research-pipeline

Compare original and translation side by side

🇺🇸

Original

English

🇨🇳

Translation

Chinese

Research Pipeline Skill

Research Pipeline 技能

Purpose

用途

The research-pipeline skill automates the complete research workflow by chaining five specialized skills in sequence. Instead of manually invoking researcher, synthesizer, devils-advocate, fact-checker, and editor with handoffs between each, this skill handles the entire pipeline with structured context passing.

This represents the "pipeline pattern" for skill orchestration: a predefined sequence of skills that together accomplish a complex goal requiring multiple specialized capabilities.

Research Pipeline 技能通过按顺序串联五个专业技能，实现了完整研究工作流的自动化。无需手动依次调用researcher、synthesizer、devils-advocate、fact-checker和editor并在各环节手动交接，本技能会通过结构化的上下文传递处理整个工作流。

这代表了技能编排的「工作流模式」：一套预定义的技能序列，可协同完成需要多种专业能力的复杂目标。

When to Use This Skill

适用场景

Invoke the research-pipeline when:

Complete research workflow needed: You need to go from research question to polished document
Quality matters: The output must withstand scrutiny (adversarial review, fact-checking)
Hands-off execution desired: You want to set the goal and receive the final product
Standard research pipeline fits: The researcher -> synthesizer -> review -> edit flow matches your needs

在以下场景下调用Research Pipeline：

需要完整研究工作流：你需要从研究问题直接得到精炼文档
对质量有要求：输出内容需经得起严格审查（对抗性评审、事实核查）
希望全程自动化：你只需设定目标，直接获取最终成果
匹配标准研究流程：你的需求符合「文献检索→内容合成→评审→编辑」的标准流程

Clear Indicators for Use

明确适用信号

User says "write a literature review on X"
User says "research X and give me a polished summary"
Task requires finding sources, synthesizing findings, and producing publication-quality output
Multi-hour research effort with quality requirements

用户提出「撰写关于X的文献综述」
用户提出「研究X并给我一份精炼的总结」
任务要求查找资料、整合发现并生成可发表级别的内容
需要数小时的高质量研究工作

When NOT to Use

不适用场景

Do not use this skill when:

You only need part of the pipeline (use individual skills instead)
Output is exploratory/draft-only (skip adversarial review and editing)
Custom skill order is needed (use technical-pm orchestration instead)
Parallel research streams are required (use technical-pm with Task tool)
Rapid iteration with user feedback between stages is needed

请勿在以下场景使用本技能：

你只需要工作流中的某一部分（请直接使用对应单个技能）
输出仅用于探索/草稿阶段（可跳过对抗性评审和编辑环节）
需要自定义技能执行顺序（请使用technical-pm进行编排）
需要并行开展多个研究分支（请结合Task工具使用technical-pm）
需要在各环节加入用户反馈进行快速迭代

Personality

特性定位

You are orchestration-focused and quality-driven. You ensure each pipeline stage receives proper context from the previous stage, validate outputs before proceeding, and maintain the user's original goal throughout the workflow. You are patient with long-running tasks but vigilant about quality gates.

You understand that the pipeline is a convenience, not a constraint. If a stage fails validation or requires user input, you pause and escalate rather than producing low-quality output.

本技能以编排为核心、质量为导向。确保每个工作流环节都能获取上一环节的完整上下文，在进入下一环节前验证输出质量，并在整个流程中始终对齐用户的初始目标。对于长时任务保持耐心，但对质量关卡严格把控。

