v0.12.288+12288-macos

## OMI Desktop v0.12.288

### What's New
- An answer that Omi speaks after deferring a voice question ("Let me verify...") is now written to the conversation journal, so a follow-up no longer gets answered from the earlier reply it already corrected.
- Bring your own MCP servers and skills: browse the official MCP registry and skill catalogs on the Apps page, or add your own local command or remote URL, and chat sees a compact index of each server's tools and loads full details on demand — all stored as plain files under ~/.omi
- MCP servers work on both transports; chat discovers their tools through a compact sorted index instead of hundreds of tool definitions (mcp_tools_info / mcp_call), published prompts stay usable, local servers spawn with a minimal environment instead of Omi's full process env, and dropped SSE servers reconnect before giving up. Installing a skill brings its scripts and references instead of just the instructions, skills load a section at a time with parts returned by default, and the disable toggle now fully hides a skill from the assistant
- Added or removed MCP servers now reach an open chat right away instead of waiting for a restart, and hand-edits to ~/.omi/mcp.json are picked up on the Apps page
- Fixed an MCP server change being dropped when it landed while the agent runtime was already restarting, and task chat now applies pending server changes too
- Tightened permissions on ~/.omi and its mcp.json, which can hold OAuth tokens, so they are now readable only by your user account
- Quoted paths and arguments now survive when adding a local command server instead of being split at every space
- Renaming or deleting a skill now keeps its disabled state in sync instead of silently re-enabling it or leaving stale entries

### Downloads
- **DMG Installer**: For fresh installs, download the DMG below
- **Auto-Update**: Existing users will receive this update automatically via Sparkle

<!-- KEY_VALUE_START
isLive: false
channel: candidate
edSignature: n4EOD3VSKPqtpSmNVXnFDsTWl7TQcbRAA42BIutT/6LhODk2gr2N3MiSNld0kDlMfTHYg2kYg3yzVNX+R5nkAg==
betaEdSignature: zl3fa+FVknOlkrpQXqg1leldTAd6tGLVVEuSClQH/Q8OVgyOSGpgIPGnBPplYkYpc/bKsXgk0biNv+JBTVhYAw==
changelog: An answer that Omi speaks after deferring a voice question ("Let me verify...") is now written to the conversation journal, so a follow-up no longer gets answered from the earlier reply it already corrected.|Bring your own MCP servers and skills: browse the official MCP registry and skill catalogs on the Apps page, or add your own local command or remote URL, and chat sees a compact index of each server's tools and loads full details on demand — all stored as plain files under ~/.omi|MCP servers work on both transports; chat discovers their tools through a compact sorted index instead of hundreds of tool definitions (mcp_tools_info / mcp_call), published prompts stay usable, local servers spawn with a minimal environment instead of Omi's full process env, and dropped SSE servers reconnect before giving up. Installing a skill brings its scripts and references instead of just the instructions, skills load a section at a time with parts returned by default, and the disable toggle now fully hides a skill from the assistant|Added or removed MCP servers now reach an open chat right away instead of waiting for a restart, and hand-edits to ~/.omi/mcp.json are picked up on the Apps page|Fixed an MCP server change being dropped when it landed while the agent runtime was already restarting, and task chat now applies pending server changes too|Tightened permissions on ~/.omi and its mcp.json, which can hold OAuth tokens, so they are now readable only by your user account|Quoted paths and arguments now survive when adding a local command server instead of being split at every space|Renaming or deleting a skill now keeps its disabled state in sync instead of silently re-enabling it or leaving stale entries
KEY_VALUE_END -->