This document details recent updates and new features, focusing on enhancements to agent interaction, user control, and contextual understanding.

A significant update is in Automated and improved rules, enhancing the agent's ability to manage and apply operational guidelines. Users can now generate rules directly from conversation context via the /Generate Cursor Rules command, allowing for the capture and reuse of specific conversational elements as structured rules. For "Auto Attached rules," the system now automatically applies relevant rules when files are read or written, based on predefined path patterns, implying an improved pattern-matching and rule-application engine. A persistent issue where "Always attached rules" would not reliably endure across extended conversations has been resolved, ensuring consistent rule application over time. Furthermore, the agent's capability to reliably edit existing rules has been introduced, granting programmatic modification of rule definitions.

More accessible history centralizes chat history management. It has been relocated to the command palette, accessible through a "Show history button" within the chat interface or directly via the "Show Chat History" command.

For Making reviews easier, a built-in diff view has been integrated at the conclusion of each conversation, specifically for agent-generated code. This feature, accessible via a "Review changes" button positioned at the bottom of the chat after an agent's message, streamlines the review process by visually highlighting code modifications. The underlying mechanism involves the system automatically computing and presenting a delta between the pre- and post-agent-modification states.

Images in MCP (presumably referring to a Multi-Context Platform or similar) now support the inclusion of images as part of the contextual input provided to servers. This enables the agent to process visual data such as screenshots, UI mocks, or diagrams, thereby enriching the context for questions or prompts and allowing for more informed responses based on visual information.

Improved agent terminal control provides users with enhanced oversight of terminals initiated by the agent. Commands executed by the agent can now be edited prior to their execution or entirely skipped. The "Pop-out" function has been renamed to "Move to background" to more accurately reflect its action of moving a terminal process to a background operation without closing it.

Global ignore files allow users to define ignore patterns that apply universally across all projects via user-level settings. This mechanism prevents specified noisy or sensitive files, such as build outputs or secrets, from being included in prompts, thereby maintaining prompt relevance and security without requiring project-specific configurations.

The system has integrated New models, significantly expanding the available large language models. These additions include Gemini 2.5 Pro, Gemini 2.5 Flash, Grok 3, Grok 3 Mini, GPT-4.1, o3, and o4-mini, accessible through model settings, offering a wider range of AI capabilities for different tasks.

Finally, a Project structure in context (Beta) feature has been introduced. This option allows for the inclusion of the directory structure of a project directly into the prompt provided to the agent. By providing this hierarchical organization of files and directories, the agent gains a clearer sense of the project's layout, which is intended to improve the quality of suggestions and navigation, particularly within large or nested monorepos. The core methodology involves augmenting the agent's textual context with a representation of the file system's topology, enabling a more informed understanding of code locations and relationships.