AnthropicFigmaLinearVercelStripeShopifyPostHogDatadogRampSeatGeekSierraHexTurbopufferAnthropicFigmaLinearVercelStripeShopifyPostHogDatadogRampSeatGeekSierraHexTurbopuffer
NVDA$1,234.56+1.23%
NVDA$1,234.56+1.23%
NVDA$1,234.56+1.23%
NVDA$1,234.56+1.23%
NVDA$1,234.56+1.23%
NVDA$1,234.56+1.23%
NVDA$1,234.56+1.23%
NVDA$1,234.56+1.23%
NVDA$1,234.56+1.23%
NVDA$1,234.56+1.23%
The Claude Code Daily
Code with Claude SF recap

Code With Claude San Francisco Event Wraps Up

The in-person Code with Claude San Francisco event has concluded. The official account thanked both attendees and remote viewers.

ClaudeDevs·Thu, May 7 8:50pm ET

60+ More Claude Code Reliability Fixes Ship This Week

Following last week's 50-fix batch, another 60+ reliability improvements landed covering long-running sessions, the agent loop, auth edge cases, and terminal rendering. The thread breaks them into categories.

ClaudeDevs·Fri, May 8 11:18am ET

Stability Fixes: Stdin, Sleep Resume, Memory Leak, Output

`claude -p` now handles more than 10MB of piped stdin. Requests resume cleanly after a Mac wakes from sleep, a stdio MCP server writing garbage to stdout no longer causes memory to balloon past 10GB, and output appears reliably after thinking completes.

ClaudeDevs·Fri, May 8 11:18am ET

Agent Loop Gets Caching, Parallelism, and Context Fixes

Sub-agent summaries now hit the prompt cache, opt-in 1-hour prompt caching is honored correctly, parallel shell calls keep running when a read-only sibling fails, and 1M-context sessions finally use their full window before hitting the "Prompt is too long" error.

ClaudeDevs·Fri, May 8 11:18am ET

New /radio Slash Command Debuts

A new /radio slash command has appeared, shared with an accompanying image. Details are sparse, but the command appears to be a fresh addition to the Claude Code slash command roster.

ClaudeDevs·Fri, May 8 2:29pm ET

Claude Code User Ditches Markdown for HTML

A workflow shift: using Claude Code to generate HTML instead of writing markdown files, with the argument that HTML is simply better for almost everything markdown is used for. The tweet includes a visual explanation.

trq212·Fri, May 8 2:03pm ET
More Stories

Long Sessions Slow Down Due to Token Accumulation

Response speed degrades as a session grows because more tokens get sent with each request. The /fast command is available for users willing to pay more, though switching back is awkward. Plan mode shouldn't affect speed independently.

jarredsumner·Thu, May 7 5:42pm ET

Prompt Injection Mitigation Behind Code-Against-Pages Limit

The current restriction on writing code against web pages is a prompt injection mitigation, not a permanent design choice. The team acknowledges it needs revisiting, and sees coding against pages as a genuinely promising direction.

trq212·Fri, May 8 2:40pm ET

OAuth and Login Now Work in WSL2, SSH, Containers

You can now paste the OAuth code directly into the terminal when the browser can't reach localhost, covering WSL2, SSH, and container environments. Login also works on slow proxies and IPv6-only devcontainers, and refresh tokens are protected against a rare concurrent-write race.

ClaudeDevs·Fri, May 8 11:18am ET

MCP Tool Listing, Images, and Reconnection Fixed

Servers that connect but fail tool listing now retry and surface a clear status in `/mcp`. Images in tool results are no longer silently dropped, and reconnecting servers announce a summary rather than re-dumping their full tool list.

ClaudeDevs·Fri, May 8 11:18am ET

Terminal Rendering Bugs Fixed Across Multiple IDEs

Scrolling is no longer too fast in Cursor, older VS Code, and JetBrains terminals. CJK text renders correctly on Windows in no-flicker mode, pasting text starting with `/` now goes to the prompt instead of misfiring, and Ctrl+L redraws the screen without losing input.

ClaudeDevs·Fri, May 8 11:18am ET
Archive

Get the daily digest in your inbox

Every weekday morning. Unsubscribe anytime.