Claude Code 2.1.329 adds clearer warnings about hooks and MCP connection failures, and fixes a wide range of problems affecting sign-in, sessions and coding workflows. For developers, the standout is a batch of practical repairs, including a Code Mode memory runaway fix and a code-review fix that makes the tool read the pull request diff rather than the checked-out branch.
Anthropic Watch analysis
What happened
Anthropic’s Claude Code 2.1.329 release notes, published on 19 November, describe new notices and fixes across Claude Code, Claude Desktop and self-hosted runners.
The release adds a trust-dialog warning showing where a folder’s hooks are declared, and a conversation warning when hooks in managed settings fail to load. It also adds scheduled tasks, /loop wake-ups and conversation forks to /tasks, and shows the JSON-RPC error message when a remote HTTP MCP server refuses a connection.
Among the fixes, Claude Code should no longer reset ~/.claude.json to defaults after a network stall or file lock makes it temporarily unreadable. Another fix addresses repeated macOS sign-outs when Claude Code can read, but not save to, the Keychain. Code review now reads the pull request diff, or says when it cannot, instead of silently reviewing the checked-out branch. Anthropic also says a runaway Code Mode loop is now stopped at a limit, rather than being allowed to fill the machine’s memory.
Why it matters
These changes touch the unglamorous parts that decide whether a coding assistant behaves predictably: credentials, hooks, session recovery, code review and memory use. A clearer error can save time; a fix that prevents a review from targeting the wrong code can change what a developer actually gets back.
Our read
This is a substantial reliability release, not just a long list of tidying. The most useful improvements make failures more visible and fix cases where the tool could lose settings, sign users out or consume excessive memory. If you rely on Code Mode, MCP servers, macOS sign-in or automated code review, check the notes against your setup. Release notes are not a substitute for testing, but at least these give you specific behaviours to look for.
What to watch
- Whether the sign-in and settings fixes resolve the problems Anthropic lists.
- Whether Code Mode’s memory limit prevents runaway use without interrupting normal work.
- Whether developers find the new hook and MCP error messages useful in practice.
Discussion spark: For an AI coding tool, which deserves priority: adding new capabilities, or making sessions, permissions and code review dependable?
Sources and evidence
- v2.1.329 (19 November 2026, 22:14 UTC)
Anthropic Watch is independently operated by WittyWires. It is not affiliated with, endorsed by, or operated by Anthropic.