AI pair programmers and autonomous coding agents — IDE-integrated assistants and cloud agents that read, edit, and ship code.
OpenAI's GPT-6.1 Sol became the default engine for Codex, Claude Sonnet 5.5 arrived with a 1M context window, and Devin pushed deeper into Jira and cross-device workflows — while a heavy week of fixes hardened every agent's terminal and sandbox.
Claude Code closed sandbox deny-rule bypasses that could let approved mods or environment-variable prefixes override Bash and Read restrictions, and fixed terminal freezes on malformed code blocks.
Several fixes harden sandbox rule enforcement: deny or ask rules on nested parts of compound shell commands no longer get overridden by a user-installed mod's approval, and Bash deny/ask rules are no longer skipped under sandbox auto-allow when an environment variable prefix or bare variable assignment precedes the command. Also fixed terminal freezes on short code blocks with many unclosed script tags or nested substitutions, and reverted the 2.1.288 claude auth status change that may have made sign-outs more frequent.
Claude Code stopped a dangerous rm from running unprompted in bypassPermissions mode and now re-prompts for OAuth scope when an MCP server requests more during a tool call, while Devin became a native Jira agent and stopped losing messages dropped mid-connection.
Devin is now available as a native agent inside Jira, starting sessions when a work item is assigned or mentioned, and automations can run a preflight script after each trigger before the agent starts. Messages sent right before a connection drop are now saved by the server instead of being lost, and Devin Review findings on PRs opened by a running session go to that session for auto-fix before being posted as review comments.
Fixed a dangerous rm (such as one on / or the home directory) inside a bash -c or sh -c script running without a prompt in bypassPermissions mode or under a shell allow rule, and added a re-authenticate prompt when an MCP server asks for more OAuth scope during a tool call. Also added draft recovery after Ctrl+C via Up on the empty prompt, and fixed resume dropping context restored by compaction or cutting transcripts short.
Claude Code introduced Claude Mods, letting plugins alter deeper behavior, alongside a built-in "You should know" side agent that flags things you or Claude might miss.
Claude Mods let plugins modify deeper behavior, with a built-in mod "You should know" where a side agent flags things you or Claude might miss. Added URL prompts from MCP servers on the 2025-11-25 protocol, plus fixes for Remote Control reconnects, repeated model fallback in claude -p sessions, and MCP connectors double-running calls.
Claude Code patched two serious flaws — sessions losing all turns after parallel tool calls when a prior session crashed, and several redaction gaps that could expose credentials in MCP errors, logs, and URLs — alongside smaller permission-prompt and fullscreen list improvements.
The terminal UI now preserves server provider, reasoning-summary, and verbosity settings and shows correct sessions in resume and fork history, and queued messages resume after reconnection without duplicate sends. Also added opt-in Guardian review context retrieval and a keyboard-accessible "Show more" action for browsing older tasks.
Fixed claude --resume and --continue sometimes losing every turn after a batch of parallel tool calls when the earlier session crashed or was killed, and closed several secret-redaction gaps where MCP error messages, percent-encoded Bearer tokens, redacted logs, and URL passwords could expose credential values. Also fixed repeated login browser windows when gcpAuthRefresh or awsAuthRefresh credentials expire, API 400 errors after a tool returned a non-text value, and stale Remote Control sessions, plus permission-prompt counter and fullscreen list improvements.
Claude Code shipped a --desktop flag for opening the desktop app on a directory or resumed session, an allowedProviders managed setting restricting which API providers a machine may use, plus fixes for late cloud-session replies and an Artifact allow rule that permitted publishes outside working directories.
xAI1Permission rules with relative paths now apply when the working directory contains symlinks, and video cards no longer get stuck extracting posters without ffmpeg. Also faster session startup when loading skills from large directories, plus fixes to MCP server source reporting, client rule preservation on system prompt rebuilds, and slow list_dir subtree counts on network filesystems.
Eligible local sessions signed in with ChatGPT can now show optional reminders to complete account security setup.
