Source: github.com/anthropics/claude-code/releases/tag/v2.1.329
The claude-code release page published v2.1.329 on 19 Nov. It adds a trust dialog that names where a folder's hooks are declared, a conversation warning when hooks in managed settings fail to load, and a "Server message" row in /mcp that shows the JSON-RPC error from a remote HTTP MCP server that refuses a connection. Code Mode can now stay on while OpenTelemetry events and metrics are enabled.
Fixes cover macOS sign-outs when the Keychain can be read but not written, a ~/.claude.json reset after a failed save, and /teleport replacing an open conversation. Several Code Mode problems are addressed too, including memory growth from runaway loops and large text files. The release also corrects /usage and /model reporting for Pro and Max users when usage credits run out.
Among the changes, most scheduled tasks saved by earlier versions need scheduling again in new signed-in sessions. Code Mode now starts even when telemetry is turned off, and a PreToolUse hook that errors no longer refuses a script's tool call unless it is set to block. Computer use on macOS also gets newer safety rules and gives up on app launches that take longer than 30 seconds.
Why it matters if you build agents
Teams running Claude Code in unattended jobs should check their scheduled tasks after upgrading, since most tasks saved by earlier versions must be scheduled again in new signed-in sessions.
If a hook must stop a Code Mode script's tool call, it now needs "onFailure": "block" on a command, HTTP or MCP tool hook, because an error alone no longer refuses the call.
Get stories like this every weekday
One short email with the day's agent tooling news.