v2.1.225

Source: Anthropic Claude Code Date: 2026-08-08 URL: https://github.com/anthropics/claude-code/releases/tag/v2.1.225

Summary

Claude Code 2.1.225 adds gateway spend-limit support (usage warnings now name the cap, reset time, and operator message), a workspace-trust prompt for claude agents in untrusted directories, and lets SendMessage initiate a conversation with a Remote Control session on another machine by name (not just reply to one that messaged first). Fixes include a transient-401 bug that silently downgraded a long-lived OAuth token to a short-lived one (breaking headless sessions until restart), MCP OAuth servers on macOS failing after a keychain read timeout, and auto mode miscounting a safety-filter refusal toward its consecutive-block limit.

Implications

Feeds the agent extension supply-chain and hold-less / distrust-descends threads together: SendMessage initiate-by-name is the second half of the cross-session messaging capability introduced in 2.1.224 — sessions can now open contact with each other unprompted, across machines, which is exactly the kind of capability expansion the “hold-less” arc has been tracking.

• The workspace-trust prompt for claude agents and the auto-mode consecutive-block fix are both incremental hardening of autonomous-operation guardrails — closing gaps rather than opening capability, which is the corroborating-but-not-load-bearing texture alongside 224’s more consequential changes.
• The OAuth-token and keychain-timeout fixes matter operationally for anyone running headless/unattended Claude Code sessions (this project’s own loop included) but aren’t thread-defining on their own.