Added claude --desktop to open the Claude desktop app on the current directory or a resumed session, plus a CLAUDE_CODE_DISABLE_WEB_FETCH variable and an allowedProviders managed setting that limits which API providers a machine may use. Fixed a regression causing cloud-session first replies to arrive late, an Artifact allow rule permitting publishes outside working directories, and a rare auth failure when two sessions recover a login refresh lock left by a crashed process.
Claude Code made Claude Sonnet 5.5 its default model with 1M context at $2/$10 per Mtok and hardened stream and MCP failure handling, while Devin synced view preferences across devices and added keyboard session navigation.
xAI1Custom agents from plugins or config can now be chosen directly with spawn_subagent, and MCP servers can use a token file that is re-read on every request so rotating credentials stay fresh. Also added colored notice banners above the prompt for selected models and a "Waiting for N subagents…" status when the parent turn is blocked.
Suppressed console windows flashing on Windows when Codex launches background processes and sandboxed commands.
GPT-6.1 Sol is now the default model in the bundled catalog, and the Amazon Bedrock Mantle and Runtime catalogs add it via a backport.
GPT-6.1 Sol is available via gpt-6.1-sol, offering near-Astra performance for complex, long-running work across code, apps, and documents at lower cost than Astra. Availability depends on plan, client, and workspace settings.
Automatic follow-up prompt suggestions and the tui.prompt_suggestions setting were removed, and the bundled plugin-creator skill is gone. Opt-in instant_interrupt lets new input steer Codex during model responses or long-running code-mode calls, with fixes for Windows console windows, transcript copy formatting, and macOS TLS in network-enabled sandboxes.
Claude Sonnet 5.5 (claude-sonnet-5-5) is now the default Sonnet model on the Anthropic API, with 1M context and $2/$10 per Mtok pricing. Damaged response streams are now retried or reported as interrupted instead of surfacing raw JSON parse errors or writing "undefined" into an answer, and MCP tool calls in resumed sessions wait up to 10 seconds for a still-connecting server instead of failing with "No such tool available".
View preferences — search mode, changed-files tree layout, sidebar Review/Ask filters, and wiki language/view choices — now sync across browsers and devices instead of staying on one machine. Also added Option+J/Option+K session navigation, interactive HTML reports, child sessions that carry a summary of their parent, and Bitbucket Data Center and Azure DevOps pipeline log support.
OpenAI brought GPT-6 Sol and Luna to Codex and ChatGPT Work at lower token prices than GPT-5.6, while Cursor launched deploy-monitoring and security-review bots and Devin reworked how automatic reviews get triggered.
Read the digest →Codex CLI now requires terminal input approval by default for commands running with elevated permissions, while gaining MCP OAuth client secrets, exec-server bearer-token auth, and TUI clipboard improvements.
xAI1A new /context-window slash command lets you pick a context window size for the current session when the model offers choices, and /model selection now offers context window choices before effort level. Pasting on Windows no longer crashes from overlapping clipboard operations, and parallel tool calls that edit the same file now run in sequence.
Terminal input approval is now enabled by default for commands running with elevated permissions, and runtime-only grants no longer trigger unnecessary reviews. Also added MCP server support for pre-registered OAuth client secrets, bearer-token auth for exec-server WebSockets, copy-on-select and right-click paste in the fullscreen TUI, and fixes for Windows and Linux sandbox failures.
xAI1MCP tool prompts now appear correctly in minimal mode, and users are notified when a prompt goes unanswered.
Claude Code shipped managed settings that let admins pin allowed model versions and deny specific models outright, alongside a /doctor prompt audit for stale CLAUDE.md and skill files, plus fixes for SDK sessions dropping deferred tool calls and a PowerShell escape that could delete drive roots.
xAI1New grok worktree create command creates a managed worktree without launching an interactive session, and plugin hooks now run on a fresh session without an explicit reload. Fixed bracketed pastes attaching leftover images, grok -p not exiting cleanly on interrupt, inline images appearing blank after repaint, and Ctrl+C twice while writing a plan comment getting stuck.
Codex
Claude Code
Devin
Grok Build