OpenHands 1.35 groups related agent actions in its chat activity

OpenHands 1.35 groups related agent actions in its chat activity.

**Dek:** The November 3 open-source release adds descriptive groups for hooks, tool calls and reasoning events. The interface change affects how people follow a coding session and investigate a surprising result.

OpenHands released version 1.35.0 on November 3 UTC, shortly after midnight on November 4 in India. Its tagged notes describe changes to how chat groups hooks, reasoning and tool calls and labels the work each group performed. The release also simplifies titles and disclosure for groups with a single tool. The update concerns the presentation of agent activity in chat. The release notes do not say that every internal reasoning step is exposed. [OpenHands 1.35.0 release](https://github.com/OpenHands/OpenHands/releases/tag/v1.35.0)

Why a useful summary still needs drill-down

A coding agent might search a repository, read several files and run tests to answer a single question. Listing every operation at the same visual level can make it difficult to see what changed or whether the agent is waiting. Grouping related events lets the reader scan the work as a sequence. It also creates a new design question. If a user wants to know why the agent stopped, can they find the failed or pending step without opening every group?

The issue becomes concrete in a review workflow. A group describing a test run could include a failed command, an omitted test or a hook that reported a warning. If the user sees only the group title, they may have difficulty distinguishing those cases. A descriptive title can help them orient themselves, while disclosure lets them inspect the underlying action when needed. OpenHands’ release describes grouping and disclosure changes. What those controls communicate in a particular session requires observing the application with representative events. [OpenHands 1.35.0 release](https://github.com/OpenHands/OpenHands/releases/tag/v1.35.0)

What operators should check

A team assessing the update can compare a successful session, an interrupted session and one that needs human approval. Try following the event order after reopening an existing conversation. Check whether a user can locate the action that requested approval, distinguish it from an action that already ran and move in and out of expanded activity using a keyboard. Labels should make sense without relying on icon color alone. These are evaluation criteria, separate from the project’s announcement about what changed. [Apple accessibility guidance](https://developer.apple.com/design/human-interface-guidelines/accessibility)

An activity feed is also different from the durable audit record an organization may need for sensitive operations. A user-facing summary can make an agent session easier to understand. An operator still needs the relevant event data, permissions and outcomes if they have to reconcile a deployment or a tool action later. For teams that already review agent work, the useful question is whether the grouping makes the evidence faster to find while preserving enough detail to act on it.