v0.2.5

✨ Feature

• Added a Skills & Prompt page where users can create, edit, enable, disable, and delete personal prompt presets.
• Added an admin Prompt Library for managing built-in platform prompts that users can view and reuse.
• Added / prompt insertion in the chat composer and user-message editor, alongside the existing @ model, file, and tool picker.
• Added reusable model capability templates, allowing admins to apply protocol presets or copy capability configs from existing models.
• Added support for creating a model with multiple upstream sources, including protocol, status, priority, and weight settings.
• Added richer file preview support for images, audio, video, PDFs, Docx, spreadsheets, Markdown, code, and plain text.
• Added a Skills & Prompt entry to the main sidebar navigation.
• Added Prompt Preset APIs for both user-owned prompts and admin-managed built-in prompts.

⚡ Optimization

• Improved mention menu placement so @ and / menus open near the current caret position, including while editing previous user messages.
• Improved prompt search and selection, with visible prompts loaded by keyword and grouped by custom vs built-in sources.
• Improved model capability editing with visual controls for default options, option controls, native tools, and image streaming.
• Improved file settings descriptions for upload limits, extraction, full-context injection, embeddings, semantic enhancement, and RAG.
• Unified file display logic for type detection, icons, sizes, timestamps, and preview handling.
• Improved subscription usage charts with cleaner rounded stacked bars.
• Hardened login and OAuth redirect path handling with shared local-path validation.
• Improved admin error messages through shared localized error resolution.
• Renamed the admin “Channels” area to “Upstreams” for clearer terminology.
• Moved shared preview, Markdown rendering, delete-file option, and loading reveal components into common modules for more consistent behavior.

🐛 Fixed

• Fixed incomplete SQLite unique-constraint detection so duplicate conflicts map to clearer errors.
• Fixed missing localized feedback when a conversation has no content suitable for automatic title generation.
• Fixed navigation search layout jitter caused by dynamic result-height calculations.
• Fixed overly low-level error messages in several admin billing, model, and bulk operation flows.
• Fixed OAuth callback logic being embedded directly in the page component, improving reuse and 2FA redirect handling.
• Fixed unclear file preview feedback for unreadable text or unsupported formats, now showing clearer preview/download guidance.