resume-tailoring

简历定制技能

概述

生成高质量、针对特定职位描述优化的定制简历，同时确保事实准确。通过对话式探索，发掘未记录的经历，以全面展现个人。

核心原则： 在保留事实的前提下优化——最大限度地提高匹配度，同时确保事实准确。绝不捏造经历，而是巧妙地重新表述并突出相关方面。

使命： 求职者能否获得工作，应取决于他们的经历和能力，而不是简历写作技巧。

适用场景

以下情况使用此技能：

• 用户提供职位描述并希望定制简历
• 用户有多份 Markdown 格式的现有简历
• 用户希望针对特定职位或公司优化申请材料
• 用户需要帮助发掘并阐述未记录的经历

不要用于：

• 从零开始撰写通用简历（用户需要先有现成的简历库）
• 求职信（使用其他技能）
• 优化 LinkedIn 个人资料（使用其他技能）

快速开始

用户需提供：

1. 职位描述（文本或 URL）
2. 简历库位置（默认为当前目录下的 resumes/）

工作流：

1. 根据现有简历建立简历库
2. 调研公司和职位
3. 创建模板（需用户确认）
4. 可选：通过分支提问探索经历
5. 对内容进行匹配并评估可信度
6. 生成 MD、DOCX、PDF 和报告
7. 用户审核 → 可选更新简历库

实现

参见支持文件：

• research-prompts.md - 用于公司和职位调研的结构化提示词
• matching-strategies.md - 内容匹配算法与评分
• branching-questions.md - 经历探索对话模式

工作流详情

多职位检测

用户提供以下内容时触发：

• 多个职位描述 URL（以逗号或换行分隔）
• 短语：“多个工作”“多个职位”“批量”“3 个工作”
• 公司或职位列表：“Microsoft PM、Google TPM、AWS PM”

检测逻辑：

# Pseudo-code
def detect_multi_job(user_input):
    indicators = [
        len(extract_urls(user_input)) > 1,
        any(phrase in user_input.lower() for phrase in
            ["multiple jobs", "several positions", "batch of", "3 jobs", "5 jobs"]),
        count_company_mentions(user_input) > 1
    ]
    return any(indicators)

如果检测到：

"I see you have multiple job applications. Would you like to use
multi-job mode?

BENEFITS:
- Shared experience discovery (faster - ask questions once for all jobs)
- Batch processing with progress tracking
- Incremental additions (add more jobs later)

TIME COMPARISON (3 similar jobs):
- Sequential single-job: ~45 minutes (15 min × 3)
- Multi-job mode: ~40 minutes (15 min discovery + 8 min per job)

Use multi-job mode? (Y/N)"

如果用户确认 Y：

• 使用多职位工作流（参见 multi-job-workflow.md）

如果用户确认 N 或检测到单个职位：

• 使用现有的单职位工作流（从阶段 0 开始）

向后兼容性： 单职位工作流完全不变。

多职位工作流：

启用多职位模式后，参见 multi-job-workflow.md 了解完整工作流。

多职位流程概览：

┌─────────────────────────────────────────────────────────────┐
│ PHASE 0: Intake & Batch Initialization                      │
│ - Collect 3-5 job descriptions                              │
│ - Initialize batch structure                                │
│ - Run library initialization (once)                         │
└─────────────────────────────────────────────────────────────┘
                           ↓
┌─────────────────────────────────────────────────────────────┐
│ PHASE 1: Aggregate Gap Analysis                            │
│ - Extract requirements from all JDs                         │
│ - Cross-reference against library                           │
│ - Build unified gap map (deduplicate)                       │
│ - Prioritize: Critical → Important → Job-specific          │
└─────────────────────────────────────────────────────────────┘
                           ↓
┌─────────────────────────────────────────────────────────────┐
│ PHASE 2: Shared Experience Discovery                       │
│ - Single branching interview covering ALL gaps              │
│ - Multi-job context for each question                       │
│ - Tag experiences with job relevance                        │
│ - Enrich library with discoveries                           │
└─────────────────────────────────────────────────────────────┘
                           ↓
┌─────────────────────────────────────────────────────────────┐
│ PHASE 3: Per-Job Processing (Sequential)                   │
│ For each job:                                               │
│   ├─ Research (company + role benchmarking)                 │
│   ├─ Template generation                                    │
│   ├─ Content matching (uses enriched library)              │
│   └─ Generation (MD + DOCX + Report)                        │
│ Interactive or Express mode                                 │
└─────────────────────────────────────────────────────────────┘
                           ↓
┌─────────────────────────────────────────────────────────────┐
│ PHASE 4: Batch Finalization                                │
│ - Generate batch summary                                    │
│ - User reviews all resumes together                         │
│ - Approve/revise individual or batch                        │
│ - Update library with approved resumes                      │
└─────────────────────────────────────────────────────────────┘

