OpenClaw Adds a Way to Turn Past Chats Into Reusable Skills

OpenClaw has released version 2026.9.4, giving users a way to turn past conversations into reusable skills through a chat they can steer. The update also makes plugins and skills easier to find, adding a more direct path from a one-off agent exchange to work that can be reused later.

The release arrives as a broad maintenance and workflow update rather than a single-purpose feature drop. Alongside the chat-to-skill workflow, OpenClaw added GPT Image 2.5 support, more control over cloud sessions, and interactive questions in the terminal. It also includes fixes affecting updates, memory, messaging, onboarding, and the web interface.

A shortcut from chat to skill

The central change is an attempt to preserve work that would otherwise remain trapped in a conversation history. OpenClaw says users can guide the conversion of an earlier chat into a reusable skill, while the release also improves discovery of both skills and plugins. That puts creation and retrieval in the same update: a user can shape a repeatable capability from previous work, then more easily locate it later.

• Past conversations can be converted into reusable skills through a steerable chat workflow.
• Plugins and skills are intended to be easier to find.
• GPT Image 2.5, cloud-session controls, and terminal questions are included in the same release.

1,558Pull requests

OpenClaw lists 1,558 pull requests in the release.

20Direct commits

The release notes list 20 direct commits.

293Contributors

OpenClaw credits 293 contributors to version 2026.9.4.

The work before the useful work

Much of the release is aimed at reducing failures before an agent can be useful. New-agent setup can now use API keys already stored in OpenClaw’s protected storage. When startup fails because of Node.js, OpenClaw can reuse a compatible local installation or offer to install a private copy used only by OpenClaw, rather than replacing the computer’s existing Node installation.

That recovery path still has limits. OpenClaw says some systems, including Alpine Linux, require manual installation. The release also separates command-line recovery from repairing a background service that may still be using an older Node version after an upgrade.

A web interface built for longer threads

The web app changes focus on navigating and continuing work. A compact message rail can preview earlier messages before a user jumps to them, covering messages already loaded in a long conversation. Users can also highlight text and add an editable quote of up to 300 characters to a draft; it is not sent until the user sends the message.

On reload, the app can show saved conversation and sidebar content while reconnecting for remembered Gateway tokens or paired-device sign-ins. Actions such as sending a message wait for the live connection, and a shared web address used for several Gateways can briefly display a conversation cached from another Gateway. OpenClaw advises waiting for the live conversation before relying on that view.

The next move is repeatability

Version 2026.9.4 is built around a practical proposition: an agent interaction should not necessarily end when its chat does. The open question is whether steering an earlier conversation into a skill will make reusable workflows easier to build than starting fresh—or simply add another layer for users to manage. The release supplies the feature and the surrounding setup repairs; its usefulness will depend on how those pieces hold together in regular work.