Anthropic shipped Claude Code 2.1.72 with 54 CLI changes, including ExitWorktree, direct /copy writes, and fixes that cut SDK query input token costs by up to 12x. Teams using long sessions or remote shells should upgrade and review the new environment variables and effort-level changes.
/copy writes, mid-session ExitWorktree, and a kill switch for scheduled cron jobs, according to the changelog thread.query() prompt-cache bug was fixed, cutting input token costs by “up to 12x” in affected cases./clear killing background tasks, stuck “Initializing…” states, and several permission-rule mismatches, as detailed in the release notes.vscode://anthropic.claude-code/open⌘FThe release is broad rather than flashy. Per the changelog thread, new workflow features include a w key in /copy to write the focused selection straight to a file “bypassing the clipboard,” an optional description argument for /plan, an ExitWorktree tool to leave an EnterWorktree session without ending the whole run, and support for marketplace Git URLs that omit the .git suffix. For remote and SSH-heavy setups, that direct write path is one of the clearest quality-of-life upgrades.
Version 2.1.72 also expands the config surface. The follow-up diff lists new environment variables including CLAUDE_CODE_DISABLE_CRON, CLAUDE_CODE_BRIEF, and CLAUDE_CODE_DISABLE_VIRTUAL_SCROLL, while the official changelog notes a new VS Code URI handler for opening a Claude Code tab programmatically with optional prompt and session parameters. The same release simplifies effort levels to low, medium, and high, removes max, and swaps the persistent icon for a brief notification.
The biggest operational change is the fix for prompt-cache invalidation in SDK query() calls. The release notes say that bug had been inflating input usage, and the fix can reduce token cost “up to 12x.” The release also swaps bash parsing to a native module for faster initialization and “no memory leak,” reduces false-positive permission prompts through better tree-sitter parsing, and fixes sandbox cases where some writes were wrongly allowed while harmless redirects still prompted.
Session reliability got a sizable cleanup too. The changelog fixes slow exits when hooks or background tasks lagged, prevents /clear from killing background agent and bash jobs, restores proper --continue behavior after --compact, and fixes crashes tied to Bedrock API errors and unusual Unicode returned by Read. The system-prompt diff in the prompt analysis also suggests a more explicit agent model, including concurrent launches, resume-by-ID, and worktree isolation rules. One practitioner complaint remains unresolved: a user screenshot shows Opus 4.6 in Claude Code auto-compacting with “Context left until auto-compact: 0%,” arguing that a 200k window is still too small for large code sessions.
Claude can now drive macOS apps, browser tabs, the keyboard, and the mouse from Claude Cowork and Claude Code, with permission prompts when it needs direct screen access. That makes legacy desktop workflows automatable, and Anthropic is pairing the push with more background-task support for longer agent loops.
releaseOpenClaw shipped version 2026.3.22 with ClawHub, OpenShell plus SSH sandboxes, side-question flows, and more search and model options, then followed with a 2026.3.23 patch. Teams get a broader plugin surface, but should patch quickly and review plugin trust boundaries as the ecosystem grows.
releaseCursor shipped Instant Grep, a local regex index built from n-grams, inverted indexes, and Bloom filters that drops large-repo searches from seconds to milliseconds. Faster candidate retrieval shortens the coding-agent loop, especially when ripgrep-style scans become the bottleneck.
breakingChatGPT now saves uploaded and generated files into an account-level Library that can be reused across conversations from the web sidebar or recent-files picker. It removes repetitive re-uploading and makes past PDFs, spreadsheets, and images part of a persistent working context.
breakingEpoch AI says GPT-5.4 Pro elicited a publishable solution to one 2019 conjecture in its FrontierMath Open Problems set, with a formal writeup planned. Treat it as an early milestone worth reproducing, not blanket evidence that frontier models can already automate math research.
⌘F in terminal coming soon to a conductor near you
Claude Code 2.1.72 further changes Flags: Added: • tengu_kairos_brief — Enables a user-visible “brief” message (markdown-supported), configurable via flag or CLAUDE_CODE_BRIEF.; notes: Context suggests a displayed message, but the exact UI surface/location isn’t shown in the Show more
Claude Code 2.1.72 further changes Flags: Added: • tengu_kairos_brief — Enables a user-visible “brief” message (markdown-supported), configurable via flag or CLAUDE_CODE_BRIEF.; notes: Context suggests a displayed message, but the exact UI surface/location isn’t shown in the Show more
Claude Code 2.1.72 has been released. 2 flag changes, 54 CLI changes, 5 system prompt changes Highlights: • Added w key in /copy to write focused selection directly to a file, bypassing clipboard • Exit EnterWorktree sessions mid-session via ExitWorktree, keeping or removing Show more