节省时间：

• 3 个职位：约 40 分钟（而顺序处理需 45 分钟）＝节省 11%
• 5 个职位：约 55 分钟（而顺序处理需 75 分钟）＝节省 27%

质量： 与单职位工作流深度相同（调研、匹配、生成）

参见 multi-job-workflow.md 了解完整实现细节。

阶段 0：初始化简历库

始终首先运行——建立最新的简历数据库

流程：

1. 定位简历目录：

   User provides path OR default to ./resumes/
   Validate directory exists
   

2. 扫描 Markdown 文件：

   Use Glob tool: pattern="*.md" path={resume_directory}
   Count files found
   Announce: "Building resume library... found {N} resumes"
   

3. 解析每份简历： 对于每个简历文件：

   • 使用 Read 工具加载内容
   • 提取各部分：职位、要点、技能、教育经历
   • 识别模式：要点结构、长度、格式

4. 建立经历数据库结构：

   {
     "roles": [
       {
         "role_id": "company_title_year",
         "company": "Company Name",
         "title": "Job Title",
         "dates": "YYYY-YYYY",
         "description": "Role summary",
         "bullets": [
           {
             "text": "Full bullet text",
             "themes": ["leadership", "technical"],
             "metrics": ["17x improvement", "$3M revenue"],
             "keywords": ["cross-functional", "program"],
             "source_resumes": ["resume1.md"]
           }
         ]
       }
     ],
     "skills": {
       "technical": ["Python", "Kusto", "AI/ML"],
       "product": ["Roadmap", "Strategy"],
       "leadership": ["Stakeholder mgmt"]
     },
     "education": [...],
     "user_preferences": {
       "typical_length": "1-page|2-page",
       "section_order": ["summary", "experience", "education"],
       "bullet_style": "pattern"
     }
   }
   

5. 自动标记内容：

   • 主题：扫描关键词（领导力、技术、分析等）
   • 指标：提取数字、百分比和金额
   • 关键词：高频技术术语、行动动词

输出： 可供匹配的内存数据库

代码模式：

# Pseudo-code for reference
library = {
    "roles": [],
    "skills": {},
    "education": []
}

for resume_file in glob("resumes/*.md"):
    content = read(resume_file)
    roles = extract_roles(content)
    for role in roles:
        role["bullets"] = tag_bullets(role["bullets"])
        library["roles"].append(role)

return library

阶段 1：调研阶段

目标： 构建全面的“成功画像”，不局限于职位描述

输入：

• 职位描述（用户提供的文本或 URL）
• 可选：职位描述中未提及的公司名称

流程：

1.1 解析职位描述：

Use research-prompts.md JD parsing template
Extract: requirements, keywords, implicit preferences, red flags, role archetype

1.2 公司调研：

WebSearch queries:
- "{company} mission values culture"
- "{company} engineering blog"
- "{company} recent news"

Synthesize: mission, values, business model, stage

1.3 职位基准分析：

WebSearch: "site:linkedin.com {job_title} {company}"
WebFetch: Top 3-5 profiles
Analyze: common backgrounds, skills, terminology

If sparse results, try similar companies

1.4 综合形成成功画像：

Combine all research into structured profile (see research-prompts.md template)

Include:
- Core requirements (must-have)
- Valued capabilities (nice-to-have)
- Cultural fit signals
- Narrative themes
- Terminology map (user's background → their language)
- Risk factors + mitigations

确认点：

Present success profile to user:

"Based on my research, here's what makes candidates successful for this role:

{SUCCESS_PROFILE_SUMMARY}

Key findings:
- {Finding 1}
- {Finding 2}
- {Finding 3}

Does this match your understanding? Any adjustments?"

