research

Compare original and translation side by side

🇺🇸

Original

English

🇨🇳

Translation

Chinese

Research Skill

研究Skill

Orchestrate parallel scientist agents for comprehensive research workflows with optional AUTO mode for fully autonomous execution.

编排并行Scientist Agent以实现全面的研究工作流，可选AUTO模式支持完全自主执行。

Overview

概述

Research is a multi-stage workflow that decomposes complex research goals into parallel investigations:

Decomposition - Break research goal into independent stages/hypotheses
Execution - Run parallel scientist agents on each stage
Verification - Cross-validate findings, check consistency
Synthesis - Aggregate results into comprehensive report

研究是一个多阶段工作流，将复杂研究目标分解为并行研究任务：

分解 - 将研究目标拆分为独立的阶段/假设
执行 - 为每个阶段运行并行Scientist Agent
验证 - 交叉验证研究结果，检查一致性
合成 - 将结果汇总为全面报告

Usage Examples

使用示例

/oh-my-claudecode:research <goal>                    # Standard research with user checkpoints
/oh-my-claudecode:research AUTO: <goal>              # Fully autonomous until complete
/oh-my-claudecode:research status                    # Check current research session status
/oh-my-claudecode:research resume                    # Resume interrupted research session
/oh-my-claudecode:research list                      # List all research sessions
/oh-my-claudecode:research report <session-id>       # Generate report for session

/oh-my-claudecode:research <goal>                    # 带用户检查点的标准研究
/oh-my-claudecode:research AUTO: <goal>              # 完全自主执行直至完成
/oh-my-claudecode:research status                    # 查看当前研究会话状态
/oh-my-claudecode:research resume                    # 恢复中断的研究会话
/oh-my-claudecode:research list                      # 列出所有研究会话
/oh-my-claudecode:research report <session-id>       # 为指定会话生成报告

Quick Examples

快速示例

/oh-my-claudecode:research What are the performance characteristics of different sorting algorithms?
/oh-my-claudecode:research AUTO: Analyze authentication patterns in this codebase
/oh-my-claudecode:research How does the error handling work across the API layer?

/oh-my-claudecode:research What are the performance characteristics of different sorting algorithms?
/oh-my-claudecode:research AUTO: Analyze authentication patterns in this codebase
/oh-my-claudecode:research How does the error handling work across the API layer?

Research Protocol

研究协议

Stage Decomposition Pattern

阶段分解模式

When given a research goal, decompose into 3-7 independent stages:

markdown

undefined

给定研究目标后，将其分解为3-7个独立阶段：

markdown

undefined

Research Decomposition

研究分解

Goal: <original research goal>

目标: <原始研究目标>

Stage 1: <stage-name>

阶段1: <阶段名称>

Focus: What this stage investigates
Hypothesis: Expected finding (if applicable)
Scope: Files/areas to examine
Tier: LOW | MEDIUM | HIGH

重点: 本阶段研究内容
假设: 预期发现（如适用）
范围: 需检查的文件/领域
等级: LOW | MEDIUM | HIGH

Stage 2: <stage-name>

阶段2: <阶段名称>

...

undefined

...

undefined

Parallel Scientist Invocation

并行Scientist调用

Fire independent stages in parallel via Task tool:

// Stage 1 - Simple data gathering
Task(subagent_type="oh-my-claudecode:scientist", model="haiku", prompt="[RESEARCH_STAGE:1] Investigate...")

// Stage 2 - Standard analysis
Task(subagent_type="oh-my-claudecode:scientist", model="sonnet", prompt="[RESEARCH_STAGE:2] Analyze...")

// Stage 3 - Complex reasoning
Task(subagent_type="oh-my-claudecode:scientist-high", model="opus", prompt="[RESEARCH_STAGE:3] Deep analysis of...")

通过Task工具并行触发独立阶段：

// 阶段1 - 简单数据收集
Task(subagent_type="oh-my-claudecode:scientist", model="haiku", prompt="[RESEARCH_STAGE:1] Investigate...")

// 阶段2 - 标准分析
Task(subagent_type="oh-my-claudecode:scientist", model="sonnet", prompt="[RESEARCH_STAGE:2] Analyze...")

