v1.18

deepseek-ai/DeepSeek-OCR-2v1.18May 26, 2026by github-actions[bot]

AI Summary

This release focuses on system stability and UI improvements, including configurable skill limits, fixes for MCP multimodal content handling, and enhanced browser DOM support.

Key Highlights

  • Configurable skills cap replacing hard-coded limits
  • Fix for MCP multimodal content handling (image, audio, resources)
  • Improved browser iframe and shadow DOM support
  • Skip transient SSH agent state during self-update

New Features

  • Configurable maximum active skills
  • MCP image and audio preservation as artifacts
  • Browser DOM helper for shadow DOM and iframe capture
  • Refined skills modal styling

Full Release Notes

## Release Notes

- **Configurable skills cap** — The maximum number of active skills is now configurable via `max_active_skills` in the skills plugin config and the settings UI, replacing the previous hard-coded limit.
- **Fix MCP multimodal content handling** — MCP image, audio, and resource tool results are now properly preserved instead of being collapsed into empty text responses. Images flow into history as data URL attachments; audio and non-image binary resources are saved as artifacts with normalized paths. Shared media artifact helpers have been extracted and reused across MCP, browser connector, and computer-use paths.
- **Improved browser iframe and shadow DOM support** — A new browser DOM helper captures shadow DOM and iframe content with frame-chain/node references, improving the agent's ability to interact with complex page structures. The helper is installed automatically for both local and host-browser runtimes.
- **Fix chat composer max-height** — Reduced the chat input box max-height to prevent it from overflowing and breaking surrounding UI elements.
- **Skip transient SSH agent state during self-update** — The Desktop profile's `.ssh/agent` runtime directory is now excluded from user backups during self-update, preventing live SSH agent sockets from aborting upgrades.
- **Refined skills modal styling** — Active Visible/Pinned toggles now visually read as active; the toggle control has been simplified and the unused visible/hidden count label has been removed.