v0.4.21

What's new

• Added a pre-send risk scan: checks your message locally for common structured sensitive data (API keys, ID numbers, bank card numbers, phone numbers) before sending and shows a warning. The scan is local-only, never blocks sending, and nothing is sent over the network.
• Added a platform readiness check in Settings: one-click health check across all 5 AI platforms (page reachable, communication alive, logged in), with per-platform quick fixes.
• Non-image files (PDF, Word, Excel, Markdown, HTML, and more) can now be attached and auto-uploaded on all 5 platforms, not just images, with automatic fallback between direct upload and paste.
• Added a self-contained HTML share card export alongside the existing Markdown export, rendering real markdown (headings, tables, code blocks) instead of raw syntax.
• Redesigned the history detail view: per-platform responses now lay out side by side instead of stacked, with a narrower list panel and clearer action buttons.
• Fixed response-capture bugs across platforms: thinking/reasoning content no longer leaks into saved answers (Claude, DeepSeek, Doubao, Gemini, ChatGPT); DeepSeek and Doubao completion detection is faster and more reliable; Claude and ChatGPT answers now preserve markdown formatting (tables, code blocks, links) instead of flattening to plain text; fixed several Doubao/Gemini rendering glitches (math formulas, code-block language labels, table headers).
• Fixed history status handling: sessions stopped by the user or that time out repeatedly are now correctly saved as failed instead of staying stuck at "pending"; genuinely uncertain captures get their own status instead of showing as a failure.
• Internal refactors to the background service worker's message handling and the chat-forwarding logic, now covered by tests; no behavior change.

No manifest permission changes in this release.

Full Changelog: v0.4.20...v0.4.21