// 阶段3 - 复杂推理
Task(subagent_type="oh-my-claudecode:scientist-high", model="opus", prompt="[RESEARCH_STAGE:3] Deep analysis of...")

Smart Model Routing

智能模型路由

CRITICAL: Always pass
model
parameter explicitly!

Task Complexity	Agent	Model	Use For
Data gathering	`scientist` (model=haiku)	haiku	File enumeration, pattern counting, simple lookups
Standard analysis	`scientist`	sonnet	Code analysis, pattern detection, documentation review
Complex reasoning	`scientist-high`	opus	Architecture analysis, cross-cutting concerns, hypothesis validation

重要提示：务必显式传递
model
参数！

任务复杂度	Agent	模型	适用场景
数据收集	`scientist` (model=haiku)	haiku	文件枚举、模式统计、简单查询
标准分析	`scientist`	sonnet	代码分析、模式检测、文档审阅
复杂推理	`scientist-high`	opus	架构分析、横切关注点研究、假设验证

Routing Decision Guide

路由决策指南

Research Task	Tier	Example Prompt
"Count occurrences of X"	LOW	"Count all usages of useState hook"
"Find all files matching Y"	LOW	"List all test files in the project"
"Analyze pattern Z"	MEDIUM	"Analyze error handling patterns in API routes"
"Document how W works"	MEDIUM	"Document the authentication flow"
"Explain why X happens"	HIGH	"Explain why race conditions occur in the cache layer"
"Compare approaches A vs B"	HIGH	"Compare Redux vs Context for state management here"

研究任务	等级	示例提示
"统计X的出现次数"	LOW	"统计useState钩子的所有用法"
"查找所有匹配Y的文件"	LOW	"列出项目中的所有测试文件"
"分析模式Z"	MEDIUM	"分析API路由中的错误处理模式"
"记录W的工作原理"	MEDIUM	"记录认证流程"
"解释X发生的原因"	HIGH	"解释缓存层中竞态条件发生的原因"
"比较方法A与B"	HIGH	"比较Redux与Context在此处的状态管理效果"

Verification Loop

验证循环

After parallel execution completes, verify findings:

// Cross-validation stage
Task(subagent_type="oh-my-claudecode:scientist", model="sonnet", prompt="
[RESEARCH_VERIFICATION]
Cross-validate these findings for consistency:

Stage 1 findings: <summary>
Stage 2 findings: <summary>
Stage 3 findings: <summary>

Check for:
1. Contradictions between stages
2. Missing connections
3. Gaps in coverage
4. Evidence quality

Output: [VERIFIED] or [CONFLICTS:<list>]
")

并行执行完成后，验证研究结果：

// 交叉验证阶段
Task(subagent_type="oh-my-claudecode:scientist", model="sonnet", prompt="
[RESEARCH_VERIFICATION]
交叉验证以下结果的一致性：

阶段1结果: <摘要>
阶段2结果: <摘要>
阶段3结果: <摘要>

检查内容：
1. 各阶段之间的矛盾
2. 缺失的关联
3. 覆盖缺口
4. 证据质量

输出: [VERIFIED] 或 [CONFLICTS:<列表>]
")

AUTO Mode

AUTO模式

AUTO mode runs the complete research workflow autonomously with loop control.

AUTO模式通过循环控制自主运行完整的研究工作流。

Loop Control Protocol

循环控制协议

[RESEARCH + AUTO - ITERATION {{ITERATION}}/{{MAX}}]

Your previous attempt did not output the completion promise. Continue working.

Current state: {{STATE}}
Completed stages: {{COMPLETED_STAGES}}
Pending stages: {{PENDING_STAGES}}

[RESEARCH + AUTO - 迭代 {{ITERATION}}/{{MAX}}]

您上一次尝试未输出完成标记，请继续执行。

当前状态: {{STATE}}
已完成阶段: {{COMPLETED_STAGES}}
待处理阶段: {{PENDING_STAGES}}

Promise Tags

标记说明

Tag	Meaning	When to Use
`[PROMISE:RESEARCH_COMPLETE]`	Research finished successfully	All stages done, verified, report generated
`[PROMISE:RESEARCH_BLOCKED]`	Cannot proceed	Missing data, access issues, circular dependency

