Coding Agents Digest
What changed in AI coding agents, selected and explained

← Latest

📰

Check · Wed, Oct 28, 2026

Claude Code 2.1.312 stops sending a safety OpenTelemetry event and changes model switching; Codex 0.165.0 turns memory v2 on for new profiles.

08:11Claude Code, Codex CLI, OpenCode · 3 releases, 28 bug fixes
Claude Code
  • ⚠️ Safety event no longer sent (2.1.312). Claude Code stops sending the claude_code.safety_event OpenTelemetry event, so change any alert or dashboard that counts it. Source
  • ⚠️ Model switches cover custom model names (2.1.312). After a flagged message, Claude Code can now switch models for unrecognized names such as custom deployment names; set CLAUDE_CODE_DISABLE_REFUSAL_FALLBACK=1 to keep the old behavior. Source
  • Click to send in fullscreen (2.1.312). Clicking the “to send now” hint under a queued message sends it right away, as its key does. Source
Codex CLI
  • Memory v2 on for new profiles (rust-v0.165.0). Fresh profiles start with memory v2, which can import memories from other agents, so review what it keeps if that matters to you. Source
  • Built-in Vertex AI provider (rust-v0.165.0). Set model_provider = "gcp" to use Google Cloud Vertex AI. Source
OpenCode
  • No user-facing changes (v1.19.5). The one listed change is a CI fix: the v1 release job uses GITHUB_TOKEN instead of the OpenCode app token. Source
Across tools
  • Defaults change again. Codex turns memory v2 on for new profiles, and Claude Code widens automatic model switches to custom model names, so both change behavior without a new command. Sources: Codex CLI, Claude Code
Antigravity CLI · no updatesKilo Code · no updates