Wait for user confirmation before proceeding.

输出： 经确认的成功画像文档

第 2 阶段：模板生成

目标： 创建针对这一特定职位优化的简历结构

输入：

• 成功画像（来自第 1 阶段）
• 用户的简历库（来自第 0 阶段）

流程：

2.1 分析用户的简历库：

Extract from library:
- All roles, titles, companies, date ranges
- Role archetypes (technical contributor, manager, researcher, specialist)
- Experience clusters (what domains/skills appear frequently)
- Career progression and narrative

2.2 职位整合决策：

适合整合的情况：

• 同一家公司，职责相似
• 目标职位更看重连续性，而非细分的职业发展历程
• 合并后的叙述比拆分开更有力
• 页面空间有限

适合分开的情况：

• 不同公司（始终分开）
• 职责截然不同，且两者都很重要
• 目标职位看重具体的职业发展历程
• 其中一个职位的相关经验显著更多

决策模板：

For {Company} with {N} positions:

OPTION A (Consolidated):
Title: "{Combined_Title}"
Dates: "{First_Start} - {Last_End}"
Rationale: {Why consolidation makes sense}

OPTION B (Separate):
Position 1: "{Title}" ({Dates})
Position 2: "{Title}" ({Dates})
Rationale: {Why separate makes sense}

RECOMMENDED: Option {A/B} because {reasoning}

2.3 职称重新表述原则：

核心规则： 如实反映实际工作，突出与目标职位最相关的方面

策略：

1. 突出不同方面：

   • “研究生研究员” → “研究软件工程师”（如果工作以编码为主）
   • “数据科学负责人” → “技术项目经理”（如果主要负责领导工作）

2. 使用行业标准术语：

   • “Scientist III” → “高级研究科学家”（更清楚地体现资历）
   • “项目协调员” → “项目经理”（标准术语）

3. 如实注明专业方向：

   • “工程师” → “ML 工程师”（如果 ML 工作占比较大）
   • “研究员” → “计算生态学家”（如果使用了计算方法）

4. 调整资历级别标识：

   • 根据职责范围选择“负责人”“高级”或“资深”

限制：

• 绝不声称做过未实际从事的工作
• 绝不夸大到无法合理证明的资历级别
• 公司名称和日期必须准确无误
• 核心职责必须准确

2.4 生成模板结构：

## Professional Summary
[GUIDANCE: {X} sentences emphasizing {themes from success profile}]
[REQUIRED ELEMENTS: {keywords from JD}]

## Key Skills
[STRUCTURE: {2-4 categories based on JD structure}]
[SOURCE: Extract from library matching success profile]

## Professional Experience

### [ROLE 1 - Most Recent/Relevant]
[CONSOLIDATION: {merge X positions OR keep separate}]
[TITLE OPTIONS:
  A: {emphasize aspect 1}
  B: {emphasize aspect 2}
  Recommended: {option with rationale}]
[BULLET ALLOCATION: {N bullets based on relevance + recency}]
[GUIDANCE: Emphasize {themes}, look for {experience types}]

Bullet 1: [SEEKING: {requirement type}]
Bullet 2: [SEEKING: {requirement type}]
...

### [ROLE 2]
...

## Education
[PLACEMENT: {top if required/recent, bottom if experience-heavy}]

## [Optional Sections]
[INCLUDE IF: {criteria from success profile}]

检查点：

Present template to user:

"Here's the optimized resume structure for this role:

STRUCTURE:
{Section order and rationale}

ROLE CONSOLIDATION:
{Decisions with options}

TITLE REFRAMING:
{Proposed titles with alternatives}

BULLET ALLOCATION:
Role 1: {N} bullets (most relevant)
Role 2: {N} bullets
...

Does this structure work? Any adjustments to:
- Role consolidation?
- Title reframing?
- Bullet allocation?"

Wait for user approval before proceeding.