本技能将工作流视为便捷工具而非约束。若某一环节未通过验证或需要用户输入，会暂停流程并反馈，而非生成低质量输出。

Pipeline Architecture

工作流架构

Stage Sequence

环节序列

researcher
    |
    | [handoff: research findings, citations, gaps]
    v
synthesizer
    |
    | [handoff: synthesized document, themes, tensions]
    v
devils-advocate
    |
    | [handoff: reviewed document, challenges addressed]
    v
fact-checker
    |
    | [handoff: verified citations, issues flagged]
    v
editor
    |
    | [final: polished document]
    v
OUTPUT

researcher
    |
    | [交接内容：研究发现、引用文献、研究缺口]
    v
synthesizer
    |
    | [交接内容：合成文档、核心主题、矛盾点]
    v
devils-advocate
    |
    | [交接内容：评审后文档、已解决的质疑]
    v
fact-checker
    |
    | [交接内容：已验证的引用、标记的问题]
    v
editor
    |
    | [最终输出：精炼文档]
    v
OUTPUT

Stage Responsibilities

环节职责

Stage	Skill	Key Output	Quality Gate
1	researcher	Literature review draft with citations	Has citations, addresses topic
2	synthesizer	Integrated analysis with themes	Cross-cutting insights present
3	devils-advocate	Strengthened arguments	Challenges addressed or documented
4	fact-checker	Verified citations	All claims have valid citations
5	editor	Polished prose	CLAUDE.md style compliant

环节	技能	核心输出	质量关卡
1	researcher	带引用的文献综述草稿	包含引用、贴合主题
2	synthesizer	整合分析后的文档及核心主题	呈现跨维度洞察
3	devils-advocate	经过强化的论证内容	已解决质疑或记录不确定性
4	fact-checker	已验证的引用文献	所有论点均有有效引用支持
5	editor	语言精炼的文档	符合CLAUDE.md格式规范

Workflow

工作流程

Step 1: Initialize Pipeline

步骤1：初始化工作流

Parse user goal:

Extract research topic or question
Determine scope (comprehensive vs focused)
Identify any constraints (time, page count, specific focus areas)
Generate workflow_id for tracking

解析用户目标:

提取研究主题或问题
确定范围（全面研究 vs 聚焦研究）
识别约束条件（时间、页数、特定聚焦领域）
生成用于追踪的workflow_id

Archival Compliance Check

归档合规性检查

Before creating the pipeline context, follow the archival compliance check pattern:

Read the reference document:

~/.claude/skills/archive-workflow/references/archival-compliance-check.md

If file not found, use graceful degradation (log warning, proceed without archival check)
Apply the 5-step pattern to all file creation operations

Store archival guidelines in the pipeline context:

yaml

pipeline:
  archival:
    guidelines_present: true/false
    naming_convention: "{from YAML}"
    output_directory_override: "{if archival says docs go elsewhere}"
    enforcement_mode: "advisory"

When setting the output location (

docs/literature/{topic}/

Validate against archival structure guidelines
If violation detected, present batch advisory options
Record the user's choice in the pipeline context
Pass archival_context to all downstream stages via handoff

Create initial context:

yaml

pipeline:
  workflow_id: "research-{uuid}"
  goal: "{user's original goal}"
  topic: "{extracted topic}"
  scope: comprehensive | focused
  constraints:
    max_pages: {N}
    focus_areas: [list]
    time_limit: {hours}
  current_stage: 1
  started_at: "{timestamp}"

在创建工作流上下文前，需执行归档合规性检查：

读取参考文档：

~/.claude/skills/archive-workflow/references/archival-compliance-check.md

若文件不存在，采用降级处理（记录警告，跳过归档检查继续执行）
对所有文件创建操作应用五步检查模式

将归档准则存储到工作流上下文：

yaml

pipeline:
  archival:
    guidelines_present: true/false
    naming_convention: "{来自YAML的配置}"
    output_directory_override: "{若归档要求文档存储到其他位置}"
    enforcement_mode: "advisory"

设置输出位置（

docs/literature/{topic}/

）时：

验证是否符合归档结构准则
若发现违规，提供批量建议选项
将用户的选择记录到工作流上下文
通过交接将archival_context传递给所有下游环节

创建初始上下文:

yaml

pipeline:
  workflow_id: "research-{uuid}"
  goal: "{用户的初始目标}"
  topic: "{提取的研究主题}"
  scope: comprehensive | focused
  constraints:
    max_pages: {N}
    focus_areas: [列表]
    time_limit: {hours}
  current_stage: 1
  started_at: "{时间戳}"

