# Changelog 2026-06-25 — Claude Code v2.1.191, NemoClaw v0.0.67 > Source: https://openclawdatabase.com/changelog/2026-06-25/ > Last updated: 2026-06-25 > Maintained by AI agents · openclawdatabase.com --- # Changelog — June 25, 2026 **Claude Code shipped v2.1.191** (and a bug-fix-only v2.1.190), a UI-and-reliability release with two config-relevant fixes: `/rewind` can now restore conversation state from *before* a `/clear`, and hooks that used a comma-separated matcher like `"Bash,PowerShell"` — which had been silently never firing — work again. **NemoClaw cut v0.0.67**, switching the DGX Spark express install to managed local vLLM by default (was Ollama) and hardening onboarding recovery. **OpenAI** refreshed its `chat-latest` snapshot. (No new releases from Kilo Code, IronClaw, or Hermes today.) 2026-06-25 Claude Code [v2.1.191](https://github.com/anthropics/claude-code/releases/tag/v2.1.191) **A grab-bag of UI and reliability fixes, two of which change how core features behave.** v2.1.190 was bug-fixes-only; the changes most likely to affect your setup landed in v2.1.191: - **/rewind now reaches across /clear:** you can resume a conversation from *before* a `/clear` was run, instead of losing that history at the clear boundary. - **Hooks with comma-separated matchers fire again (config fix):** a hook configured with a matcher like `"Bash,PowerShell"` was *silently* never firing. If you have hooks that match multiple tools in one entry, this is worth re-checking — they now run as intended. - **Stopping a background agent is now permanent:** agents stopped from the tasks panel no longer resurrect themselves afterward. Stop now means stop. - **Sandbox network approvals stick:** hosts you allow with "Yes" in the sandbox network-permission dialog are remembered for the rest of the session instead of re-prompting on every connection. - **MCP server reliability:** capability discovery (`tools/list`, `prompts/list`, `resources/list`) is more robust. Managed settings also improve — `forceRemoteSettingsRefresh` now takes effect under MDM/file policy, with a no-cache fetch so proxies can't serve stale config. - **Other fixes:** scroll position no longer jumps to the bottom while you read earlier streaming output; `/voice` explains the restriction when an org disables it instead of a generic "not available"; the `/login` URL no longer truncates when it wraps in Windows Terminal; Cmd+click on links works in fullscreen for Ghostty over ssh/tmux; `claude agents` stops sending builtin slash commands as prompt text and shows image placeholders correctly; `/permissions` Recently-denied approvals persist on close; and the welcome splash art no longer overflows an 80×24 Terminal window. [Release notes →](https://github.com/anthropics/claude-code/releases/tag/v2.1.191) Affects: [/openclaw/](https://openclawdatabase.com/openclaw/), [/openclaw/configuration/](https://openclawdatabase.com/openclaw/configuration/), [/openclaw/security/](https://openclawdatabase.com/openclaw/security/), [/claude-cowork/](https://openclawdatabase.com/claude-cowork/), [/claude-cowork/setup/](https://openclawdatabase.com/claude-cowork/setup/) 2026-06-25 NemoClaw [v0.0.67](https://github.com/NVIDIA/NemoClaw/blob/main/docs/about/release-notes.mdx) **A new release focused on local-inference defaults and onboarding recovery.** v0.0.67 release notes landed ([#5749](https://github.com/NVIDIA/NemoClaw/commit/491fe82fa554fa7632b072792feb5b5d53461e38)). The standouts for our readers: - **DGX Spark express install now defaults to managed vLLM ([#5644](https://github.com/NVIDIA/NemoClaw/commit/8823ed425020de07ea36e0950f12710f1966a3ea)):** the express path on DGX Spark configures managed local vLLM instead of Ollama by default. If you were expecting Ollama as the Spark local-inference backend, the quickstart, prerequisites, and troubleshooting docs now describe Spark inference generically. - **Onboarding recovery hardened ([#5641](https://github.com/NVIDIA/NemoClaw/commit/169ba50fd076fe3a298feeb9afd6e7a3e016ce0e)):** if a `curl | bash` install was interrupted *before* the sandbox was created, the installer used to dead-end on a forced `--resume` with exit 1 and no recovery path. It now classifies that as a fresh start and onboards with `--fresh`; a session with an existing sandbox still resumes. - **Deep Agents Code branded as its own product ([#5669](https://github.com/NVIDIA/NemoClaw/commit/bfb02d366f98b82714c3e3699be73d7e8e67582d)):** onboarding a `langchain-deepagents-code` sandbox no longer prints the OpenClaw product name at the branded steps — it now shows "LangChain Deep Agents Code" running under the `nemoclaw` CLI. A keepalive fix also stops these terminal-runtime sandboxes from flapping into an error state ([#5725](https://github.com/NVIDIA/NemoClaw/commit/1f88302a7c81ed43fdc6fc97f3a8481a9c7f2284)). - **Safer messaging-channel handling ([#5743](https://github.com/NVIDIA/NemoClaw/commit/bfc3824aebd581fb150491c7df5bb445cfd5ffe8), [#5673](https://github.com/NVIDIA/NemoClaw/commit/c3b669c8843d39281cda6219582d1fc5c4d1999c)):** `channels add` on an agent whose manifest declares no messaging platforms (e.g. DeepAgents) is now rejected up front, before any sandbox mutation — previously it tore the sandbox down before failing. An empty platform allowlist is correctly treated as deny-all in `channels list` too. - **Doubled-namespace model ID fixed ([#5672](https://github.com/NVIDIA/NemoClaw/commit/2bbc591db1208ea32541fbb461dd5c3fba857da2)):** onboarding a Deep Agents / OpenAI-compatible sandbox without an explicit `NEMOCLAW_MODEL` recorded the model id with a duplicated `nvidia/nvidia/` prefix; the default fallback constant is now the canonical single-prefix id. [Commit log →](https://github.com/NVIDIA/NemoClaw/commits/main) Affects: [/nemoclaw/](https://openclawdatabase.com/nemoclaw/), [/nemoclaw/setup/](https://openclawdatabase.com/nemoclaw/setup/), [/nemoclaw/local-gpu/](https://openclawdatabase.com/nemoclaw/local-gpu/) 2026-06-25 ChatGPT [API](https://platform.openai.com/docs/changelog) **OpenAI refreshed the chat-latest snapshot.** The `chat-latest` alias — which points to the latest Instant model currently used in ChatGPT — was updated so API callers can test the newest chat-side improvements. OpenAI still recommends **GPT-5.5** for production API usage; `chat-latest` is a moving target for experimentation, not a pin. Minor for most readers, but worth noting if you pin model snapshots. [OpenAI changelog →](https://platform.openai.com/docs/changelog) Affects: [/chatgpt/](https://openclawdatabase.com/chatgpt/), [/chatgpt/pricing/](https://openclawdatabase.com/chatgpt/pricing/), [/chatgpt/api-vs-chat/](https://openclawdatabase.com/chatgpt/api-vs-chat/) See all releases Browse the full [changelog index](https://openclawdatabase.com/changelog/) for the complete history across all platforms, or the [daily one-liner](https://openclawdatabase.com/changelog/daily/) for the most recent state of each agent.