v2.2.1

## What's Changed

## ✨ Added

**Models, Agents & Memory**

- Configure model routing separately for each Agent, including provider preferences and fallback behavior ([#7501](https://github.com/agentscope-ai/QwenPaw/pull/7501))
- Add Auto Fin proactive memory review and upgrade ReMe for more reliable long-term-memory processing ([#7441](https://github.com/agentscope-ai/QwenPaw/pull/7441))
- Add PowerContext as an optional long-term-memory backend with configuration in the Console ([#7080](https://github.com/agentscope-ai/QwenPaw/pull/7080))
- Strengthen built-in Agent behavior around request scope, untrusted content, confirmations, tool use, and completion ([#7526](https://github.com/agentscope-ai/QwenPaw/pull/7526))

**QwenPaw Creator**

- Add a blueprint workbench, multi-timeline comparison, reusable video templates, MiniMax H3 support, project snapshot rollback, clearer Agent progress, and Docker deployment in Creator 1.2.0 ([#7486](https://github.com/agentscope-ai/QwenPaw/pull/7486))

**Console & Workspaces**

- Manage environment variables in a unified settings page shared by Agents, MCP services, tools, and local models ([#7538](https://github.com/agentscope-ai/QwenPaw/pull/7538))
- Detect available plugin and PawApp updates and install them with one click ([#7605](https://github.com/agentscope-ai/QwenPaw/pull/7605))
- Redesign the sidebar and Settings experience with customizable navigation and improved session browsing ([#7502](https://github.com/agentscope-ai/QwenPaw/pull/7502))
- Restore direct path entry when choosing a session project directory ([#7593](https://github.com/agentscope-ai/QwenPaw/pull/7593))
- Add Chinese and English localization to Agent Kanban ([#7482](https://github.com/agentscope-ai/QwenPaw/pull/7482))

**Skills & MCP**

- Let selected Skills preload their full instructions so they are ready without an extra activation step ([#7183](https://github.com/agentscope-ai/QwenPaw/pull/7183))
- Upgrade Make Skill with guided drafting, validation, behavior tests, and publishing tools ([#7509](https://github.com/agentscope-ai/QwenPaw/pull/7509))
- Show Skill versions and validate declared dependencies before use ([#7609](https://github.com/agentscope-ai/QwenPaw/pull/7609))
- Configure connection timeouts for HTTP and SSE MCP clients ([#7649](https://github.com/agentscope-ai/QwenPaw/pull/7649))

## 🔄 Changed

- Give automatic memory backends a consistent lifecycle and action interface, and deliver ADBPG and PowerContext as plugins ([#7561](https://github.com/agentscope-ai/QwenPaw/pull/7561), [#7616](https://github.com/agentscope-ai/QwenPaw/pull/7616))
- Standardize Console colors and component states across light and dark themes ([#7487](https://github.com/agentscope-ai/QwenPaw/pull/7487))
- Improve the Agent selector on smaller screens ([#7623](https://github.com/agentscope-ai/QwenPaw/pull/7623))

## 🐛 Fixed

**Console, Chat & Files**

- Prevent stale session state and queued-send races when streaming, switching conversations, or submitting messages quickly ([#7523](https://github.com/agentscope-ai/QwenPaw/pull/7523), [#7610](https://github.com/agentscope-ai/QwenPaw/pull/7610), [#7237](https://github.com/agentscope-ai/QwenPaw/pull/7237))
- Keep free models separate from the Pro model tab ([#7524](https://github.com/agentscope-ai/QwenPaw/pull/7524))
- Preserve the selected loop mode when opening a chat from a link ([#7560](https://github.com/agentscope-ai/QwenPaw/pull/7560))
- Use the same language options throughout the Console ([#7595](https://github.com/agentscope-ai/QwenPaw/pull/7595))
- Keep Chrome extension tabs grouped with the correct QwenPaw session ([#7457](https://github.com/agentscope-ai/QwenPaw/pull/7457))
- Make navigation and theme-toggle icons visually consistent ([#7499](https://github.com/agentscope-ai/QwenPaw/pull/7499))
- Show PawApp updates whenever the installed and available versions differ ([#7651](https://github.com/agentscope-ai/QwenPaw/pull/7651))
- Show file upload controls only inside a workspace ([#7667](https://github.com/agentscope-ai/QwenPaw/pull/7667))
- Restore consistent QwenPaw styling for forms and the Git panel ([#7646](https://github.com/agentscope-ai/QwenPaw/pull/7646))

**Runtime, Security & Compatibility**

- Require approval for critical security findings unless policy explicitly denies the action ([#7525](https://github.com/agentscope-ai/QwenPaw/pull/7525))
- Enforce each MCP server's tool allowlist on the Agent runtime path ([#7504](https://github.com/agentscope-ai/QwenPaw/pull/7504))
- Move managed Chromium installation out of the startup path so browser setup cannot block QwenPaw launch ([#7539](https://github.com/agentscope-ai/QwenPaw/pull/7539))
- Prevent ACP Agents from stalling during workspace startup on Windows ([#7401](https://github.com/agentscope-ai/QwenPaw/pull/7401))
- Accept numeric-looking tool arguments emitted as JSON numbers by some models ([#6936](https://github.com/agentscope-ai/QwenPaw/pull/6936))
- Convert PDF content safely when a selected model accepts text but not document blocks ([#7621](https://github.com/agentscope-ai/QwenPaw/pull/7621))
- Keep long conversations within context limits by folding thinking content that has already been consumed ([#7521](https://github.com/agentscope-ai/QwenPaw/pull/7521))
- Restore compatibility with MCP servers whose legacy authentication flow responds to discovery with HTTP 401 ([#7627](https://github.com/agentscope-ai/QwenPaw/pull/7627))
- Prevent shell child processes from competing with QwenPaw for interactive input ([#7598](https://github.com/agentscope-ai/QwenPaw/pull/7598))
- Return the correct not-found response when configuration is requested for an unknown tool ([#7498](https://github.com/agentscope-ai/QwenPaw/pull/7498))
- Preserve full exception details in logs when a downstream tool-call handler fails, while returning a safe error to the model ([#7578](https://github.com/agentscope-ai/QwenPaw/pull/7578))

**Memory, Hub & Operations**

- Return actionable memory status when a configured backend is unavailable, and fall back safely when its plugin cannot load ([#7544](https://github.com/agentscope-ai/QwenPaw/pull/7544), [#7663](https://github.com/agentscope-ai/QwenPaw/pull/7663))
- Start Hub runtimes without blocking lifecycle operations and authenticate CLI requests to the active local runtime ([#7566](https://github.com/agentscope-ai/QwenPaw/pull/7566), [#7631](https://github.com/agentscope-ai/QwenPaw/pull/7631))
- Restore Skill discovery through the current ClawHub endpoint ([#7640](https://github.com/agentscope-ai/QwenPaw/pull/7640))
- Preserve Unix permission bits for restored secrets and master keys ([#7658](https://github.com/agentscope-ai/QwenPaw/pull/7658))
- Deliver Base64-encoded images and other media correctly across supported chat channels ([#7647](https://github.com/agentscope-ai/QwenPaw/pull/7647))

## New Contributors

- [@kic635](https://github.com/kic635)
- [@wuyak](https://github.com/wuyak)
- [@Bruce-Yii](https://github.com/Bruce-Yii)
- [@kabishou11](https://github.com/kabishou11)

**Full Changelog**: [v2.2.0...v2.2.1](https://github.com/agentscope-ai/QwenPaw/compare/v2.2.0...v2.2.1)