Step 2: Execute Stage 1 - Researcher

步骤2：执行环节1 - Researcher

Invoke:

Skill(researcher, topic="{topic}", constraints={...})

Researcher executes:

Literature search via PubMed, bioRxiv, OpenAlex
Paper reading and note-taking
Draft literature review with Nature-style citations

Validate output:

Document exists at expected location
Contains inline citations (superscripts)
Addresses the stated topic
Minimum length achieved (varies by scope)

If validation fails: Pause pipeline, report issue to user with options:

Retry researcher with narrowed scope
Accept partial output and continue
Abort pipeline

Create handoff document (see Handoff Format section)

调用方式:

Skill(researcher, topic="{topic}", constraints={...})

Researcher执行内容:

通过PubMed、bioRxiv、OpenAlex进行文献检索
论文阅读与笔记记录
生成带有Nature风格引用的文献综述草稿

输出验证:

文档存在于预期位置
包含嵌入式引用（上标格式）
贴合指定研究主题
达到最低长度要求（根据范围不同而变化）

若验证失败: 暂停工作流，向用户报告问题并提供选项：

缩小范围后重试Researcher
接受部分输出并继续流程
终止工作流

创建交接文档（详见交接格式部分）

Step 3: Execute Stage 2 - Synthesizer

步骤3：执行环节2 - Synthesizer

Invoke:

Skill(synthesizer, handoff="{handoff-file}")

Synthesizer executes:

Read researcher's output via handoff
Identify cross-cutting themes
Highlight tensions and contradictions
Draw project-specific implications
Create synthesis document

Validate output:

Document exists
Adds value beyond summary (themes, tensions, implications)
Maintains citations from source
Addresses original goal

Create handoff document

调用方式:

Skill(synthesizer, handoff="{handoff-file}")

Synthesizer执行内容:

通过交接文档读取Researcher的输出
识别跨维度核心主题
突出矛盾点与对立观点
提炼与项目相关的启示
创建合成文档

输出验证:

文档已生成
提供了超出摘要的价值（主题、矛盾点、启示）
保留了原始引用
贴合初始目标

创建交接文档

Step 4: Execute Stage 3 - Devil's Advocate

步骤4：执行环节3 - Devil's Advocate

Invoke:

Skill(devils-advocate, handoff="{handoff-file}")

Devil's Advocate executes:

Identify thesis of synthesized document
Evaluate strategic coherence
Challenge thesis-critical claims
Propose counter-arguments
Document exchange with synthesizer (up to 2 rounds)

Validate output:

Review report generated
Challenges addressed or uncertainty documented
Synthesizer has responded to critical challenges

Create handoff document including:

Which challenges were addressed
Which uncertainties remain
Devil's advocate approval status

调用方式:

Skill(devils-advocate, handoff="{handoff-file}")

Devil's Advocate执行内容:

识别合成文档的核心论点
评估论点的逻辑连贯性
对核心论点提出质疑
提出反论点
记录与Synthesizer的交互（最多2轮）

输出验证:

已生成评审报告
质疑已解决或不确定性已记录
Synthesizer已对关键质疑做出回应

创建交接文档，包含：

已解决的质疑
仍存在的不确定性
Devil's Advocate的审批状态

Step 5: Execute Stage 4 - Fact-Checker

步骤5：执行环节4 - Fact-Checker

Invoke:

Skill(fact-checker, handoff="{handoff-file}")

Fact-Checker executes:

Inventory all quantitative claims
Verify each citation exists and supports claim
Check citation format (superscripts in text)
Flag missing or incorrect citations
Generate verification report

Validate output:

Verification report generated
All critical citations verified
Issues flagged for correction

If issues found:

Minor issues: Correct inline and proceed
Major issues: Return to synthesizer for corrections, then re-verify

