v1.18
Skills Controls, MCP Media, & Browser DOM Capture
May 26, 2026
v1.18 is a focused polish release for skills, MCP media handling, complex browser pages, chat input sizing, and self-update reliability.
🧠 Skills
- Configurable active skills cap — The maximum number of active skills is now configurable through
max_active_skillsin the skills plugin config and the settings UI, replacing the previous hard-coded limit. - Skills modal styling refined — Active Visible and Pinned toggles now visually read as active, the toggle control has been simplified, and the unused visible/hidden count label was removed.
🔌 MCP & Media Artifacts
- MCP multimodal results preserved — MCP image, audio, and resource tool results are no longer collapsed into empty text responses.
- Images flow into history — MCP images are preserved as data URL attachments so visual results remain available in chat history.
- Binary resources save as artifacts — Audio and non-image binary resources are saved as artifacts with normalized paths.
- Shared media helpers extracted — Common media artifact helpers are now reused across MCP, browser connector, and computer-use paths.
🌐 Browser
- Iframe and shadow DOM support improved — A new browser DOM helper captures shadow DOM and iframe content with frame-chain and node references, improving interaction with complex page structures.
- Helper installs automatically — The DOM helper is installed for both local Docker browser and host-browser runtimes.
🎨 UI & Maintenance
- Chat composer max-height fixed — The chat input box max-height was reduced so it no longer overflows and breaks nearby UI elements.
- Self-update skips transient SSH agent state — The Desktop profile's
.ssh/agentruntime directory is excluded from user backups during self-update, preventing live SSH agent sockets from aborting upgrades.