Patch release: keyless Console sign-in, sandboxed Bash host prompting, MCP and fullscreen fixes, Remote Control recovery, and hardened claude.ai skills.
Changelog
[new] Keyless sign-in under /login → Anthropic Console: Sign in with your Console account (recommended) alongside creating an API key, so organizations that don’t allow API keys can still sign in
[changed] Sandboxed Bash no longer lists allowed network hosts in its prompt — Claude attempts requests and you can approve new hosts instead of treating unlisted hosts as blocked
[changed] Skills synced from claude.ai no longer shadow local commands or MCP prompts; descriptions are sanitized and labeled, and on your machine their bodies don’t run ! commands or expand @ files
[fix] MCP elicitation dialogs blanked out for URLs longer than 4,096 characters; permission prompts could drop “don’t ask again” when the project path didn’t fit the terminal width
[fix] MCP tools marked requiresUserInteraction still offered “Yes, and don’t ask again”; that option wrote an allow rule the tool then ignored
[fix] A Remote Control session whose process crashed stayed unavailable until claude remote-control was restarted; it can now be reused on the next message
[fix] Feature flags could be evaluated without the user’s subscription tier when a session started with an expired login token, which could wrongly prompt Max plan users to enable usage credits for Fable
[fix] Fullscreen mode could answer a permission prompt or press a button when you clicked the terminal only to refocus the window
[fix] Fullscreen mode could move keyboard focus onto the control under the pointer on that same refocus click
[changed] VS Code: improved attribution of the extension’s own usage telemetry to your organization when signed in with a Claude account, so its data-handling settings apply
[changed] Subagent guidance tightened: don’t spawn agents unless the user asks or names an agent type — multi-part or “thorough” tasks are handled inline by default
[changed] Background subagents remain the default when available; pass run_in_background: false only when the very next step depends on the result
Artifact workflows: clearer rules for reading every line before treating content as viewed, re-fetching when a handed-off artifact changed, and publishing from your own file; cloud sessions note that deleting Artifacts isn’t available there (use claude.ai or local /artifacts).
Automated/CI auth: federation-oriented path for non-interactive sign-in (federation rule with oauth_scope: org:admin on an admin service account, then federation env vars so the SDK/CLI exchanges and refreshes tokens without interactive login).
Focus mode: only the final text of each response is user-visible — skip interim progress narration between tool calls.
Memory and corrections: write durable memories in the same turn you engage a correction or preference; don’t defer past an offered fix.
PR follow-up: after creating/updating a PR, if CLAUDE.md mentions Slack, search for Slack send tools and ask before posting; skip quietly if none are available.
Writing and delivery polish: fewer decorative headers, less self-narration of reasoning, plain corrections, and stash-safety reminders across shared worktrees/sessions.
Under the hood
Organization management API surface expanded (orgs, workspaces, members, invites, service accounts, API/external keys, rate limits, federation issuers/rules) in support of Console/federation auth flows.
Experimental/opt-in work toward artifact preview, CCR request-body gzip, lazy CCR subagent hydrate, remote-tools session channels, and related session restore / background-shell hardening (env and feature-gate controlled).
SDK: gitOperation on command results is documented as a stable client-facing classification (commit/push/merge/rebase/PR) for UI without re-parsing stdout; internal harness hand-back bookkeeping fields were removed from AgentOutput.
Background-task level snapshots can include an ambient membership flag and still snapshot on repeated initialize after reconnect.
SDK compatibility matrix no longer lists wrapper 0.3.209 among tested versions (0.3.210+ remain).
Upgrade notes
If your org disallows API keys, try Sign in with your Console account under /login.
Expect sandboxed network access to prompt for hosts that previously looked “blocked” only because they weren’t pre-listed.
Skills pulled from claude.ai may show sanitized labels and will not override same-named local commands, MCP prompts, ! shell expansion, or @ file expansion on the machine.
CI/automation setups can move to federation rules + service accounts instead of interactive /login; regular (non-admin) API keys are not used on the organization admin endpoints.
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.