Create handoff document

调用方式:

Skill(fact-checker, handoff="{handoff-file}")

Fact-Checker执行内容:

梳理所有量化论点
验证每个引用的存在性及对论点的支持性
检查引用格式（文本中的上标）
标记缺失或错误的引用
生成验证报告

输出验证:

已生成验证报告
所有关键引用已验证
已标记需修正的问题

若发现问题:

小问题：直接修正并继续流程
大问题：返回给Synthesizer修正后，重新验证

创建交接文档

Step 6: Execute Stage 5 - Editor

步骤6：执行环节5 - Editor

Invoke:

Skill(editor, handoff="{handoff-file}")

Editor executes:

Apply CLAUDE.md style guidelines
Convert bullets to prose where appropriate
Add bridging transitions
Verify glossary placement
Final polish for publication

Validate output:

CLAUDE.md checklist passed
Document flows smoothly
Ready for archival/publication

调用方式:

Skill(editor, handoff="{handoff-file}")

Editor执行内容:

应用CLAUDE.md格式准则
酌情将项目符号转换为连贯文本
添加过渡语句
验证术语表位置
为发表做最终润色

输出验证:

通过CLAUDE.md检查清单
文档逻辑流畅
已准备好归档/发表

Step 7: Complete Pipeline

步骤7：完成工作流

Generate completion report:

markdown

undefined

生成完成报告:

markdown

undefined

Research Pipeline Complete

研究工作流已完成

Workflow ID: {workflow_id} Original Goal: {goal} Duration: {total time} Stages Completed: 5/5

工作流ID: {workflow_id} 初始目标: {goal} 总耗时: {total time} 已完成环节: 5/5

Final Output

最终输出

Location: {path to final document}

位置: {最终文档路径}

Pipeline Summary

工作流摘要

Stage	Duration	Status	Notes
Researcher	2h 15m	Complete	8 papers reviewed
Synthesizer	45m	Complete	3 themes identified
Devil's Advocate	30m	Complete	2 challenges, 1 uncertainty
Fact-Checker	20m	Complete	12 citations verified
Editor	25m	Complete	CLAUDE.md compliant

环节	耗时	状态	备注
Researcher	2h 15m	已完成	已评审8篇论文
Synthesizer	45m	已完成	识别出3个核心主题
Devil's Advocate	30m	已完成	2个质疑，1个不确定性
Fact-Checker	20m	已完成	12个引用已验证
Editor	25m	已完成	符合CLAUDE.md规范

Quality Indicators

质量指标

Citations verified: 12/12
Challenges addressed: 2/2
Uncertainties documented: 1
Style compliance: PASS

已验证引用：12/12
已解决质疑：2/2
已记录不确定性：1
格式合规性：通过

Output Files

输出文件

Final document: {path}
Researcher notes: {path}
Synthesis draft: {path}
Review report: {path}
Fact-check report: {path}

undefined

最终文档：{路径}
Researcher笔记：{路径}
合成草稿：{路径}
评审报告：{路径}
事实核查报告：{路径}

undefined

Optional: Git Strategy Advisory

可选：Git策略建议

After generating the completion report, you MAY invoke

git-strategy-advisor

via Task tool in post-work mode to recommend git strategy for the pipeline output files:

Invocation (via Task tool):

Use git-strategy-advisor to determine git strategy for completed work.

mode: post-work

The advisor analyzes the collection of output files (final document, intermediate outputs, notes, reports) and recommends branch strategy, push timing, and PR creation based on the actual scope.

Response handling: Read the advisor's

summary

field. Include in the completion report if available.

Confidence handling: If the advisor returns confidence "none" or "low", silently skip the git strategy section.

Note: git-strategy-advisor analyzes changes within the current git repository only. If pipeline output files are written outside the repository (e.g., to /tmp/), the advisor will not detect them.

This is advisory only. If

git-strategy-advisor

is not available or returns an error, skip this step. Include the advisor's recommendation in the completion report if available.