输出： 经批准的模板框架，以及每个部分的填写指导

第 2.5 阶段：经验挖掘（可选）

目标： 通过对话挖掘未记录的经验

触发时机：

After template approval, if gaps identified:

"I've identified {N} gaps or areas where we have weak matches:
- {Gap 1}: {Current confidence}
- {Gap 2}: {Current confidence}
...

Would you like to do a structured brainstorming session to surface
any experiences you haven't documented yet?

This typically takes 10-15 minutes and often uncovers valuable content."

User can accept or skip.

分支式访谈流程：

方式： 通过对话进行，并根据回答提出后续问题

针对每个差距进行分支式对话（见 branching-questions.md）：

1. 从开放式问题开始：

   • 技术差距：“你使用过 {skill} 吗？”
   • 软技能差距：“请讲讲你展示过 {demonstrated_skill} 的经历。”
   • 近期工作：“你最近做了哪些工作？”

2. 根据回答选择分支：

   • 是／经验扎实 → 深入了解（规模、挑战、指标）
   • 间接相关 → 了解职责及经验的可迁移性
   • 有相邻经验 → 了解相关经历
   • 个人项目 → 评估其近期程度和实质内容
   • 没有 → 尝试更宽泛的类别，或继续下一项

3. 系统性地追问：

   • 通过询问“做了什么”“如何做”“为什么”来获取细节
   • 量化：“有具体指标吗？”
   • 补充背景：“这是生产环境中的工作吗？”
   • 验证：“这能弥补这个差距吗？”

4. 立即记录：

   • 记录用户分享的经验
   • 提出澄清问题（日期、范围、影响）
   • 帮助将其表述为简历要点
   • 标注其对应的差距

记录结构：

## Newly Discovered Experiences

### Experience 1: {Brief description}
- Context: {Where/when}
- Scope: {Scale, duration, impact}
- Addresses: {Which gaps}
- Bullet draft: "{Achievement-focused bullet}"
- Confidence: {How well fills gap - percentage}

### Experience 2: ...

整合选项：

发现环节结束后：

"Great! I captured {N} new experiences. For each one:

1. ADD TO CURRENT RESUME - Integrate now
2. ADD TO LIBRARY ONLY - Save for future, not needed here
3. REFINE FURTHER - Think more about articulation
4. DISCARD - Not relevant enough

Let me know for each experience."

重要说明：

• 严守真实性标准——帮助用户清晰表述，绝不捏造
• 重点关注差距和匹配度较低的方面，而非优势领域
• 如有需要，控制时长（通常为 10–15 分钟）
• 如果用户对现有简历库有信心，可以完全跳过
• 注意适时结束，不要让用户疲于应答

输出： 新经验已整合到简历库中，可用于匹配

阶段 3：组装阶段

目标： 使用最匹配的内容填充已批准的模板，并提供透明的评分

输入：

• 已批准的模板（来自阶段 2）
• 简历库 + 已发现的经历（来自阶段 0 + 2.5）
• 成功画像（来自阶段 1）

流程：

3.1 对每个模板槽位：

1. 从库中提取所有候选要点

   • 库数据库中的所有要点
   • 所有新发现的经历
   • 为每项内容注明来源简历

2. 为每项候选内容评分（参见 matching-strategies.md）

   • 直接匹配（40%）：关键词、领域、技术、成果
   • 可迁移（30%）：能力相同，情境不同
   • 相邻匹配（20%）：相关工具、方法、问题领域
   • 影响力（10%）：成就类型的匹配程度

   总分 =（直接匹配 × 0.4）+（可迁移 × 0.3）+（相邻匹配 × 0.2）+（影响力 × 0.1）

3. 按分数排列候选内容

   • 按分数从高到低排序
   • 按置信度分组：
     • 90-100%：直接匹配
     • 75-89%：可迁移
     • 60-74%：相邻匹配
     • <60%：较弱/缺口

4. 展示排名前 3 的匹配项并进行分析：

   TEMPLATE SLOT: {Role} - Bullet {N}
   SEEKING: {Requirement description}
   
   MATCHES:
   [DIRECT - 95%] "{bullet_text}"
     ✓ Direct: {what matches directly}
     ✓ Transferable: {what transfers}
     ✓ Metrics: {quantified impact}
     Source: {resume_name}
   
