AI coding agent for the terminal and IDE
npx @buildinternet/releases get claude-codeFixed a 2.1.265 regression where the undocumented CLAUDE_CODE_USE_GATEWAY environment variable forced Cloud-gateway sign-in on its own, breaking requests for LLM-gateway and proxy configurations using an API key, apiKeyHelper, or custom auth headers. The variable is ignored again on its own; no configuration change is needed.
Added support for pointing --plugin-dir at a folder of plugins, a 1 GB cap on tool results saved to disk, and user.email/user.groups in telemetry. Fixed dozens of issues including broken prompt-cache reuse for resumed subagents, resume after a crashed process losing the interrupted tool call, and a symlink check bypass on macOS and Linux.
Added bashOutputMaxChars and taskOutputMaxChars settings to raise inline command output up to 128K, plus /skill-doctor to identify unused loaded skills. Fixed Remote Control showing stale permission mode, stuck spinners after stopping a turn, and other session-state issues, along with dozens more bug fixes and improvements across Bedrock, VS Code, and CLI.
Fixed several sandbox permission holes, including Edit/Write/Read rules with parentheses being dropped as invalid (leaving read-only folders writable), zsh REPORTTIME/REPORTMEMORY/DIRSTACKSIZE assignments auto-approving command substitutions, and uncompilable patterns causing every file edit to fail. Reverted the 2.1.259 change that applied Read() deny rules to Bash arguments. Also added a fullscreen diff panel (/diff), text form of /advisor for headless sessions, /reload-plugins in headless mode, and multiple fixes for rewind, model switching, and prompting on Fable.
Organizations can now provide HTTP/SSE MCP servers to all users via the managedMcpServers setting. Fixed concurrent sessions silently reverting each other's ~/.claude.json changes, an OAuth-refresh prompt cache invalidation, and a sleep/wake OAuth race that could log out sessions.
Fixed Claude Code failing to launch on macOS 12 (Monterey), a regression introduced in 2.1.255. Also fixed remote and scheduled sessions failing with "user messages must have non-empty content" after a re-sent permission approval could not be applied.
Added Claude Fable 5.1 (claude-fable-5-1) — now the default Fable model with 1M context — plus a Containment Escape rule that stops auto-approving cloud credential fetches, egress evasion, and cross-tenant reach. Fixes include unbounded memory growth from piped non-JSONL data, an OAuth credential leak risk, and dozens of smaller fixes across subagents, time formatting, and sandboxing.
Fixed Bash commands failing with "task output swap refused" on some Macs when the tasks directory was moved or linked, and resolved Remote Control sessions stalling for minutes after a tool finished under degraded claude.ai connections. Also fixed "always allow" not saving in projects without a settings.local.json and background task notifications with very large failure output exceeding API limits.
Added PreModelSwitch and PostModelSwitch hook events, live streaming of foreground subagent tool calls to Remote Control clients, and a Spend limit bar to /usage. Fixed multiple security issues including file tools following swapped symlinks, plugin commands with path traversal, and beta tracing bypasses, plus dozens of correctness fixes across sessions, MCP, and teams.
Release notes describe general bug fixes and reliability improvements without specifying individual changes.
Added --restricted mode that removes command/code tools and WebFetch (unless named), keeps file tools inside the working directory, refuses bypassPermissions, and ignores settings files. Fixed a prompt-cache miss (with lost extended-thinking context) roughly once an hour in long sessions caused by tool definitions re-rendering after OAuth token refresh, plus dozens of smaller fixes across agents, hooks, MCP, and remote control.
Added /claude-api cost-optimize to profile and reduce a project's Claude API spend, plus the SendFeedback tool for drafting feedback reports. Fixed sub-agents failing on model 404s (now use the session's fallback chain), unbounded memory growth when hook output files couldn't be written, and a dotfile-managed settings symlink being deleted by Bash sandbox cleanup. Also hardened plugin marketplace against control/invisible characters and improved MCP failure reporting in Bedrock, Vertex, and Foundry sessions.
Fixed telemetry and metrics requests to Anthropic carrying the API key for a third-party gateway, so a credential is only sent to its own host. Also fixed fullscreen blank-transcript and scrolling regressions after resize, a transcript slowdown on very long lines, and dozens of other fixes spanning MCP, plugins, background sessions, and Bash tooling, plus an Auto mode tab in /permissions.
Fixed a crash on startup for Linux distributions with glibc 2.44, such as Arch Linux, CachyOS, and Fedora Rawhide.
Fixed a Bedrock streaming bug behind proxies that stripped response Content-Type headers, silently doubling billed API calls, plus a startup hang with SSO profiles behind HTTPS proxies. Added /claude-api upgrade to migrate Python projects from anthropic 0.x to 1.x, a one-time fullscreen renderer offer on Bedrock/Vertex/Foundry setups, and cost estimates that now include the 1.1× US-only inference premium for data-residency workspaces. Another four dozen fixes span vim mode, fullscreen rendering, remote sessions, and /resume behavior.
Fixed unbounded memory growth in long interactive sessions by releasing subagent tool results once they leave the recent display window, and added a keybindingFlavor setting with a "readline" option for Bash-style Ctrl+W. Also fixed output styles drifting back to default voice mid-session and dozens of Remote Control, MCP, and cross-session messaging fixes.
Fixed prompt caching for sessions using an LLM gateway or custom base URL. Added a built-in "Concise" output style that leads with results and skips preamble, selectable under Output style in /config.