生成完成报告后，可在工作流后期通过Task工具调用

git-strategy-advisor

，为工作流输出文件推荐Git策略：

调用方式（通过Task工具）:

使用git-strategy-advisor为已完成的工作确定Git策略。

mode: post-work

该工具会分析输出文件集合（最终文档、中间输出、笔记、报告），并根据实际范围推荐分支策略、推送时机及PR创建建议。

响应处理: 读取工具返回的

summary

字段，若可用则加入完成报告。

置信度处理: 若工具返回置信度为「none」或「low」，则跳过Git策略部分。

注意: git-strategy-advisor仅分析当前Git仓库内的变更。若工作流输出文件存储在仓库外（如/tmp/），工具将无法检测到。

此步骤仅为建议。若git-strategy-advisor不可用或返回错误，直接跳过。若可用则将建议加入完成报告。

Handoff Format

交接格式

Each stage-to-stage transition uses the standardized handoff format from technical-pm:

yaml

handoff:
  version: "1.0"
  source_skill: "{previous skill}"
  target_skill: "{next skill}"
  timestamp: "{ISO8601}"
  workflow_id: "{pipeline workflow_id}"

deliverable:
  type: document
  location: "{path to output file}"
  format: markdown
  summary: "{50+ char description of what was produced}"
  checksum: "{sha256}"

context:
  original_goal: "{user's original goal}"
  completed_skills: [list of completed stages]
  focus_areas: [key topics/themes]
  known_gaps: [limitations identified]
  open_questions: [unresolved items]

quality:
  completion_status: complete | partial
  confidence: high | medium | low
  warnings: [concerns for downstream]

Validation before each handoff:

Schema validation: All required fields present
Content validation: Summary >= 50 chars, file exists
Checksum validation: Recompute and compare

On validation failure: STOP pipeline, report to user

环节间的所有过渡均采用technical-pm定义的标准化交接格式：

yaml

handoff:
  version: "1.0"
  source_skill: "{上一技能名称}"
  target_skill: "{下一技能名称}"
  timestamp: "{ISO8601格式时间}"
  workflow_id: "{工作流ID}"

deliverable:
  type: document
  location: "{输出文件路径}"
  format: markdown
  summary: "{50字以上的产出描述}"
  checksum: "{sha256校验值}"

context:
  original_goal: "{用户的初始目标}"
  completed_skills: [已完成环节列表]
  focus_areas: [关键主题/领域]
  known_gaps: [已识别的局限性]
  open_questions: [未解决的问题]

quality:
  completion_status: complete | partial
  confidence: high | medium | low
  warnings: [给下游环节的注意事项]

每次交接前的验证:

schema验证：所有必填字段均存在
内容验证：摘要不少于50字，文件存在
校验值验证：重新计算并对比

若验证失败: 终止工作流，向用户报告

Configuration Options

配置选项

Scope Settings

范围设置

Comprehensive (default):

Researcher: Full literature review (8+ papers)
Synthesizer: Multi-theme analysis
Devil's Advocate: 2 exchange rounds
Fact-Checker: All citations verified
Editor: Full CLAUDE.md polish

Focused:

Researcher: Targeted review (3-5 papers)
Synthesizer: Single-theme summary
Devil's Advocate: 1 exchange round
Fact-Checker: Critical citations only
Editor: Essential polish only

全面模式（默认）:

Researcher：完整文献综述（8篇以上论文）
Synthesizer：多主题分析
Devil's Advocate：2轮交互
Fact-Checker：验证所有引用
Editor：完整CLAUDE.md格式润色

聚焦模式:

Researcher：针对性综述（3-5篇论文）
Synthesizer：单主题摘要
Devil's Advocate：1轮交互
Fact-Checker：仅验证关键引用
Editor：仅做必要润色

Skip Options

跳过环节选项

For experienced users who want to skip stages:

research-pipeline topic="X" --skip=devils-advocate
research-pipeline topic="X" --skip=fact-checker,editor

Warning: Skipping stages reduces quality guarantees. Pipeline will note skipped stages in completion report.

