A focused set of quality improvements today: Kilo Code ships a pre-release with a new Fork Session button for Agent Manager and several JetBrains and UX polish fixes. NemoClaw merges two important reliability patches — automatic sandbox recovery after a host reboot and correct CDI GPU mode selection on Ubuntu 24.04+ systems. Anthropic's Claude apps changelog page also received an update.
Fork Session button in Agent Manager. Completed Agent Manager sessions now show a "Fork Session" button, letting you branch from any finished session directly from the session history view without restarting the task from scratch.
Timeline pinning and auto-scroll improvements. Task timelines now stay in place while you review earlier activity across Kilo chat surfaces — the view no longer snaps back to the latest bar while you are reading history, then resumes following updates once you return to the bottom. Chat auto-scroll also now pauses correctly whenever you scroll upward through nested tool output, preventing the display from jumping away mid-read.
Improved JetBrains reasoning blocks. Reasoning blocks in the JetBrains plugin now stream as they arrive, automatically collapse once the model finishes reasoning, hide completely when empty, and render adjacent reasoning steps as a single block instead of separate stacked panels.
Release notes →
Affects: /kilocode/, /kilocode/setup/, /kilocode/models/, /kilocode/orchestrator/, /kilocode/security/, /kilocode/vs-claude-code/
2026-06-10
NemoClaw
2026-06-10 commit batch
NemoClaw merged a large batch of commits on June 10. Most are internal test-infrastructure refactors (migrating legacy bash E2E tests to Vitest) and onboarding code cleanups with no user-visible effect. Two commits fix real-world reliability issues:
Sandbox auto-recovery after host reboot (#4423): Previously, if your sandbox was running before a host reboot, running nemoclaw <name> status after restart would print a "retry in a moment" message and leave the sandbox in a broken state. NemoClaw now walks Docker by OpenShell labels, restarts the stopped container (or renames a GPU-patch backup container back to the original name), and returns a fully live sandbox — no manual intervention needed.
CDI GPU mode on Ubuntu 24.04+ / 26.04 (#4956): On Docker-driver GPU hosts with an NVIDIA CDI spec (e.g. /etc/cdi/nvidia.yaml on Ubuntu 24.04 and 26.04), nemoclaw onboard was selecting --gpus all for the GPU patch. The create probe accepted this flag, but the OpenShell supervisor never reconnected to the recreated container, causing onboard to abort. NemoClaw now prefers the CDI device injection mode (--device nvidia.com/gpu=all) over --gpus whenever a CDI spec is detected, matching how OpenShell's own gateway start --gpu injects GPU devices. Non-CDI hosts are unaffected.
Commit log →
Affects: /nemoclaw/, /nemoclaw/setup/, /nemoclaw/policy/, /nemoclaw/local-gpu/, /nemoclaw/switching-providers/, /nemoclaw/skills/
2026-06-09
Claude Cowork
Apps changelog updated
Anthropic apps changelog updated. Anthropic's Claude apps release notes page was updated on June 9. The release monitor detected the change; check the official page for the full details of what was added.
Release notes →
Affects: /claude-cowork/, /claude-cowork/projects/, /claude-cowork/skills-guide/, /claude-cowork/skills-database/