Skip to content

Claude Code

Claude Code stopped erasing your scheduled tasks

3 min read

Written by an AI model from the linked sources and checked against them by a second model. How we work

All about: Claude Code versions and changes →

Claude Code 2.1.328 is out. Like the last few releases, it is mostly fixes. One change matters for teams that run Claude through cloud providers: auto mode now covers Opus 4.6 and Sonnet 4.6 on Amazon Bedrock, Google Cloud’s Agent Platform, Microsoft Foundry and the Claude apps gateway. Another fix closes a trap that could quietly erase a project’s scheduled tasks.

Auto mode reaches older models

By default Claude Code asks before it runs a command or changes a file. Anthropic says this wears people down with approvals, and that users accept 93% of these prompts anyway. Auto mode hands the approvals to a classifier, a separate model that checks each action before it runs and stops the ones that look dangerous or don’t match what the user asked for. Your own deny and ask rules still come first, and routine reading and editing inside the project folder goes through without a check.

From this release, interactive sessions with Opus 4.6 or Sonnet 4.6 on those providers start in auto mode when no permission mode is set. Plan mode and Bash prompts on these models now use and offer it too, as they already did on newer models. If your team doesn’t want that, the disableAutoMode setting turns both off.

A related fix: behind a gateway that rejects beta features, with CLAUDE_CODE_DISABLE_EXPERIMENTAL_BETAS set, auto mode failed on every turn outside Claude Desktop on Bedrock, Agent Platform, Foundry or the Claude apps gateway. That no longer happens.

Where you could lose work

Saving a scheduled task could erase the project’s other saved tasks when Claude Code couldn’t read the scheduled tasks file. This hit people on third-party providers or with telemetry or feature flags turned off. Separately, a task with a typo in that file was deleted the next time the file was saved. Both are fixed.

Two smaller fixes are along the same lines. A resumed cloud session could lose its compaction summary, the condensed record of the conversation so far, or its recent messages when a message was saved late. And on self-hosted runners, a temporary disk error could lead to the deletion of a kept clone’s objects or refs, including commits that existed only there. Now that start fails instead.

What else changed

/code-review on most models now reports only bugs and violations of the project’s CLAUDE.md file. Suggestions about reuse, simplification and efficiency moved to /simplify. Each finding is now marked Important, Pre-existing or Nit, and each summary ends with a suggested fix.

When usage credits ran out mid-response, Claude Code showed an error that named a rate limit. That message is fixed. /bug now asks Yes or No before sending, so pressing Enter repeatedly no longer sends a report or opens GitHub. And Code Mode is faster in auto and bypass-permissions modes: the 0.75-second pause after most scripts is gone.

No comments yet. Be the first.

Join the conversation

Sign in to leave a comment. It appears after moderation.