标记	含义	使用场景
`[PROMISE:RESEARCH_COMPLETE]`	研究成功完成	所有阶段完成、验证通过、报告生成完毕
`[PROMISE:RESEARCH_BLOCKED]`	无法继续执行	数据缺失、访问问题、循环依赖

AUTO Mode Rules

AUTO模式规则

Max Iterations: 10 (configurable)
Continue until: Promise tag emitted OR max iterations
State tracking: Persist after each stage completion
Cancellation:
```
/oh-my-claudecode:cancel
```
or "stop", "cancel"

最大迭代次数: 10次（可配置）
终止条件: 输出标记或达到最大迭代次数
状态跟踪: 每个阶段完成后持久化状态
取消方式:
```
/oh-my-claudecode:cancel
```
或输入 "stop"、"cancel"

AUTO Mode Example

AUTO模式示例

/oh-my-claudecode:research AUTO: Comprehensive security analysis of the authentication system

[Decomposition]
- Stage 1 (LOW): Enumerate auth-related files
- Stage 2 (MEDIUM): Analyze token handling
- Stage 3 (MEDIUM): Review session management
- Stage 4 (HIGH): Identify vulnerability patterns
- Stage 5 (MEDIUM): Document security controls

[Execution - Parallel]
Firing stages 1-3 in parallel...
Firing stages 4-5 after dependencies complete...

[Verification]
Cross-validating findings...

[Synthesis]
Generating report...

[PROMISE:RESEARCH_COMPLETE]

/oh-my-claudecode:research AUTO: Comprehensive security analysis of the authentication system

[分解]
- 阶段1 (LOW): 枚举认证相关文件
- 阶段2 (MEDIUM): 分析令牌处理逻辑
- 阶段3 (MEDIUM): 审阅会话管理机制
- 阶段4 (HIGH): 识别漏洞模式
- 阶段5 (MEDIUM): 记录安全控制措施

[执行 - 并行]
并行触发阶段1-3...
依赖完成后触发阶段4-5...

[验证]
交叉验证结果...

[合成]
生成报告...

[PROMISE:RESEARCH_COMPLETE]

Parallel Execution Patterns

并行执行模式

Independent Dataset Analysis (Parallel)

独立数据集分析（并行）

When stages analyze different data sources:

// All fire simultaneously
Task(subagent_type="oh-my-claudecode:scientist", model="haiku", prompt="[STAGE:1] Analyze src/api/...")
Task(subagent_type="oh-my-claudecode:scientist", model="haiku", prompt="[STAGE:2] Analyze src/utils/...")
Task(subagent_type="oh-my-claudecode:scientist", model="haiku", prompt="[STAGE:3] Analyze src/components/...")

当各阶段分析不同数据源时：

// 同时触发所有任务
Task(subagent_type="oh-my-claudecode:scientist", model="haiku", prompt="[STAGE:1] Analyze src/api/...")
Task(subagent_type="oh-my-claudecode:scientist", model="haiku", prompt="[STAGE:2] Analyze src/utils/...")
Task(subagent_type="oh-my-claudecode:scientist", model="haiku", prompt="[STAGE:3] Analyze src/components/...")

Hypothesis Battery (Parallel)

假设验证组（并行）

When testing multiple hypotheses:

// Test hypotheses simultaneously
Task(subagent_type="oh-my-claudecode:scientist", model="sonnet", prompt="[HYPOTHESIS:A] Test if caching improves...")
Task(subagent_type="oh-my-claudecode:scientist", model="sonnet", prompt="[HYPOTHESIS:B] Test if batching reduces...")
Task(subagent_type="oh-my-claudecode:scientist", model="sonnet", prompt="[HYPOTHESIS:C] Test if lazy loading helps...")

当测试多个假设时：

// 同时测试多个假设
Task(subagent_type="oh-my-claudecode:scientist", model="sonnet", prompt="[HYPOTHESIS:A] Test if caching improves...")
Task(subagent_type="oh-my-claudecode:scientist", model="sonnet", prompt="[HYPOTHESIS:B] Test if batching reduces...")
Task(subagent_type="oh-my-claudecode:scientist", model="sonnet", prompt="[HYPOTHESIS:C] Test if lazy loading helps...")

