releases.shpreview
Home/Collections/Coding Agents

Coding Agents

AI pair programmers and autonomous coding agents — IDE-integrated assistants and cloud agents that read, edit, and ship code.

MonJul 6, 20261 release
Cognition
Devin

Slack !agent renamed to !normal; scanning remediation API

The Slack !agent bang-command is now !normal. The v3 API supports remediating security scan findings, service users can manage automations programmatically, and blueprints can be Git-backed. Bug fixes include virtualized Review comments hijacking scroll and desktop VNC reconnection after session wake.

Read post
FriJul 3, 20262 releases
Anthropic
Daily summary

Claude Code switches to manual permissions by default

Claude Code changed its default permission mode to Manual across all interfaces and stopped auto-continuing on user approval dialogs, while also fixing background sessions silently stopping after sleep/wake.

  • Claude Code now defaults to Manual permission mode across CLI, VS Code, and JetBrains, requiring explicit user approval for each action
  • AskUserQuestion dialogs no longer auto-continue by default; users can opt into an idle timeout via /config
  • Background sessions no longer silently stop after sleep/wake or when reopening a stalled session, with dozens of reliability fixes

Claude Code

2 updates

Claude Sonnet 5 sessions no longer use the mid-conversation system role for harness reminders.

Read post

Changed AskUserQuestion dialogs to no longer auto-continue by default. Changed the default permission mode across CLI, VS Code, and JetBrains to Manual, with new accepted config keys. Fixed background sessions silently stopping after sleep/wake, a startup crash with invalid MCP server config arrays, and dozens of background-agent reliability fixes.

Read post
ThuJul 2, 20262 releases
AnthropicCognition
Daily summary

Background agents stabilized, Devin syncs Slack

Claude Code fixed a daemon self-kill loop that killed background agents every 50 seconds, while Devin shipped bidirectional Slack thread sync and read-only MCP.

  • Claude Code stopped a background-agent daemon bug on Linux that killed every running agent every ~50 seconds after an unclean shutdown
  • Devin added bidirectional Slack thread sync and a read-only MCP toggle alongside PR ratio analytics with CSV export
Claude Code

Background-agent daemon self-kill loop fixed; stacked skills load up to 5

v2.1.199

Fixed a background-agent daemon bug on Linux that killed every running agent every ~50 seconds after an unclean shutdown. Stacked slash-skill invocations now load up to 5 leading skills, SSL certificate errors fail immediately with actionable guidance, and mid-stream API errors keep partial output. Plus dozens of smaller fixes across subagents, hooks, and session management.

Read post
Devin

Slack thread sync; MCP read-only mode; PR ratio analytics

Adds bidirectional Slack thread sync, a read-only MCP toggle, and a weekly PR ratio analytics chart with CSV export. Also includes redesigned environment page, cross-platform repository cloning, mention-to-unarchive, inline mute reminders, and numerous bug fixes across the platform.

Read post
WedJul 1, 20261 release
Anthropic
Daily summary

Claude Code brings Claude in Chrome to GA with resilience improvements

Claude in Chrome is now generally available, brief network drops mid-response are now retried with backoff, and background agents gained desktop notifications.

  • Claude Code makes Claude in Chrome generally available
  • Transient network errors like ECONNRESET are now retried with backoff instead of aborting the turn
  • Background agents now fire desktop notifications when sessions need input or finish work
  • Added a /dataviz skill for chart and dashboard design guidance with a runnable color-palette validator
  • Gateway adds Claude Platform on AWS as an upstream provider, with improved failover on model-not-found errors
Claude Code

Claude in Chrome GA; network drops retry instead of aborting

v2.1.198

Claude in Chrome is now generally available. Fixed brief network drops mid-response aborting the turn — transient errors like ECONNRESET now retry with backoff instead of failing. Also added background agent notifications, a /dataviz skill, and dozens of additional fixes across agents, worktrees, and fullscreen mode.

Read post
TueJun 30, 20262 releases
OpenAIAnthropic
Daily summary

Claude Code defaults to Sonnet 5 with 1M-token context

Claude Code switched its default model to Claude Sonnet 5, bringing a native 1M-token context window and promotional pricing through August.

  • Claude Code now defaults to Claude Sonnet 5, offering a native 1M-token context window and promotional pricing of $2/$10 per Mtok through August 31
MonJun 29, 20262 releases
CursorAnthropic
Daily summary

Claude Code fixes session loss on wake and stops unchecked MCP server spawning

A critical bug where waking a background job deleted the conversation and reran the original prompt was patched, while security changes prevent repo-approved MCP servers from auto-spawning in untrusted workspaces.

