Release v0.3.0

Added

• Progressive streaming for AI responses using Telegram draft messages with real-time updates
• Telegram draft API compatibility layer with graceful fallback to regular messages
• Automatic detection of numbered options in responses (not just AskUserQuestion tool)
• Streaming configuration via DRAFT_UPDATE_MIN_CHARS and DRAFT_UPDATE_INTERVAL environment variables

Fixed

• Duplicate message issue when responses contain option buttons: streamed messages are no longer re-sent
• Improved AskUserQuestion denial message with clearer formatting instructions for the AI

Changed

• Streaming message handler now uses regular send_message for initial draft creation to ensure message_id availability
• Large text chunks are split into progressive updates for smoother streaming experience

Release v0.2.1

Added

• Session progress summary: show last assistant message when switching sessions via /resume

Changed

• Remove hardcoded zh-CN language policy; bot preset strings stay minimal English, LLM handles language adaptation naturally

Release v0.2.0

Added

• Long message auto-splitting: responses are split at paragraph/line boundaries (4000-char limit) and sent as multiple messages instead of being truncated
• Typing keepalive loop: background task sends typing indicator at regular intervals during long tool calls to prevent Telegram from dropping the typing status

Fixed

• Removed 4000-character hard truncation from _clean_response; full response content is now preserved
• Inline option keyboard now only appears for AskUserQuestion degraded responses (via force_options flag), preventing false positives on numbered lists in regular replies

Release v0.1.0

Added

• Telegram bot integration with Claude Code SDK for running Claude sessions from Telegram
• Per-user persistent Claude SDK streams with session history browsing
• Permission gating for file access: auto-allow inside PROJECT_ROOT, inline button confirmation for outside
• Message queue per user (max 3 concurrent tasks with overflow rejection)
• AskUserQuestion tool degraded to Telegram inline keyboard buttons
• Auto-send media files (photos/documents) when response contains matching file paths
• Session persistence via JSON store (PROJECT_ROOT/.telegram_bot/sessions.json)
• Bilingual documentation (English and Chinese)
• start.sh lifecycle manager with venv creation, dependency caching, log rotation (14 days), and crash detection
• macOS launchd auto-start support via --install / --uninstall
• Debug mode with verbose logging and per-session chat file logging
• Proxy support via PROXY_URL environment variable