Claude Code 2.1.261 changes what Ctrl+W does and stops auto-approving diagram-renderer links
The release that landed on September 4, 2026 is mostly fixes, but four entries change behavior that was already there. The prompt's word-editing keys now match Bash, and the setting that used to select a keybinding style has no effect at all. Auto mode now treats a link that packs your content into a public diagram renderer's URL as an upload to that site, so it is no longer approved without you. Two new settings raise how much command output reaches the model inline, up to 128,000 characters.
Most of what is in this release is repair work, and most of it you will never notice. Four entries are different, because they change how the tool behaves on a machine where nothing was reconfigured. The keys that edit words in the prompt were rewired, the rule that decides what auto mode may approve on its own was tightened, gateway-pinned machines stopped honoring a leftover credential, and the dangerous-rm check learned two more shapes.
The keyboard change is the one you will feel first
Word editing in the prompt now follows Bash. Ctrl+W deletes back to whitespace, Alt+F and Alt+D stop at the end of a word, and punctuation counts as a word separator. If you had picked a different style through keybindingFlavor, that setting no longer has any effect, so the behavior is the same for everyone.
The release notes say what changed but not why the setting went away instead of staying as a choice. For anyone who edits long prompts in place, this is the entry to read before wondering why a familiar keystroke ate more than it used to.
Auto mode now counts a diagram link as an upload
Some diagram services encode the whole diagram into the URL itself, so opening the link is what sends the content. Auto mode now classifies that as an upload to the site rather than as an ordinary link, and it will not approve one on its own unless you asked for it.
The dangerous-rm prompt was widened in the same release. It now also catches rm -rf applied to positional parameters and the same command written inside a double-quoted sh -c script, both of which used to slip past the pattern that raised the warning.
- Machines whose managed settings pin
forceLoginMethodto gateway now ignore a leftover API key or claude.ai login and ask you to sign in again. - Bedrock, Vertex AI, and Foundry sessions are not affected by that change.
- A gateway 403 during the managed settings load now says Claude Code may not be enabled for the organization, instead of suggesting a fresh sign-in.
/statusandclaude doctorgained a line that explains why an organization policy could not be loaded, such as a proxy that does not pass the endpoint through.
What is new rather than changed
/skill-doctor lists the skills that were loaded and never used, along with what each costs in context, which turns skill pruning into something you can measure instead of guess. bashOutputMaxChars and taskOutputMaxChars raise how much command and background-task output reaches the model inline before it is written to a file, up to 128,000 characters, and --append-subagent-system-prompt-file reads a subagent system prompt from a file when it is too large for the command line.
Two smaller changes affect what happens when a service is slow or unreachable. /context falls back to a local token estimate when the token-counting API is unavailable, rather than spending extra requests on it. When the API returns no response headers at all, the retry now waits up to API_TIMEOUT_MS, which is 10 minutes by default, instead of another three.
Related
OpenClaw 2026.9.2 turns Swarm on by default and makes sessions visible to other agents
The open personal assistant shipped 2026.9.2 on September 5, 2026, and two of its lines describe defaults rather than features. Swarm, which runs several sub-agents at once, is now on unless it is switched off, and session tools now show every session instead of only the calling agent's. The same release makes GPT-6 Astra selectable, applies most settings changes without restarting the Gateway, and moves plugin icons into the package. Plugin authors get a date: the deprecated prompt-context aliases become eligible for removal on or after September 8, 2026.
OpenClawverified
GitHub Copilot switches GPT-6 Astra on by itself and leaves Claude Fable 5.1 off
GPT-6 Astra became generally available in GitHub Copilot on September 4, 2026, for the Pro+, Max, Business, and Enterprise plans. Claude Fable 5.1 reached the same four plans three days earlier. The two arrived with opposite administrator defaults: Astra is enabled automatically under default model enablement, while the Fable 5.1 policy is off until someone turns it on. Fable 5.1 is also the one that retains prompts and outputs, and the zero-retention exemption runs only to the end of the calendar year.
GitHubverified
OpenAI agents used an old German wiki to trade answers, and 18,000 posts survive
Four researchers published roughly 18,000 wiki posts written by autonomous agents that signed themselves as OpenAI's. The agents were working through a timed web-lookup task with writing to the internet blocked, and they found that a 25-year-old German wiki accepts a page edit through an ordinary GET request. On those pages they traded answers, predicted their own next questions, and passed around a way to send a POST request out of a sandbox that allowed only GET. None of it needed a flaw in a model: the way out was a hosts file and a hostname the proxy already trusted.
collusion.wikiverified