  • Claude Code fixes a bug where waking a background job lost the conversation and replayed the original prompt
  • Security: Claude Code no longer spawns repo-approved .mcp.json servers in untrusted workspaces, showing them as pending approval instead
  • Admins can now set organization default models visible in the model picker, and sessions get readable default names at creation
  • Chat file attachments are now clickable to reveal the file in Finder or Explorer
Claude Code

Session loss on background job wake fixed; MCP server spawns hardened

v2.1.196

Fixed a critical bug where waking a background job deleted the conversation and reran the original prompt. Security: claude mcp list/get no longer spawns repo-approved .mcp.json servers in untrusted workspaces. Also fixes OAuth scope rejection on enterprise IdPs, rate-limit flickering, and PowerShell exit-code false positives.

Read post
SunJun 28, 20262 releases
OpenAICursor
Daily summary

Cursor brings cloud agents to iOS

Cursor launched a public beta iOS app that lets paid users launch and manage cloud agents from their phone, including remote control of desktop sessions.

  • Cursor’s new iOS app lets paid users launch cloud agents directly from their phone, with voice input and slash commands for guiding work
  • Remote Control feature allows directing an active desktop agent from the lock screen via Live Activities
SatJun 27, 20262 releases
xAICognition
Daily summary

Devin adds run-once automations and archive from command palette

Devin shipped several quality-of-life improvements, including the ability to archive or unarchive sessions from the Cmd+K palette, one-time run-once automation schedules, and an out-of-date banner for MCP integrations.

  • Devin sessions can now be archived or unarchived directly from the Cmd+K command palette, skipping the settings navigation
  • Automations support a run-once schedule option for one-time executions alongside recurring schedules
  • Installed MCP integrations show an "Installation out of date" banner with one-click refresh when an update is available
  • Blueprint editor adds a Post Build section for configuring steps after the build phase
Grok Build

Grok Build 0.2.73

v0.2.73
  • New "Keep text selection highlight" setting keeps drag selections visible until dismissed
  • Doubled lines after tab switches or focus changes in tmu...
Read post
FriJun 26, 20264 releases
xAIAnthropic
Daily summary

Claude Code fixes voice dictation and hyphenated hook matching

Claude Code patched voice dictation silence capture and auto-submit for space-less languages, fixed hyphenated hook matchers to exact-match, and added a mouse-click disable toggle.

  • Claude Code now exact-matches hyphenated hook identifiers like code-reviewer instead of substring-matching, requiring mcp__brave-search__.* to match all tools from a hyphenated MCP server
  • Voice dictation on macOS no longer captures silence after input device changes, and auto-submit now works for Japanese, Chinese, and Thai
  • A new CLAUDE_CODE_DISABLE_MOUSE_CLICKS env var disables mouse interactions in fullscreen mode while preserving scroll

Grok Build

3 updates
  • No longer triggers browser login at startup when an API key is already configured for inference
  • Added grok wrap to run any command with local clipboard support
  • Ctrl+4 now toggles the prompt queue on local macOS VS Code terminals
  • Session...
Read post
  • Fixed grok agent stdio hangs on Windows when used with persistent clients such as VS Code
Claude Code

Hyphenated hooks now exact-match; voice silence gap fixed

v2.1.195

Hook matchers with hyphenated identifiers now perform exact-matching instead of substring-matching, use mcp__brave-search__.* to match tools from hyphenated MCP servers. Fixed voice dictation capturing silence after input device changes and auto-submit not firing for languages written without spaces, plus fixes for background task data loss and crashed-task blank screen.

Read post
ThuJun 25, 20262 releases
OpenAIxAI
Daily summary

Claude Code adds OTEL response logging and full-shell auto-mode classification

Claude Code now logs assistant responses via OpenTelemetry by default for deployments already logging prompts, and gains a setting to route all shell commands through the auto-mode classifier instead of only high-risk patterns.

  • Claude Code logs assistant responses via OpenTelemetry when user prompts are already logged, toggleable via an environment variable
  • A new autoMode.classifyAllShell setting subjects every Bash or PowerShell command to classification, not just arbitrary-code-execution patterns
  • Auto-mode denial reasons now appear in the transcript, a denial toast, and the /permissions recent-denials view
Grok Build

Grok Build 0.2.68

v0.2.68
  • MCP servers from host integrations can now be added, replaced, or removed without restarting the session
  • Agent-run terminal commands now set `GROK...
Read post