agentscope-ai/QwenPaw v1.1.3 on GitHub

✨ Added

Agent System

Backup & Restore: Backup and restore system — create scoped snapshots of agents, skills, memory, and sessions with per-agent selection, import/export as zip files (#3534, #3655)
ACP Server: Expose QwenPaw agents as ACP (Agent Communication Protocol) endpoints over stdio via qwenpaw acp (#3487, #3589, #3595)
Proactive Agent Messaging: Agents can proactively send messages, using session memory and screen context to provide timely information (#3466, #3643)
Cross-Provider Message Normalization: Seamlessly switch LLM providers mid-conversation — provider-specific message fields are automatically cleaned and normalized when changing models (#3530)

Console & UI

Console Plugin System: Extensible plugin architecture for the console — third-party plugins can inject sidebar pages, register routes, and share host modules (#3357, #3512, #3631)
Agent Statistics Page: New dashboard with session and message trend charts, token usage over time, and channel breakdown pie charts (#3365, #3608, #3632)

Skills

Skill Page Improvements: Redesigned Skills and Skill Pool pages with batch select/manage, separated search and filter controls (#3616, #3634)
Builtin Skill Language Switch: All builtin skills now ship with English and Chinese variants. Switch language from the Skill Pool page, with default language support when importing skills to agents (#3558, #3644)

Security

Shell Evasion Guard: New Tool Guard guardian that detects shell command obfuscation patterns (#3400, #3668)
Enhanced Tool Guard Response: Tool guard messages now include risk severity level and a localized explanation of why a tool call was blocked or requires approval (#3515)

Providers

Local Model Management: Configure custom server ports for local models (llama.cpp) and monitor server logs in real-time (#3596, #3604)
OpenRouter Multimodal Detection: OpenRouter models now automatically report multimodal capabilities (image/video support) from the model registry (#3584, #3604)
Aliyun Coding Plan International: Added international region support for the Aliyun Coding Plan provider (#3609)

Channels

QQ Instant Acknowledgment: Configurable instant reply for QQ channel — acknowledge the user immediately while the agent processes the request (#3246)
Telegram Typing Indicator: Typing indicator now stays active during tool execution, providing continuous visual feedback while the agent works (#3585)
DingTalk @Mention on Reply: Automatically @mention the original sender when replying in DingTalk group chats (#3591)
Channel Health Check & Restart API: Per-agent HTTP endpoints to check individual channel health status and hot-restart channels (#3649)

CLI

Update Provider Base URLs: qwenpaw providers update now supports updating provider base URLs from the command line (#3536)

Debug Page Redesign: Relocated the Debug page under Settings with improved component structure, dedicated log viewer, and dark mode styling fixes (#3539, #3547, #3590)
make_plan Skill Update: Cleaned up stale documentation references and streamlined the make_plan skill (#3535)
Unified Channel Media Directory: All channels now resolve media_dir from the agent workspace consistently, ensuring isolated file storage across channel types (#3610)

Console & UI

Preview URL Prefix: Fixed duplicated URL prefixes in file preview paths (#3355)
Markdown Rendering: Fixed markdown rendering and code block copy styling in workspace file editor (#3639)
Table Scrolling: Fixed table scrolling on Sessions and Cron Jobs pages (#3654)

Channels

WeCom Attachment Access: Fixed WeCom attachment access failing in server deployments (#3079)
WeCom File Upload: Fixed event loop blocking in WeCom file uploads and added relative path support for send_file_to_user (#3128)
WeCom Duplicate Messages: Fixed duplicate chat entries for WeCom group messages (#3529)
WeCom Screenshot Filename: Fixed CJK characters in WeCom screenshot filenames causing download failures (#3633)

Full Changelog: v1.1.2...v1.1.3