Patch release (2.1.224 → 2.1.226): webhook triggers for remote automation, clearer focus-mode and correction/memory behavior, and sturdier device-bridge / remote-session plumbing.
Changelog
[new] Remote triggers can create webhook triggers (create_webhook_trigger), backed by a dedicated webhook-triggers API.
[changed] Focus mode: when it is on, only the final text of each turn is shown — progress chatter between tool calls is skipped so everything the user needs lands in that last message.
[changed] Corrections and memory: stronger guidance to write durable memories in the same turn that engages a correction (not after the conversation “settles”), including when the reply ends in an offered next step.
[changed] Reply-only artifact threads: the composer can answer questions substantively and flag change requests for the owning session, without claiming edits or narrating its own limitations.
[changed] Thread listing prefers threads with comments sent to you, then other threads newest-first.
[changed] Auto-mode classifier prompts push block-first staging (full effect of an action, not surface form) and note that read-only work does not need the classifier.
[changed] Permission-denied tool feedback is clearer when a use was rejected (e.g. an edit was not applied), including optional user-provided denial text.
[changed] After opening or updating a PR, if CLAUDE.md mentions Slack channels, Claude can offer to post the PR URL there — only after confirmation, and it stays quiet if no Slack tool is available.
[changed] Safer multi-session git stash guidance: avoid bare stash/pop on the shared stack; prefer WIP commits or tagged stash apply/drop by SHA.
Models & prompts
Context-management copy spells out long-conversation summarization: work continues from the summary plus remaining unsummarized context — no need to wrap up early mid-task.
Subagent defaults remain background-with-notification; prompts still cover synchronous run_in_background: false, remote isolation, and teammate limits where those apply.
Coordinator-mode, onboarding-guide, ultrareview/high-effort review, and computer-use MCP instructions were refreshed for consistency with the above behaviors.
Push/PR shipping policy text was tightened (no force-push or merge to main; worktree-friendly shipping when the checkout isn’t cleanly yours).
Under the hood
Device-bridge lifecycle is instrumented end-to-end (start/connect/heartbeat/reconnect/reject/stop and related failures), with auth-revive paths for bridge and REPL sessions.
New configuration knobs and session controls, including nonessential-traffic disable, scroll speed, effort level, session name, resume prompt, host auth refresh timeout, container id, remote-tools bridge URL, remote-control session name prefix, and workshop progress tracking.
Additional backend routes for notification preferences, CLI OAuth API-key creation, and v2 logs.
SDK surface: RemoteTriggerInput.action extended with create_webhook_trigger; tested SDK wrapper list updated (adds 0.3.216, drops 0.3.183 from the tested set).
Internal background-backend flag naming cleaned up (CLAUDE_BG_BACKEND removed in favor of newer controls).
Upgrade notes
If you integrate remote triggers via the SDK, create_webhook_trigger is available on the remote-trigger action union — update types/clients if you pin them.
Optional env-based tuning (CLAUDE_CODE_DISABLE_NONESSENTIAL_TRAFFIC, CLAUDE_CODE_SCROLL_SPEED, CLAUDE_CODE_EFFORT_LEVEL, CLAUDE_CODE_SESSION_NAME, and related) is present for environments that opt in; defaults are unchanged if unset.
content/
Browse captured content/
Same tree as claude-code/latest/content for this version. Open the full browser for sidebar navigation between docs and other files.
