Skip to content

Weizhena/Deep-Research-skills / skills/research-zh/research-deep

research-deep

Verified

读取调研outline,为每个item启动独立agent进行深度调研。禁用task output。

Research Deep - 深度调研

触发方式

/research-deep

执行流程

Step 1: 自动定位Outline

在当前工作目录查找 */outline.yaml 文件,读取items列表、execution配置(含items_per_agent)。

Step 2: 断点续传检查

  • 检查output_dir下已完成的JSON文件
  • 跳过已完成的items

Step 3: 分批执行

  • 按batch_size分批(完成一批需要得到用户同意才可进行下一批)
  • 每个agent负责items_per_agent个项目
  • 启动web-search-agent(后台并行,禁用task output)

参数获取:

  • {topic}: outline.yaml中的topic字段
  • {item_name}: item的name字段
  • {item_related_info}: item的完整yaml内容(name + category + description等)
  • {output_dir}: outline.yaml中execution.output_dir(默认./results)
  • {fields_path}: {topic}/fields.yaml的绝对路径
  • {output_path}: {output_dir}/{item_name_slug}.json的绝对路径(slugify处理item_name:空格替换为_,移除特殊字符)

硬约束:以下prompt必须严格复述,仅替换{xxx}中的变量,禁止改写结构或措辞。

Prompt模板:

prompt = f"""## 任务
调研 {item_related_info},输出结构化JSON到 {output_path}

## 字段定义
读取 {fields_path} 获取所有字段定义

## 输出要求
1. 按fields.yaml定义的字段输出JSON
2. 不确定的字段值标注[不确定]
3. JSON末尾添加uncertain数组,列出所有不确定的字段名
4. 所有字段值必须使用中文输出(调研过程可用英文,但最终JSON值为中文)

## 输出路径
{output_path}

## 验证
完成JSON输出后,运行验证脚本确保字段完整覆盖:
python ~/.claude/skills/research/validate_json.py -f {fields_path} -j {output_path}
验证通过后才算完成任务。
"""

One-shot示例(假设调研GitHub Copilot):

## 任务
调研 name: GitHub Copilot
category: 国际产品
description: Microsoft/GitHub开发,首个主流AI编程助手,市场份额约40%,输出结构化JSON到 {project_dir}/results/GitHub_Copilot.json

## 字段定义
读取 {project_dir}/fields.yaml 获取所有字段定义

## 输出要求
1. 按fields.yaml定义的字段输出JSON
2. 不确定的字段值标注[不确定]
3. JSON末尾添加uncertain数组,列出所有不确定的字段名
4. 所有字段值必须使用中文输出(调研过程可用英文,但最终JSON值为中文)

## 输出路径
{project_dir}/results/GitHub_Copilot.json

## 验证
完成JSON输出后,运行验证脚本确保字段完整覆盖:
python ~/.claude/skills/research/validate_json.py -f {project_dir}/fields.yaml -j {project_dir}/results/GitHub_Copilot.json
验证通过后才算完成任务。

Step 4: 等待与监控

  • 等待当前批次完成
  • 启动下一批
  • 显示进度

Step 5: 汇总报告

全部完成后输出:

  • 完成数量
  • 失败/不确定标记的items
  • 输出目录

Agent配置

  • 后台执行: 是
  • Task Output: 禁用(agent完成时有明确输出文件)
  • 断点续传: 是
reverse-engineer-anythingmorluto/reaReverse engineer native, managed, Electron/JavaScript, packaged, firmware, and browser targets with REA, and build agent environments (AgentEnv bundles) of web applications from a live instance or a capture, and of Linux desktop apps with a task from a request. Use shipped-artifact or requested runtime evidence to explain features, compare versions, decompile code, or guide a reconstruction. Skip REA for ordinary source-repository architecture analysis.Stars: 471kUpdated Claude CodeCodexVerifiedInstall
diagnosing-superpowersobra/superpowersUse when a superpowers session went wrong and your human partner wants to know why — repeated work, ignored plans, stumbles, poor results, a skill that didn't fire, "it took too long", "why is it so expensive", "what is it doing" — or wants to build a bug report for the superpowers maintainers, for the current session or a past one identified by id or path, on any harness.Stars: 322kUpdated Claude CodeCodexVerifiedInstall
dispatching-parallel-agentsobra/superpowersUse when facing 2+ independent tasks that can be worked on without shared state or sequential dependenciesStars: 322kUpdated Claude CodeCodexVerifiedInstall
executing-plansobra/superpowersUse when executing an implementation plan in the current session as the implementer yourself — your human partner chose inline execution, or no subagent tool is availableStars: 322kUpdated Claude CodeCodexVerifiedInstall