lab Claude Code Releases · 1h ago · agents

Claude Code v2.1.218 changes /code-review to background subagent

from English

Claude Code v2.1.218 updates the `/code-review` command to run as a background subagent, preventing review work from filling the conversation and allowing stacked slash commands to use it as their target.

Added screen-reader announcements for deleted text in `--ax-screen-reader` mode.
Fixed Windows paths with `\`-prefixed segments being corrupted into CJK characters.
Fixed `/ultrareview` failing on descriptive arguments by applying the text as a note to findings.
Changed `/deep-research` to start only when invoked manually, removing automatic launches.
Improved auto mode so dangerous checks no longer open permission dialogs but are adjudicated by the classifier.
Fixed gateway spend metering for Bedrock application-inference-profile ARNs and other config-mapped upstream model IDs.

These changes improve workflow continuity by keeping the main conversation clear and enhance reliability through fixes for path corruption, screen-reader accessibility, and session stability.

Importance 1/3 Trust 2/3 Claude Code Releases Anthropic AI agents Code generation

Read original

lab Claude Code Releases · 1d ago

Claude Code v2.1.217 fixes memory leaks, auto-update failures, and adds subagent concurrency caps

Claude Code version 2.1.217 introduces several bug fixes and configuration changes, including emoji shortcode autocomplete in the prompt input and warnings for failing transcript writes. The update addresses critical stability issues such as a memory leak involving truncated MCP tool outputs and Windows auto-update failures that could leave claude.exe missing.

lab Claude Code Releases · 2d ago

Claude Code v2.1.216 fixes quadratic slowdown and adds sandbox filesystem setting

Claude Code version 2.1.216 addresses a critical performance regression where message normalization cost grew quadratically with the number of turns, causing multi-second stalls in long sessions. The update also introduces a new `sandbox.filesystem.disabled` setting that allows users to skip filesystem isolation while retaining network egress control.

lab Claude Code Releases · 4d ago · 3 views

Claude Code v2.1.215 makes /verify and /code-review skills manual

Claude Code version 2.1.215 updates the behavior of the /verify and /code-review skills so that they are no longer executed automatically.

lab Claude Code Releases · 5d ago · 2 views

Claude Code v2.1.214 fixes permission bypasses and adds EndConversation tool

Claude Code version 2.1.214 addresses critical security vulnerabilities, including a permission-check bypass in Windows PowerShell 5.1 and issues with Bash permission checks. The update also introduces the EndConversation tool, allowing Claude to terminate sessions with abusive users or those attempting jailbreaks.

media TLDR AI · 5d ago

Thinking Machines releases Inkling; Anthropic prepares IPO; OpenAI trains GPT-Red

Thinking Machines has released Inkling, a 975B-parameter mixture-of-experts model with 41B active parameters and multimodal reasoning. Simultaneously, Anthropic is preparing for a potential IPO later this year following a $65 billion funding round at a $965 billion valuation. OpenAI has also trained GPT-Red to iteratively generate adversarial prompts, which reduced failures on a prompt-injection benchmark by sixfold for GPT-5.6 Sol.