set CLAUDE_CODE_RETRY_WATCHDOG=1. Claude Code then retries 429 errors indefinitely, waiting until the limit resets when the response carries a reset time, so a run that hits a usage limit waits out the window instead of failing. CLAUDE_CODE_RETRY_WATCHDOG_MAX_WAIT_MS caps the wait per request, which stops a run from sitting for days on a weekly limit.

Pick up in a new session or another agent

Sometimes resuming isn't the right move: the transcript is huge, the weekly reset is days out, or you want Codex or Cursor to finish the job now. A new session doesn't know which step was open or what was decided unless something outside the conversation records it. A plan file in the repo works for one tool. Taskhold keeps the same state outside every tool: the task's steps, which one is running, the last progress message and the notes, on the live board.

• A new Claude Code session that follows the Taskhold workflow in CLAUDE.md calls taskhold_get_current, sees the task in progress and its open step, and continues from there.
• To move the task to another agent, open it in Taskhold and use Continue in another agent. It copies a prompt for Codex, Cursor or any client that already has Taskhold connected.
• The other agent's progress shows on the same card, with its own name, so you can see who finished which step.

Continue the Taskhold task "Move billing emails to the new queue" in project api-server.
Claude Code started it; you're taking over.

1. Read it first: taskhold_get {task_id: "<task id>"}. It has the description, the steps, the notes and what was done so far.
2. Pick it up: taskhold_begin {task_id: "<task id>", project: "api-server", agent: "<your client> in <repo folder>"}. Don't pass steps unless you want to add new ones; the plan is already there.
3. Carry on from "Retry failed sends" (step <step id>). Finish each step with taskhold_complete {task_id: <step id>, start_next: true}.
4. If you need me, ask with taskhold_block {task_id: "<task id>", reason: "<one question>"} and wait for my reply with taskhold_updates.

• Taskhold doesn't watch your usage or restart a session. Something still has to start the next one: you, automatic continue, or a script.
• The open step is as current as the agent's last report. Uncommitted changes live in your working tree, so tell the next agent to check git status first.
• Setup is on the Claude Code page and the Codex page; a CLAUDE.md task workflow covers the instructions that make a new session check the list first.