What Anthropic says the release fixes
Catch-up briefing on the November 3 release (Central time).
Claude Code is Anthropic’s coding assistant. Anthropic says 2.1.318 fixes permission prompts being answered by keys typed before they appeared. [1]
A shell command is a text instruction that runs a computer tool. The notes also fix approved commands running in the session’s starting folder after a cloud or self-hosted runner restart. A command now stops if the requested folder is unavailable. [1]
Another fix addresses an automated job incorrectly reporting success in a particular custom extension workflow. [1]
For technicians
That success-reporting fix concerns claude -p runs without interactive chat, or Agent SDK software, when an error in the API, the connection between programs, cut off a reply while a mod hooked turn.step. The SDK is a toolkit connecting Claude to another program; the hook is a custom extension responding to a step in the run. [1]
Our take
Try a small job in a reviewable copy before trusting restarted automation with important work. For a website repair, use a copy with a harmless text change and a clear expected result.
First, check the working folder. Next, inspect the proposed command and approve it deliberately. Finally, read the changed files and check the output. Keep error messages with the result so an incomplete job can return to review.
A completion label is useful only when the work supports it. We reviewed the release notes; we have not installed or tested this version.