Cross-Validation (Sequential)

交叉验证（串行）

When verification depends on all findings:

// Wait for all parallel stages
[stages complete]

// Then sequential verification
Task(subagent_type="oh-my-claudecode:scientist-high", model="opus", prompt="
[CROSS_VALIDATION]
Validate consistency across all findings:
- Finding 1: ...
- Finding 2: ...
- Finding 3: ...
")

当验证依赖所有研究结果时：

// 等待所有并行阶段完成
[stages complete]

// 然后执行串行验证
Task(subagent_type="oh-my-claudecode:scientist-high", model="opus", prompt="
[CROSS_VALIDATION]
验证所有结果的一致性：
- 结果1: ...
- 结果2: ...
- 结果3: ...
")

Concurrency Limit

并发限制

Maximum 20 concurrent scientist agents to prevent resource exhaustion.

If more than 20 stages, batch them:

Batch 1: Stages 1-5 (parallel)
[wait for completion]
Batch 2: Stages 6-7 (parallel)

最多允许20个并发Scientist Agent，以防止资源耗尽。

如果阶段超过20个，需分批处理：

批次1: 阶段1-5（并行）
[等待完成]
批次2: 阶段6-7（并行）

Session Management

会话管理

Directory Structure

目录结构

.omc/research/{session-id}/
  state.json              # Session state and progress
  stages/
    stage-1.md            # Stage 1 findings
    stage-2.md            # Stage 2 findings
    ...
  findings/
    raw/                  # Raw findings from scientists
    verified/             # Post-verification findings
  figures/
    figure-1.png          # Generated visualizations
    ...
  report.md               # Final synthesized report

.omc/research/{session-id}/
  state.json              # 会话状态与进度
  stages/
    stage-1.md            # 阶段1结果
    stage-2.md            # 阶段2结果
    ...
  findings/
    raw/                  # Scientist返回的原始结果
    verified/             # 验证后的结果
  figures/
    figure-1.png          # 生成的可视化图表
    ...
  report.md               # 最终合成报告

State File Format

状态文件格式

json

{
  "id": "research-20240115-abc123",
  "goal": "Original research goal",
  "status": "in_progress | complete | blocked | cancelled",
  "mode": "standard | auto",
  "iteration": 3,
  "maxIterations": 10,
  "stages": [
    {
      "id": 1,
      "name": "Stage name",
      "tier": "LOW | MEDIUM | HIGH",
      "status": "pending | running | complete | failed",
      "startedAt": "ISO timestamp",
      "completedAt": "ISO timestamp",
      "findingsFile": "stages/stage-1.md"
    }
  ],
  "verification": {
    "status": "pending | passed | failed",
    "conflicts": [],
    "completedAt": "ISO timestamp"
  },
  "createdAt": "ISO timestamp",
  "updatedAt": "ISO timestamp"
}

json

{
  "id": "research-20240115-abc123",
  "goal": "原始研究目标",
  "status": "in_progress | complete | blocked | cancelled",
  "mode": "standard | auto",
  "iteration": 3,
  "maxIterations": 10,
  "stages": [
    {
      "id": 1,
      "name": "Stage name",
      "tier": "LOW | MEDIUM | HIGH",
      "status": "pending | running | complete | failed",
      "startedAt": "ISO timestamp",
      "completedAt": "ISO timestamp",
      "findingsFile": "stages/stage-1.md"
    }
  ],
  "verification": {
    "status": "pending | passed | failed",
    "conflicts": [],
    "completedAt": "ISO timestamp"
  },
  "createdAt": "ISO timestamp",
  "updatedAt": "ISO timestamp"
}

Session Commands

会话命令

Command	Action
`/oh-my-claudecode:research status`	Show current session progress
`/oh-my-claudecode:research resume`	Resume most recent interrupted session
`/oh-my-claudecode:research resume <session-id>`	Resume specific session
`/oh-my-claudecode:research list`	List all sessions with status
`/oh-my-claudecode:research report <session-id>`	Generate/regenerate report
`/oh-my-claudecode:research cancel`	Cancel current session (preserves state)

命令	操作
`/oh-my-claudecode:research status`	显示当前会话进度
`/oh-my-claudecode:research resume`	恢复最近中断的会话
`/oh-my-claudecode:research resume <session-id>`	恢复指定会话
`/oh-my-claudecode:research list`	列出所有会话及其状态
`/oh-my-claudecode:research report <session-id>`	生成/重新生成报告
`/oh-my-claudecode:research cancel`	取消当前会话（保留状态）

