clauding.de
DE

Claude Code

Claude Code 2.1.311: Standard Context Compacts Instead of Stopping

2 min read AI-generated

A lone backslash inside quotes or brackets used to vanish from Claude's replies. And startup is faster for Pro and Max accounts, because two network requests are skipped.

Featured image for "Claude Code 2.1.311: Standard Context Compacts Instead of Stopping"

Claude Code 2.1.311 landed in the npm registry on October 26 at 19:47 UTC. It is the longest changelog in weeks, and two of the fixes in it are not cosmetic.

The 200K stop is gone

Until now, a standard-context session could come to a halt with “Usage credits required for 1M context” as soon as a single request ran past 200,000 tokens. That hits exactly the moments where you least want it: a long conversation, a large set of files, and then nothing. Claude Code now compacts the conversation and carries on.

The second fix matters to anyone using subagents. Subagents could not use MCP tools from servers that finish connecting after the subagent starts. On top of that, SubagentStop hooks ran about twice a minute for each running subagent instead of once when it finishes. If you hung a notification or a script off that hook, you have been running it on a loop.

What else stands out

  • Long freezes at startup and after reading or editing files when a CLAUDE.md holds a very long list item.
  • Interrupting Claude only took effect after a timeout while an MCP server had not answered a resource read.
  • Write and Edit stalled for up to five seconds after running and reported “0 lines changed” when thousands of lines were modified.
  • Claude Code retried requests the API had explicitly marked not to retry — indefinitely in unattended retry mode.
  • Agent teams showed an empty task list after --resume, --continue or /clear.
  • A lone backslash inside quotes or brackets vanished from Claude’s replies.
  • New: --hold-idle-file and --max-hold-min for self-hosted-runner, so an idle session stays open while the host still reports processes of its own.
  • Startup is faster for Pro and Max accounts, because two network requests are skipped within an hour of the last check.

Two bugs that actually cost work

I wrote about 2.1.310 on Saturday, back then with the Monitor tool on Bedrock, Vertex and Foundry. This version brings no feature of that size, but two fixes you could feel: the 200K stop ended sessions that had work in them, and the double-firing SubagentStop hooks made any automation built on them useless. If you run subagents or agent teams, this is the best reason to update in several versions.

Sources

Claude CodeReleaseSubagentsMCP