针对有经验的用户，支持跳过特定环节：

research-pipeline topic="X" --skip=devils-advocate
research-pipeline topic="X" --skip=fact-checker,editor

警告: 跳过环节会降低质量保障。工作流会在完成报告中记录跳过的环节。

Output Location

输出位置

Default:

docs/literature/{topic}/

Override:

research-pipeline topic="X" --output="{custom path}"

默认位置:

docs/literature/{topic}/

自定义位置:

research-pipeline topic="X" --output="{自定义路径}"

Error Handling

错误处理

Stage Failure

环节执行失败

If any stage fails:

Preserve completed work: All previous stage outputs kept
Save partial output: Current stage's work-in-progress saved
Pause pipeline: Do NOT proceed to next stage
Report to user:

Pipeline paused: Stage 3 (devils-advocate) failed

Error: Could not identify thesis in synthesizer output

Completed outputs preserved:
- docs/literature/topic/researcher-draft.md
- docs/literature/topic/synthesis.md (partial)

Options:
(A) Retry devils-advocate with clarified thesis
(B) Add thesis statement to synthesis, then retry
(C) Skip devils-advocate, proceed to fact-checker
(D) Abort pipeline (keep completed outputs)

若任意环节执行失败：

保留已完成工作: 保留所有上一环节的输出
保存部分输出: 保存当前环节的半成品
暂停工作流: 不进入下一环节
向用户报告:

工作流已暂停：环节3（devils-advocate）执行失败

错误信息：无法识别Synthesizer输出中的核心论点

已保留的输出：
- docs/literature/topic/researcher-draft.md
- docs/literature/topic/synthesis.md（半成品）

可选操作：
(A) 明确核心论点后重试devils-advocate
(B) 为合成文档添加核心论点后重试
(C) 跳过devils-advocate，进入fact-checker环节
(D) 终止工作流（保留已完成输出）

Timeout Handling

超时处理

For long-running stages (especially researcher):

Progress updates: Every 30 minutes during researcher stage
Timeout threshold: 4 hours for researcher, 1 hour for other stages
On timeout: Pause pipeline, show progress, offer options

针对长时运行环节（尤其是Researcher）:

进度更新: Researcher环节每30分钟更新一次进度
超时阈值: Researcher为4小时，其他环节为1小时
超时处理: 暂停工作流，展示当前进度并提供选项

Interruption Recovery

中断恢复

Pipeline state is saved after each stage completion:

Location:

/tmp/pipeline-state-{workflow_id}.yaml

On resume:

Found interrupted pipeline: research-abc123
Topic: "hepatocyte oxygenation"
Progress: 3/5 stages complete

Completed:
- [x] Researcher
- [x] Synthesizer
- [x] Devil's Advocate
- [ ] Fact-Checker
- [ ] Editor

Options:
(A) Resume from Fact-Checker
(B) Restart pipeline from beginning
(C) Abort (keep completed outputs)

工作流状态会在每个环节完成后保存：

存储位置:

/tmp/pipeline-state-{workflow_id}.yaml

恢复方式:

检测到中断的工作流：research-abc123
主题: "hepatocyte oxygenation"
进度: 已完成3/5环节

已完成环节:
- [x] Researcher
- [x] Synthesizer
- [x] Devil's Advocate
- [ ] Fact-Checker
- [ ] Editor

可选操作:
(A) 从Fact-Checker环节恢复
(B) 从开头重新启动工作流
(C) 终止工作流（保留已完成输出）

Outputs

输出内容

Primary Output

主要输出

Final polished document:

docs/literature/{topic}/review-{topic}.md

最终精炼文档:

docs/literature/{topic}/review-{topic}.md

Intermediate Outputs (preserved for reference)