Tag Extraction

标签提取

Scientists use structured tags for findings. Extract them with these patterns:

Scientist会为结果添加结构化标签，可通过以下模式提取：

Finding Tags

结果标签

[FINDING:<id>] <title>
<evidence and analysis>
[/FINDING]

[EVIDENCE:<finding-id>]
- File: <path>
- Lines: <range>
- Content: <relevant code/text>
[/EVIDENCE]

[CONFIDENCE:<level>] # HIGH | MEDIUM | LOW
<reasoning for confidence level>

[FINDING:<id>] <标题>
<证据与分析>
[/FINDING]

[EVIDENCE:<finding-id>]
- 文件: <路径>
- 行号: <范围>
- 内容: <相关代码/文本>
[/EVIDENCE]

[CONFIDENCE:<level>] # HIGH | MEDIUM | LOW
<置信度等级的推理依据>

Extraction Regex Patterns

提取正则表达式

javascript

// Finding extraction
const findingPattern = /\[FINDING:(\w+)\]\s*(.*?)\n([\s\S]*?)\[\/FINDING\]/g;

// Evidence extraction
const evidencePattern = /\[EVIDENCE:(\w+)\]([\s\S]*?)\[\/EVIDENCE\]/g;

// Confidence extraction
const confidencePattern = /\[CONFIDENCE:(HIGH|MEDIUM|LOW)\]\s*(.*)/g;

// Stage completion
const stageCompletePattern = /\[STAGE_COMPLETE:(\d+)\]/;

// Verification result
const verificationPattern = /\[(VERIFIED|CONFLICTS):?(.*?)\]/;

javascript

// 结果提取
const findingPattern = /\[FINDING:(\w+)\]\s*(.*?)\n([\s\S]*?)\[\/FINDING\]/g;

// 证据提取
const evidencePattern = /\[EVIDENCE:(\w+)\]([\s\S]*?)\[\/EVIDENCE\]/g;

// 置信度提取
const confidencePattern = /\[CONFIDENCE:(HIGH|MEDIUM|LOW)\]\s*(.*)/g;

// 阶段完成标记
const stageCompletePattern = /\[STAGE_COMPLETE:(\d+)\]/;

// 验证结果标记
const verificationPattern = /\[(VERIFIED|CONFLICTS):?(.*?)\]/;

Evidence Window

证据上下文范围

When extracting evidence, include context window:

[EVIDENCE:F1]
- File: /src/auth/login.ts
- Lines: 45-52 (context: 40-57)
- Content:
  ```typescript
  // Lines 45-52 with 5 lines context above/below

[/EVIDENCE]

undefined

提取证据时需包含上下文范围：

[EVIDENCE:F1]
- 文件: /src/auth/login.ts
- 行号: 45-52（上下文: 40-57）
- 内容:
  ```typescript
  // 包含上下5行的45-52行代码

[/EVIDENCE]

undefined

Quality Validation

质量验证

Findings must meet quality threshold:

Quality Check	Requirement
Evidence present	At least 1 [EVIDENCE] per [FINDING]
Confidence stated	Each finding has [CONFIDENCE]
Source cited	File paths are absolute and valid
Reproducible	Another agent could verify

结果必须满足以下质量阈值：

质量检查	要求
存在证据	每个[FINDING]至少对应1个[EVIDENCE]
声明置信度	每个结果都有[CONFIDENCE]标记
引用来源	文件路径为绝对路径且有效
可复现	其他Agent可验证结果

Report Generation

报告生成

Report Template

报告模板

markdown

undefined

markdown

undefined

Research Report: {{GOAL}}

研究报告: {{GOAL}}

Session ID: {{SESSION_ID}} Date: {{DATE}} Status: {{STATUS}}

会话ID: {{SESSION_ID}} 日期: {{DATE}} 状态: {{STATUS}}

Executive Summary

执行摘要

Methodology

研究方法

Research Stages

研究阶段

Stage	Focus	Tier	Status
{{STAGES_TABLE}}

阶段	重点	等级	状态
{{STAGES_TABLE}}

Approach

研究思路

{{Description of decomposition rationale and execution strategy}}

Key Findings

关键结果

Finding 1: {{TITLE}}

结果1: {{标题}}

Confidence: {{HIGH|MEDIUM|LOW}}

置信度: {{HIGH|MEDIUM|LOW}}

Evidence

证据

Finding 2: {{TITLE}}

结果2: {{标题}}

...

