claudekit / updates / claude-code-2-1-225
[ PATCH · ]

Claude Code 2.1.225

Claude Code's usage warning now supports gateway spend limits, so the limit-reached message names the cap, its reset time, and the operator's message (it requires the gateway on 2.1.225). claude agents gained a workspace trust prompt for untrusted directories, matching the behavior of claude, and SendMessage can start a conversation with your Remote Control sessions on other machines by name, with ListAgents showing them as name [ref]. On the fix side, a transient 401 no longer replaces a long-lived CLAUDE_CODE_OAUTH_TOKEN with a stored login's short-lived token, MCP OAuth servers on macOS no longer fail with a burst of 401 errors after a keychain read times out, and conversation history no longer breaks on Remote Control session resume after very large conversations were compacted.

Claude Code
Official announcement →

This article is a summary based on official documentation.

What changed

Claude Code 2.1.225 shipped on August 8, 2026. Most of it is about auth tokens and session state drifting apart in ways that quietly broke headless and Remote Control sessions. Alongside those, gateway spend limits now surface in the usage warning, claude agents gained a workspace trust prompt, and SendMessage can start a conversation with a Remote Control session on another machine instead of only replying to one.

New features

  • Gateway spend-limit support in the usage warning

    When a gateway enforced a spend limit, the message you hit didn’t say what the cap was or when it would lift. Claude Code’s usage warning now supports gateway spend limits, and the limit-reached message names the cap, its reset time, and the operator’s message. It requires the gateway on 2.1.225.

  • Workspace trust prompt for claude agents

    claude prompted for workspace trust in untrusted directories; claude agents didn’t. claude agents now shows a workspace trust prompt for untrusted directories, matching the behavior of claude.

  • Starting conversations with Remote Control sessions by name

    SendMessage could only reply to Remote Control sessions on other machines after they messaged you first. It can now start a conversation with those sessions by name, and ListAgents shows them as name [ref].

Key improvements

  • Photos from the Claude app go straight to Claude

    A photo attached in the Claude app had to be read back from disk with a separate tool call before Claude could see it. Remote Control now shows attached photos to Claude directly.

  • Safety-filter refusals stop counting toward the consecutive-block limit

    Auto mode counted a safety-filter refusal of its own permission check toward the consecutive-block limit, pushing a session toward that limit over something that wasn’t a real block. The action is still denied, but the model is now told to move on rather than retry.

  • claude self-hosted-runner fails fast on an unusable --base-dir

    The runner registered even when --base-dir couldn’t be created or written, then failed every session afterward. It now exits at startup with a clear error.

Bug fixes

Authentication

  • A transient 401 replacing a long-lived CLAUDE_CODE_OAUTH_TOKEN with a stored login’s short-lived token, breaking headless sessions until restart — fixed.
  • MCP OAuth servers on macOS intermittently failing with a burst of 401 errors, as if never authenticated, after a keychain read timed out — fixed.

Sessions & messaging

  • Cross-session messages staying parked without a notice or expiry in headless sessions and during startup — fixed.
  • Conversation history breaking on Remote Control session resume after very large conversations were compacted — fixed.
  • Claude Code on the web sessions being misreported as stuck, re-sending a growing event backlog on every reconnect — fixed.
  • A Remote Control recipient you already confirmed in SendMessage being swapped for a same-named session on this machine when its own list couldn’t be checked — fixed.

Agents list & editor

  • Hovering over a session in another project in the agents list changing the directory the next agent starts in — fixed.
  • [VSCode] Focus view folding away the latest to-do list, a pending question’s context, and settled answers — fixed; thinking-only folds show “Thought for Ns” and re-collapse when their turn completes.

Notes

  • The spend-limit message needs both sides on 2.1.225 — upgrading Claude Code alone isn’t enough; it requires the gateway on 2.1.225.
  • Headless setups get the most out of this one — the CLAUDE_CODE_OAUTH_TOKEN swap and the parked cross-session messages both surfaced in headless sessions. Worth upgrading if you run Claude Code in CI or automation.
  • A new prompt can appear in claude agents — running it in an untrusted directory now shows a workspace trust prompt that wasn’t there before, which matters if you launch it from a script.
  • Auto mode’s consecutive-block limit itself is unchanged — only safety-filter refusals of its own permission check stop counting toward it. The action is still denied.