ClaudeDevs
@ClaudeDevs
In the last four Claude Code CLI releases, we’ve shipped 50+ stability and performance fixes. Faster resume, stable auth, lower memory, fewer hangs: 🧵
20retweets487likes
View on XAnthropic Ships 50 Stability Fixes to Harden Claude Code Agentic Workflows
· Updated
Anthropic updated Claude Code, its terminal-based agentic coding tool, with version 2.1.121 to resolve persistent stability issues. The release fixes several memory leaks, including multi-gigabyte growth during image processing. It also improves Claude Code session mobility by making the resume command more resilient to corrupted transcript files.
This update follows a series of quality regressions that impacted the CLI during recent feature launches. While the models remained capable, orchestration-layer bugs like terminal scrollback duplication created friction. Hardening these mechanics moves the tool toward a dependable production environment for autonomous bug hunting.
You can apply these updates by running claude update in your terminal. The version includes an alwaysLoad option for MCP (Model Context Protocol) servers and adds certificate-based authentication for Vertex AI. These improvements are available to all subscribers, including those using session recaps to manage parallel coding tasks.
Still wondering? A few quick answers below.
To get the latest performance and stability improvements, you should run the command claude update in your terminal. This will install version 2.1.121, which includes over 50 fixes for memory management, terminal rendering, and connection reliability. Keeping the CLI updated ensures you have the most resilient version of the agentic coding tool for your local environment.
This update resolves several critical memory leaks, including unbounded growth that reached multiple gigabytes when processing images. It also fixes a 2GB leak associated with the usage command and a leak occurring when long-running tools fail to emit progress events. These changes prevent the CLI from consuming excessive system resources during extended, high-intensity agentic coding sessions.
The alwaysLoad option is a new configuration setting for Model Context Protocol servers. When set to true, it allows all tools from a specific server to skip the standard tool-search deferral process. This ensures those tools are always available to the model immediately, rather than being loaded only after a search identifies them as necessary for a task.
Yes, version 2.1.121 adds support for X.509 certificate-based Workload Identity Federation for users on Google Vertex AI. This allows for more secure, certificate-driven authentication using the mTLS Application Default Credentials standard. It is part of a broader effort to improve enterprise compatibility and connection stability for developers using third-party cloud providers to run Claude.
The update fixes scrollback duplication issues in terminals like tmux, GNOME Terminal, and Windows Terminal. It also improves fullscreen mode so that typing into the prompt no longer forces the view to snap back to the bottom if you have scrolled up to read earlier output. These fixes make it easier to review long conversation histories and tool results.