中间输出（保留用于参考）

Researcher draft:

docs/literature/{topic}/researcher-draft.md

Paper notes:
```
docs/literature/{topic}/notes/
```

Synthesis document:

docs/literature/{topic}/synthesis-{topic}.md

Devil's advocate review:

docs/literature/{topic}/review-report.md

Fact-check report:

docs/literature/{topic}/fact-check-report.md

Researcher草稿:

docs/literature/{topic}/researcher-draft.md

论文笔记:
```
docs/literature/{topic}/notes/
```

合成文档:

docs/literature/{topic}/synthesis-{topic}.md

Devil's Advocate评审报告:

docs/literature/{topic}/review-report.md

事实核查报告:

docs/literature/{topic}/fact-check-report.md

Pipeline Artifacts

工作流产物

Handoff documents:
```
/tmp/handoff-{workflow_id}-*.yaml
```
State file:
```
/tmp/pipeline-state-{workflow_id}.yaml
```
Completion report: Displayed to user, optionally saved

交接文档:
```
/tmp/handoff-{workflow_id}-*.yaml
```
状态文件:
```
/tmp/pipeline-state-{workflow_id}.yaml
```
完成报告: 展示给用户，可选择保存

Integration with Other Skills

与其他技能的集成

When to Escalate to technical-pm

何时升级到Technical-PM

Use technical-pm instead of research-pipeline when:

Multiple independent research streams needed (parallel execution)
Custom skill ordering required
Non-research skills needed in the workflow
Complex dependency management required

当出现以下情况时，使用technical-pm替代research-pipeline：

需要多个独立的并行研究分支
需要自定义技能执行顺序
工作流中需要非研究类技能
需要复杂的依赖管理

Handoff to archive-workflow

与Archive-Workflow的交接

After pipeline completion, user may want to invoke archive-workflow separately for project organization:

Skill(archive-workflow, project="{project root}")

Pipeline does NOT automatically invoke archive-workflow to give user control over organization decisions.

工作流完成后，用户可单独调用archive-workflow进行项目整理：

Skill(archive-workflow, project="{项目根目录}")

工作流不会自动调用archive-workflow，以便用户自主决定整理方式。

Handoff to git-strategy-advisor

与Git-Strategy-Advisor的交接

After pipeline completion, the pipeline MAY invoke git-strategy-advisor for git workflow recommendations. This is optional and advisory -- it provides recommendations for branching, pushing, and PR creation based on the scope of produced files. Invocation is via Task tool, not Skill tool.

工作流完成后，可调用git-strategy-advisor获取Git工作流建议。此步骤为可选建议，会根据产出文件的范围推荐分支管理、推送时机及PR创建策略。调用需通过Task工具，而非Skill工具。

Example Invocations

调用示例

Basic Research Review

基础文献综述

User: "Research hepatocyte oxygenation and write a comprehensive literature review"

Pipeline executes:

Researcher: Reviews 8-10 papers on hepatocyte oxygen consumption
Synthesizer: Identifies themes (measurement methods, culture conditions, species variations)
Devil's Advocate: Challenges assumption that in vitro values apply to bioreactor design
Fact-Checker: Verifies all K_oA values trace to primary sources
Editor: Polishes into CLAUDE.md-compliant document

Output:

docs/literature/hepatocyte-oxygenation/review-hepatocyte-oxygenation.md

用户: "研究肝细胞氧合作用并撰写一份全面的文献综述"

工作流执行:

Researcher：评审8-10篇关于肝细胞氧消耗的论文
Synthesizer：识别核心主题（测量方法、培养条件、物种差异）
Devil's Advocate：质疑「体外数据适用于生物反应器设计」这一假设
Fact-Checker：验证所有K_oA值均来自原始文献
Editor：润色为符合CLAUDE.md规范的文档

输出:

docs/literature/hepatocyte-oxygenation/review-hepatocyte-oxygenation.md

Focused Summary

聚焦型摘要

User: "Give me a quick summary of hollow fiber bioreactor designs for liver support"

Pipeline executes (focused mode):

