A plugin skill that stops responding after an update can look like a broken installation. With Kimi Code CLI 2.3.0, there is a simpler possibility to check first: the name used to invoke it has changed. The October 21 release arrived alongside Desktop 1.1.1, which also changes what happens when you open an untrusted workspace.
For existing users, they are worth noticing before the next session.
Plugin skills now need their full names
Skills supplied by plugins must now be invoked as “pluginID:skillName.” The bare skill name no longer works. Including the plugin ID lets Kimi distinguish skills with the same name instead of allowing one plugin’s skill to obscure another’s.
If a plugin skill seems to have disappeared after upgrading, checking its invocation name is a better first step than reinstalling everything. Any saved prompts or personal notes that use the old name are worth reviewing too. The change concerns plugin-provided skills; it does not mean that every custom skill has stopped working.
An older permission setting deserves a second look as well. In CLI 2.3.0, dangerous_command_guard = false disables all built-in confirmation guards, rather than only those for dangerous commands.
The default remains true. The people who need to pay attention are those who previously turned it off: built-in confirmation checks for sensitive-file access and Git control paths are now included in its scope. There is an important limit to that statement. The documentation says hard blocks on secret files in Read, Write and Edit remain in place, so describing this as the removal of every safety protection would be misleading.
Kimi’s documentation reserves this opt-out for environments where command permissions are already centrally restricted outside the agent. If an old configuration was copied merely to reduce the number of approval prompts, this is a good time to reconsider it.
Desktop asks about trust, and adds a useful reload option
Desktop 1.1.1 asks whether to trust a workspace when opening a folder that has not been trusted. Declining prevents messages from being sent there. If that happens, it is worth checking the workspace decision before assuming there is a network or account problem. Confirm which folder the dialog refers to and whether you know where the project came from, rather than accepting it simply to make the prompt go away.
The new /reload command should also save some unnecessary restarts. After editing AGENTS.md, skills or configuration outside the app, you can reload the session to apply the changes. The same action is available from the session menu.
Both the CLI and Desktop now ask what to do when switching models: start a new session, compact the current context and continue, or carry on directly. That makes the fate of an existing conversation more explicit when changing models mid-task.
The permission change is the one least likely to explain itself on screen. An old setting can stay in place while its effect becomes broader, so it deserves a closer look than the version number alone. These are client updates; the announcement does not introduce a new model. We have checked the official documentation but have not tested the upgrade ourselves.
