Guangzhou, China
Joined on 2023-08-24
c23de38b21
fix(dashboard): convert UTC times to Beijing time before today/yesterday delta calculation
a517768083
feat: AI hint supports multi-turn conversation with context - hint_history list stored in conversation, accumulates across rounds - Each new hint appended, full history passed to AI as combined prompt - Format: 'hint1 → hint2 → hint3' so AI understands the conversation flow
e0c2110c09
fix: use original email subject for reply, not AI-extracted title Reply emails should use the original subject from IMAP header (e.g. 'Re: [original]'), not the AI-generated Chinese summary title.
b18dcb0527
fix: wrap original text in code block to avoid MarkdownV2 parse errors Email bodies contain *, _, etc. that break MarkdownV2 entity parsing. Wrapping in backtick code block prevents Telegram from parsing internals.
9adb32cb0c
feat: cleanup reply flow messages + AI comparison explanation - Track extra_msg_ids in conversation for code block messages - _cleanup_conv_messages: delete all intermediate messages on cancel/send - Cancel/send/edit all clean up extra messages before restoring summary view - AI prompts now return 'comparison' field with Chinese explanation of differences - _show_ai_suggestions displays comparison text above button list - generate_more_replies returns {suggestions, comparison}
ae1a23143e
feat(ui): optimize landing page and showcase for mobile (hamburger nav, 2-col grid, compact lightbox)
beaa3f146e
feat: show full recipient and CC in email summary - Email class: add cc field - fetch_unseen_emails: parse Cc header - _format_normal: display full recipient, add CC line if present - send_summary/ai_process: pass cc through to email_context - save_email_context: store cc for callback use
1c53209acf
fix(templates): replace all TemplateResponse with _render to bypass Jinja2 unhashable dict cache bug
b80a1859a0
fix(templates): fix unhashable dict in Jinja2 cache by converting available_languages global to function
7a01b29655
fix: update email_client and smtp_client for new config structure - email_client: use account.imap.server/port/ssl/starttls instead of flat fields - smtp_client: use s.ssl/starttls instead of s.use_ssl/use_starttls - tg_bot: fix _do_send_reply to use find_user_by_chat_id for email_accounts
f2b3793d4c
feat: multi-user support - Config format: JSON with bot_token (global), users array - Each user has chat_id, AI config, email_accounts - Each email account has separate IMAP/SMTP with ssl/starttls - Config module: new dataclasses (UserConfig, ImapConfig, SmtpConfig) - find_user_by_chat_id() for callback dispatch - Main: email poller iterates all users, queues carry user object - Summarizer: accepts UserConfig, uses user-specific AI and chat_id - TG bot: all functions use bot_token string instead of TelegramConfig - Callback/message handlers look up user by chat_id - AI regen/hint threads use find_user_by_chat_id for AI config - Config example: config.example.json - Tests: 3/3 passing
9906f1675f
chore: remove reply_suggestions from email summarization prompt AI reply suggestions are now generated on-demand when user clicks full/short reply buttons, using style-specific prompts.
8386ab3c89
fix: always regenerate AI replies when selecting style, ignore summary cache The cached reply_suggestions from summarization use a generic prompt, not matching full/short style. Now always triggers fresh generation.
bedf5b887c
feat: 2-col reply buttons, direct send, language-matched prompts - Buttons: [回复N 直接发送] [回复N 编辑] in 2-column layout - CALLBACK_SEND_REPLY: send immediately without editing - Prompts enforce matching email language (no Chinese forced on English) - Full mode: greeting + comma only, no [Your Name] placeholder - Tests: 5/5 passing
6ce9a806ea
fix: clear reply suggestions cache on cancel, regenerate when switching styles - Cancel now clears reply_suggestions from email_context - Selecting style with empty cache triggers fresh AI generation - Selecting style with cache uses cached suggestions - Tests: 3/3 passing