/auto-mode-setup
Claude Code’s /auto-mode-setup teaches auto mode your project
- Reads
- your project and recent sessions
- Also on
- Bedrock, Vertex, Foundry, the Claude apps gateway, telemetry-disabled sessions
Claude Code adds /auto-mode-setup, a guided setup that reads your project and recent sessions to teach auto mode about your environment. It also works on Bedrock, Vertex, Foundry and the Claude apps gateway.
Resume warns about usage
>Resuming an old conversation now warns before it eats your 5-hour limit
- Who sees it
- Pro and Max plans
- Your choice
- resume anyway, or start a new conversation
- Starting fresh
- Claude can look up the previous one if you refer to it
On Pro and Max plans, resuming an inactive conversation that would use a noticeable share of your five-hour limit now warns you first. Resume, or start fresh and Claude can look up the old one.
Worktrees and drives stay safe
>claude --worktree asks before deleting your branch, and rm -r of a drive gets the safety prompt
- claude --worktree
- no longer removes the worktree and its branch at exit without asking when HEAD was detached or on another branch
- rm -r
- of a mounted drive or a top-level folder like /Volumes, /media or /mnt/c now gets the safety prompt
After a branch switch, Claude Code's worktree mode could delete the worktree and its branch at exit without asking, losing commits. Now it asks. And deleting a whole mounted drive now gets the safety prompt.
Check a Managed Agents agent
>/claude-api managed-agents-secure checks an agent against Anthropic’s security guide
- Compares
- a Managed Agents agent’s settings with Anthropic’s published security guide
- Changes
- nothing; it only reports
If you build on Managed Agents, the Claude API skill gains a managed-agents-secure command. It compares your agent's settings with Anthropic's published security guide and changes nothing.
LLM gateways
>Claude Code works better behind LLM gateways
- context_management errors
- Claude Code resends the request without that field
- Tool search
- on by default behind a custom ANTHROPIC_BASE_URL; a 400 or 422 retries without it
Behind an LLM gateway, Claude Code now resends a request rejected with “Extra inputs are not permitted” without that field. Tool search is on by default there too, and retries without it when refused.
Fixes you may have hit
>Fixed: the 200k compact window, hour-long /loop wakeups, forgetful subagents
- CLAUDE_CODE_AUTO_COMPACT_WINDOW
- 200k no longer means 100K; k and M suffixes work and bad values warn
- /loop
- wakeups at the 1-hour maximum no longer miss the prompt cache
- Subagents
- keep their latest work when messaged or resumed right after finishing
Setting the auto compact window to 200k no longer gives you 100K. Hour-long /loop wakeups stop missing the prompt cache. And subagents keep their latest work when you message them right after they finish.
The full changelog
Every change is in the Claude Code changelog
https://github.com/anthropics/claude-code/blob/main/CHANGELOG.md
Do you run Claude Code through Bedrock, Vertex or your own gateway?
The Claude Code changelog has the rest, including Code Review levels and new hooks for mods. So, do you run Claude Code through Bedrock, Vertex or your own gateway?
