研究问题解析 (Agentic AI 科研平台)

解析用户的自然语言临床科研问题，输出结构化研究参数（研究类型、终点、变量、推荐workflow）。当用户提出一个科研问题并需要将其转为结构化任务时触发。

Security Scan

The skill's stated purpose is plausible, but its runtime instructions ask the agent to read a specific local file and POST status to a localhost endpoint—actions that are unnecessary for simple question parsing and could access local data or services without clear justification.

Purpose & Capability

Name/description: parse clinical research questions into structured parameters — this is coherent. However, the SKILL.md requires reading a fixed filesystem path (/home/ubuntu/workspace/demo/mock_data/task_create.json) and calling http://localhost:5001/api/report for start/complete reports. Neither the fixed local file nor the local HTTP reporting are explained by the skill description and are not strictly necessary to perform natural-language parsing, so they are unexpected behavior even if not overtly malicious.

Instruction Scope

The instructions explicitly tell the agent to: (1) curl a localhost HTTP endpoint twice (start and complete reports) and (2) read a hard-coded file path on disk and use its contents to form output. These actions introduce scope creep: they access local services and local files outside the simple parse/transform purpose. The instructions also require the agent to display fields from that mock JSON rather than parse the user's live input in some cases, which may be surprising to users.

Install Mechanism

This is an instruction-only skill with no install spec and no code files, so nothing is written to disk and no external packages are pulled in. That is low-risk from installation perspective.

Credentials

The skill requests no environment variables, no credentials, and no declared config paths. However, it does implicitly require read access to a specific filesystem location and network access to localhost:5001 at runtime — these implicit resource requirements are not declared in metadata and should be considered by users.

Persistence & Privilege

The skill is not marked always:true and uses normal user-invocable/autonomous defaults. It does not request persistent platform privileges or modify other skills. The main privilege concern is the runtime ability to access local files and localhost services, which is not a platform-level persistence request but a runtime behavior to be aware of.

What to consider before installing

Before installing or enabling this skill, be aware it will (without asking) try to: 1) read /home/ubuntu/workspace/demo/mock_data/task_create.json on the host, and 2) POST small JSON reports to http://localhost:5001/api/report. These actions are not explained in the public description and are not strictly necessary to parse a user question. If you plan to use it: - Inspect the mock_data file to confirm it contains only non-sensitive demo data. - Confirm what (if anything) is listening on localhost:5001 in your environment; the skill will contact it. - If you do not want the skill contacting local services or reading files, ask the author to remove those steps or to make them optional/conditional. - Run the skill in a sandboxed agent or with limited filesystem/network permissions first. Overall: the skill could be legitimate for a demo pipeline, but the hard-coded local file path and localhost callbacks are unexpected and warrant caution.

研究问题解析 Skill

何时使用

当用户输入一个自然语言的临床科研问题（如「评估胃癌术后感染风险」「分析某某治疗方案差异」），需要将其解析为结构化研究任务时，使用本 skill。

执行步骤

1. 上报开始

curl -s -X POST http://localhost:5001/api/report \
  -H "Content-Type: application/json" \
  -d '{"skill":"research-question-parser","display_name":"研究问题解析","status":"running","message":"正在解析研究问题，提取结构化参数..."}'

2. 读取数据并输出结果

读取文件 /home/ubuntu/workspace/demo/mock_data/task_create.json。

根据其中的数据，向用户展示以下内容（使用清晰的格式化输出）：

• 原始研究问题：task_input.raw_query
• 结构化参数：时间范围、疾病领域、手术类型、主要终点、对比分组、输出偏好（来自 task_form）
• 任务解析预览：研究类型、目标人群、主要结局、候选变量列表（来自 task_parse_preview）
• 推荐 Workflow：列出推荐的执行步骤（来自 workflow_preview）
• 推荐 Skills：如果数据中有 recommended_skills，一并展示

3. 上报完成

curl -s -X POST http://localhost:5001/api/report \
  -H "Content-Type: application/json" \
  -d '{"skill":"research-question-parser","display_name":"研究问题解析","status":"completed","message":"研究问题解析完成，已生成结构化任务参数与推荐 Workflow"}'

输出要求

• 使用结构化格式（表格或分段列表）展示，不要输出原始 JSON
• 如果用户的问题与胃癌/手术后感染无关，仍使用 mock 数据演示流程，但说明这是 demo 数据