Category: provider
Model Studio Qwen Deep Research
Validation
CODEBLOCK0
Pass criteria: command exits 0 and output/aliyun-qwen-deep-research/validate.txt is generated.
Output And Evidence
- Save research goals, confirmation answers, normalized request payloads, and final report snapshots under output/aliyun-qwen-deep-research/. Keep the exact model, region, and enable_feedback setting with each saved run. Use this skill when the user wants a deep, multi-stage research workflow rather than a single chat completion.
Critical model names
Use one of these exact model strings:
Selection guidance:
- Use qwen-deep-research for the current mainline model. Use qwen-deep-research-2025-12-15 when you need the snapshot with MCP tool-calling support and stronger reproducibility. Prerequisites
- Install SDK in a virtual environment: CODEBLOCK1
- Set DASHSCOPE_API_KEY in your environment, or add dashscope_api_key to ~/.alibabacloud/credentials. This model currently applies to the China mainland (Beijing) region and uses its own API shape rather than OpenAI-compatible mode. Normalized interface (research.run)
Request
- topic (string, required) INLINECODE11 (string, optional): default INLINECODE12 INLINECODE13 (array, optional) INLINECODE14 (bool, optional): default INLINECODE15 INLINECODE16 (bool, optional): must be INLINECODE17 INLINECODE18 (array, optional): image URLs and related context Response
- status (string): stage status such as thinking, researching, or INLINECODE22 INLINECODE23 (string, optional): streamed content chunk INLINECODE24 (string, optional): final structured research report INLINECODE25 (object, optional) Quick start
CODEBLOCK2
Operational guidance
- Expect streaming output only. Keep the initial topic concrete and bounded; broad topics can trigger long iterative search plans. If the model asks follow-up questions and you already know the constraints, answer them explicitly to avoid wasted rounds. Use the snapshot model when you need stable evaluation runs or MCP tool-calling support. Output location
- Default output: INLINECODE26 Override base dir with OUTPUT_DIR. References
技能名称: aliyun-qwen-deep-research
详细描述:
Model Studio Qwen Deep Research
验证
bashcompile skills/ai/research/aliyun-qwen-deep-research/scripts/prepare deepresearch request.py && echo pycompile ok > output/aliyun-qwen-deep-research/validate.txt
通过标准：命令退出码为0，且生成了 output/aliyun-qwen-deep-research/validate.txt 文件。
输出与证据
- 将研究目标、确认答案、标准化请求负载以及最终报告快照保存在 output/aliyun-qwen-deep-research/ 目录下。 每次保存运行时，需保留确切的模型、区域以及 enable_feedback 设置。 当用户需要深度、多阶段的研究工作流，而非单次对话补全时，请使用此技能。
关键模型名称
使用以下确切的模型字符串之一：
- qwen-deep-research qwen-deep-research-2025-12-15 选择指南：
- 对于当前主线模型，使用 qwen-deep-research。 当需要支持MCP工具调用且具备更强可复现性的快照时，使用 qwen-deep-research-2025-12-15。 前提条件
bash
- 在环境中设置 DASHSCOPEAPI KEY，或者将 dashscopeapi key 添加到 ~/.alibabacloud/credentials 文件中。 该模型目前适用于中国大陆（北京）区域，并使用其自身的API形态，而非兼容OpenAI的模式。 标准化接口 (research.run)
请求
- topic (字符串，必填) model (字符串，可选)：默认为 qwen-deep-research messages (对象数组，可选) enable_feedback (布尔值，可选)：默认为 true stream (布尔值，可选)：必须为 true attachments (对象数组，可选)：图片URL及相关上下文 响应
- status (字符串)：阶段状态，例如 thinking、researching 或 finished text (字符串，可选)：流式内容片段 report (字符串，可选)：最终的结构化研究报告 raw (对象，可选) 快速开始
bashdeep research_request.py \
操作指南
- 只期望流式输出。 保持初始主题具体且有边界；宽泛的主题可能触发冗长的迭代搜索计划。 如果模型提出后续问题，而您已知约束条件，请明确回答，以避免不必要的轮次。 当您需要稳定的评估运行或MCP工具调用支持时，请使用快照模型。 输出位置
- 默认输出：output/aliyun-qwen-deep-research/requests/ 可通过 OUTPUT_DIR 覆盖基础目录。 参考资料