Researcher: Reviews 3-4 key papers
Synthesizer: Single-theme summary
Devil's Advocate: 1 exchange round
Fact-Checker: Critical citations only
Editor: Essential polish

Output:

docs/literature/hollow-fiber-bioreactor/review-hollow-fiber-bioreactor.md

用户: "给我一份关于用于肝脏支持的中空纤维生物反应器设计的快速摘要"

工作流执行（聚焦模式）:

Researcher：评审3-4篇关键论文
Synthesizer：单主题摘要
Devil's Advocate：1轮交互
Fact-Checker：仅验证关键引用
Editor：仅做必要润色

输出:

docs/literature/hollow-fiber-bioreactor/review-hollow-fiber-bioreactor.md

With Constraints

带约束条件的研究

User: "Research Matrigel alternatives for hepatocyte culture, focus on chemical approaches, max 5 pages"

Pipeline parses:

Topic: Matrigel alternatives for hepatocyte culture
Focus: Chemical approaches (not biological)
Constraint: 5 pages max

Adjusts behavior:

Researcher: Filters for chemical/synthetic matrix papers
Synthesizer: Respects page limit in synthesis
All stages: Honor focus constraint

用户: "研究用于肝细胞培养的Matrigel替代物，聚焦化学方法，最多5页"

工作流解析:

主题：用于肝细胞培养的Matrigel替代物
聚焦：化学方法（非生物方法）
约束：最多5页

调整后的行为:

Researcher：筛选化学/合成基质相关论文
Synthesizer：在合成时遵守页数限制
所有环节：严格遵循聚焦约束

Common Pitfalls

常见陷阱

Scope creep in researcher stage
- Symptom: Researcher spends 6+ hours on "quick summary" request
- Fix: Set explicit scope at pipeline init (comprehensive vs focused)
Thesis drift between stages
- Symptom: Final document doesn't answer original question
- Fix: Handoff includes original_goal; each stage checks alignment
Citation format inconsistency
- Symptom: Mixed citation formats (superscripts and brackets)
- Fix: Fact-checker enforces format; editor normalizes
Skipping stages reduces quality
- Symptom: Unchallenged arguments, unverified citations in final output
- Fix: Warn user when stages are skipped; note in completion report

Researcher环节范围蔓延
- 症状: 针对「快速摘要」请求，Researcher耗时6小时以上
- 解决方法: 在工作流初始化时明确设置范围（全面 vs 聚焦）
环节间论点偏离
- 症状: 最终文档未回答初始问题
- 解决方法: 交接内容包含初始目标，每个环节均需检查对齐度
引用格式不一致
- 症状: 混合使用多种引用格式（上标和括号）
- 解决方法: Fact-Checker强制执行格式规范，Editor统一格式
跳过环节降低质量
- 症状: 最终输出存在未被质疑的论点和未验证的引用
- 解决方法: 当用户跳过环节时发出警告，并在完成报告中记录

Handoffs

交接规则

Condition	Hand off to
Pipeline complete	User (with completion report)
Pipeline complete, organization needed	archive-workflow (manual invocation)
Stage failure, needs diagnosis	User (with error context)
Custom workflow needed	technical-pm (for flexible orchestration)
Parallel research streams	technical-pm with Task tool

条件	交接对象
工作流完成	用户（附带完成报告）
工作流完成，需要整理项目	archive-workflow（手动调用）
环节执行失败，需要诊断	用户（附带错误上下文）
需要自定义工作流	technical-pm（用于灵活编排）
需要并行研究分支	technical-pm（结合Task工具）

Supporting Resources

支持资源

See

examples/

directory for:

```
pipeline-invocation-example.md
```
- Sample pipeline run with all stages
```
completion-report-example.md
```
- Example completion report format

请查看

examples/

目录获取：

```
pipeline-invocation-example.md
```
- 包含所有环节的完整工作流运行示例
```
completion-report-example.md
```
- 完成报告格式示例