Claude Code updates 2.1.79 with /remote-control, Console auth, and stricter memory saving

• Anthropic shipped Claude Code 2.1.79 one day after 2.1.78, adding claude auth login --console for Anthropic Console billing, a 2-minute timeout on non-streaming fallback requests, and about 18 MB lower startup memory use, according to the 2.1.79 changelog and the prompt update thread.
• The biggest behavior change is memory handling: Claude Code now enforces “do not save” rules even when asked, storing only “surprising” or “non-obvious” durable insights instead of raw PR lists or activity logs, as shown in the system prompt diff.
• Anthropic also slipped in /remote-control for VS Code, which the release thread says can “bridge your session” to a browser or phone; supporting posts show Dispatch rolling out to Pro users and expanding to launch Claude Code sessions from mobile Dispatch rollout Code from Dispatch.
• This follows 2.1.78’s line-by-line response streaming, StopFailure hooks for API errors, and fixes for large-session resume, permission leaks, and sandbox warnings, making the last two Claude Code releases a meaningful workflow update rather than a routine patch 2.1.78 changelog 2.1.78 extras.

Claude Code 2.1.79 is a small version bump with real operational changes. The release adds --console to claude auth login, so teams using Anthropic Console billing no longer need to route everything through the older auth path, and it adds a “Show turn duration” toggle in /config release thread. The same changelog says non-streaming fallback now uses “a 2-minute per-attempt timeout,” which matters for wrappers that previously risked hanging sessions indefinitely.

The reliability fixes are concentrated in places engineers actually hit: claude -p hanging when spawned without explicit stdin, Ctrl+C not working in print mode, enterprise users being unable to retry on 429s, and SessionEnd hooks not firing when /resume switches sessions release thread. The companion diff thread also notes new CLI surface for --claudeai and --console, plus CLAUDE_CODE_SUBPROCESS_ENV_SCRUB and several new config keys, while estimating only a modest bundle increase of 26.1 kB and 434 prettified lines of code extra diff thread release diff.

The most consequential change is in the system prompt. The prompt diff says these exclusions apply “even when the user explicitly asks,” and if someone asks to save a PR list or activity summary, Claude should ask what was “surprising” or “non-obvious” before writing memory system prompt diff. It also now treats memory as a time-stamped claim: before recommending a remembered file, function, or flag, it should verify that the path or symbol still exists memory verification.

Anthropic is also pushing Claude Code beyond the local terminal. The 2.1.79 notes add /remote-control in VS Code to “bridge your session” so work can continue from a browser or phone remote-control note. Separate rollout posts show the mobile side arriving under Cowork Dispatch for Claude Pro users, with an explicit warning that Claude “will access your desktop” and that users should pair only devices they trust mobile screenshots. A follow-on repost says Dispatch can now “launch Claude Code sessions,” and another says “Skills and a bunch of /slash commands” already work over mobile Remote Control launch sessions mobile commands.

For engineering teams, that means internal guidance now has to cover cross-device session handling, not just terminal usage. The same is true for model selection: a practitioner post points out that the Claude API can now list per-model capabilities such as context window, thinking mode, and context-management support programmatically through the models.list API endpoint, which makes it easier to generate model pickers and guardrails from live metadata instead of hardcoded tables capabilities listing.

This all lands immediately after 2.1.78, which added line-by-line streaming, a StopFailure hook for rate-limit and auth failures, persistent plugin data, and fixes for large-session resume truncation, sandbox misconfiguration visibility, and permission-rule leaks around MCP tools 2.1.78 changelog. The prior release also changed memory guidance by treating third-party uploads as effectively public and by telling Claude to trust current observations over stale memory snapshots prompt changes. Together, the two releases tighten memory persistence, widen remote control, and reduce failure modes in real agent workflows.