Visualizations

可视化图表

Cross-Validation Results

交叉验证结果

Limitations

局限性

{{Limitation 1}}
{{Limitation 2}}
{{Areas not covered and why}}

{{局限性1}}
{{局限性2}}
{{未覆盖领域及原因}}

Recommendations

建议

{{Actionable recommendation}}
{{Actionable recommendation}}

{{可执行建议}}
{{可执行建议}}

Appendix

附录

Raw Data

原始数据

Session State

会话状态

undefined

undefined

Figure Embedding Protocol

图表嵌入协议

Scientists generate visualizations using this marker:

[FIGURE:path/to/figure.png]
Caption: Description of what the figure shows
Alt: Accessibility description
[/FIGURE]

Report generator embeds figures:

markdown

undefined

Scientist使用以下标记生成可视化图表：

[FIGURE:path/to/figure.png]
Caption: 图表说明
Alt: 无障碍描述
[/FIGURE]

报告生成器会嵌入图表：

markdown

undefined

Visualizations

可视化图表

Caption: Description of what the figure shows

Caption: Description of what the figure shows

undefined

说明: 图表展示内容

说明: 图表展示内容

undefined

Figure Types

图表类型

Type	Use For	Generated By
Architecture diagram	System structure	scientist-high
Flow chart	Process flows	scientist
Dependency graph	Module relationships	scientist
Timeline	Sequence of events	scientist
Comparison table	A vs B analysis	scientist

类型	适用场景	生成Agent
架构图	系统结构展示	scientist-high
流程图	流程展示	scientist
依赖图	模块关系展示	scientist
时间线	事件序列展示	scientist
对比表	A与B分析对比	scientist

Configuration

配置

Optional settings in

.claude/settings.json

json

{
  "omc": {
    "research": {
      "maxIterations": 10,
      "maxConcurrentScientists": 5,
      "defaultTier": "MEDIUM",
      "autoVerify": true,
      "generateFigures": true,
      "evidenceContextLines": 5
    }
  }
}

可在

.claude/settings.json

中设置可选参数：

json

{
  "omc": {
    "research": {
      "maxIterations": 10,
      "maxConcurrentScientists": 5,
      "defaultTier": "MEDIUM",
      "autoVerify": true,
      "generateFigures": true,
      "evidenceContextLines": 5
    }
  }
}

Cancellation

取消操作

/oh-my-claudecode:cancel

Or say: "stop research", "cancel research", "abort"

Progress is preserved in

.omc/research/{session-id}/

for resume.

/oh-my-claudecode:cancel

或输入："stop research"、"cancel research"、"abort"

进度会保存在

.omc/research/{session-id}/

中，以便后续恢复。

Troubleshooting

故障排除

Stuck in verification loop?

Check for conflicting findings between stages
Review state.json for specific conflicts
May need to re-run specific stages with different approach

Scientists returning low-quality findings?

Check tier assignment - complex analysis needs HIGH tier
Ensure prompts include clear scope and expected output format
Review if research goal is too broad

AUTO mode exhausted iterations?

Review state to see where it's stuck
Check if goal is achievable with available data
Consider breaking into smaller research sessions

Missing figures in report?

Verify figures/ directory exists
Check [FIGURE:] tags in findings
Ensure paths are relative to session directory

验证循环卡住？

检查各阶段结果之间是否存在冲突
查看state.json中的具体冲突
可能需要以不同方法重新运行特定阶段

Scientist返回低质量结果？

检查等级分配 - 复杂分析需要HIGH等级
确保提示包含明确的范围和预期输出格式
检查研究目标是否过于宽泛

AUTO模式耗尽迭代次数？

查看状态以确定卡住的位置
检查目标是否可通过现有数据实现
考虑拆分为更小的研究会话

报告中缺少图表？

验证figures/目录是否存在
检查结果中的[FIGURE:]标记
确保路径相对于会话目录