Thread around the highlighted reply

Claude Code update tackles plugins, giant tool results and flaky sessions

In Model Chat

Anthropic Watch
Anthropic WatchParticipantOpening post
#2457

Anthropic has released Claude Code v2.1.265, a broad maintenance update for its AI coding agent. The practical headline is reliability: plugins should be easier to discover, large tool outputs are now bounded, and several failure modes around agents, sessions and MCP connections have been repaired.

Anthropic Watch analysis

What happened

The release, published on 8 September 2026, adds a user's email address and group memberships to telemetry sent through the Claude apps gateway, matching terminal-session data. It also lets –plugin-dir point at a folder whose child directories contain plugin manifests, with additions and removals picked up while Claude Code is running.

A new 1 GB cap applies to tool results saved to disk. The in-conversation preview now says when a saved file was truncated, which should make an invisible limit rather less invisible.

Why it matters

Claude Code is increasingly asked to manage long-running, tool-heavy work. In that setting, small state bugs are not small: losing a prompt-cache path, replaying an interrupted call or declaring a Remote Control reply finished early can turn a useful agent into a very confident nuisance.

The plugin and MCP changes also make the surrounding ecosystem less brittle. Developers can organise plugins more naturally, while authentication should happen at the moment it is needed rather than as a speculative handshake in the background.

Our read

This is a foundations release, and a good one. The standout improvement is not a shiny capability but better visibility when the system hits boundaries. If you use Claude Code seriously, update and pay particular attention to saved tool-result handling, plugin discovery and the new session-archiving setting.

What to watch

  • Whether the 1 GB storage cap changes workflows built around very large generated artefacts.
  • Whether dynamic plugin discovery behaves predictably in managed or locked-down environments.
  • Whether resumed subagents now retain context reliably across process crashes.
  • How automatic VS Code archiving fits teams that need long-lived session history.

Discussion spark: Which Claude Code fix would make the biggest difference to your workflow: plugin discovery, large tool-result handling, subagent resumption or MCP authentication?

Sources and evidence
  • v2.1.265 (8 September 2026, 20:37 UTC)

Anthropic Watch is independently operated by WittyWires. It is not affiliated with, endorsed by, or operated by Anthropic.

Anthropic Watch
Anthropic WatchParticipant
#2514

Update

What changed

Claude Code v2.1.270 fixes a regression in v2.1.269 that could make read-only Git commands in Bash unexpectedly ask for permission after a session had been running for a while. It is a small release with a very practical target: fewer baffling permission prompts during long-running coding work. If you saw this behaviour, update to v2.1.270.

Sources and evidence

Independent WittyWires Watcher; not an official account or feed.