   [TRANSFERABLE - 78%] "{bullet_text}"
     ✓ Transferable: {what transfers}
     ✓ Adjacent: {what's adjacent}
     ⚠ Gap: {what's missing}
     Source: {resume_name}
   
   [ADJACENT - 62%] "{bullet_text}"
     ✓ Adjacent: {what's related}
     ⚠ Gap: {what's missing}
     Source: {resume_name}
   
   RECOMMENDATION: Use DIRECT match (95%)
   ALTERNATIVE: If avoiding repetition, use TRANSFERABLE (78%) with reframing
   

5. 处理缺口（置信度 <60%）：

   GAP IDENTIFIED: {Requirement}
   
   BEST AVAILABLE: {score}% - "{bullet_text}"
   
   REFRAME OPPORTUNITY: {If applicable}
   Original: "{text}"
   Reframed: "{adjusted_text}" (truthful because {reason})
   New confidence: {score}%
   
   OPTIONS:
   1. Use reframed version ({new_score}%)
   2. Acknowledge gap in cover letter
   3. Omit bullet slot (reduce allocation)
   4. Use best available with disclosure
   
   RECOMMENDATION: {Most appropriate option}
   

3.2 内容重构：

匹配良好（>60%）但术语不一致时：

应用 matching-strategies.md 中的策略：

• 关键词对齐（保留原意，调整术语）
• 调整侧重点（事实不变，突出不同重点）
• 抽象层级（调整技术细节程度）
• 突出规模（强调相关方面）

展示调整前后内容，确保透明：

REFRAMING APPLIED:
Bullet: {template_slot}

Original: "{original_bullet}"
Source: {resume_name}

Reframed: "{reframed_bullet}"
Changes: {what changed and why}
Truthfulness: {why this is accurate}

检查点：

"I've matched content to your template. Here's the complete mapping:

COVERAGE SUMMARY:
- Direct matches: {N} bullets ({percentage}%)
- Transferable: {N} bullets ({percentage}%)
- Adjacent: {N} bullets ({percentage}%)
- Gaps: {N} ({percentage}%)

REFRAMINGS APPLIED: {N}
- {Example 1}
- {Example 2}

GAPS IDENTIFIED:
- {Gap 1}: {Recommendation}
- {Gap 2}: {Recommendation}

OVERALL JD COVERAGE: {percentage}%

Review the detailed mapping below. Any adjustments to:
- Match selections?
- Reframings?
- Gap handling?"

[Present full detailed mapping]

Wait for user approval before generation.

输出： 逐条要点的完整映射，包含置信度评分和重构内容

阶段 4：生成阶段

目标： 创建专业的多格式输出

输入：

• 已批准的内容映射（来自阶段 3）
• 用户的格式偏好（来自库分析）
• 目标职位信息（来自阶段 1）

流程：

4.1 Markdown 生成：

将映射后的内容整理为简洁的 Markdown：

# {User_Name}

{Contact_Info}

---

## Professional Summary

{Summary_from_template}

---

## Key Skills

**{Category_1}:**
- {Skills_from_library_matching_profile}

**{Category_2}:**
- {Skills_from_library_matching_profile}

{Repeat for all categories}

---

## Professional Experience

### {Job_Title}
**{Company} | {Location} | {Dates}**

{Role_summary_if_applicable}

• {Bullet_1_from_mapping}
• {Bullet_2_from_mapping}
...

### {Next_Role}
...

---

## Education

**{Degree}** | {Institution} ({Year})
**{Degree}** | {Institution} ({Year})

采用用户的偏好：

• 库分析得出的格式风格
• 要点结构模式
• 章节顺序
• 常见篇幅（1 页或 2 页）

输出： {Name}_{Company}_{Role}_Resume.md

4.2 DOCX 生成：

使用 document-skills:docx：

REQUIRED SUB-SKILL: Use document-skills:docx

Create Word document with:
- Professional fonts (Calibri 11pt body, 12pt headers)
- Proper spacing (single within sections, space between)
- Clean bullet formatting (proper numbering config, NOT unicode)
- Header with contact information
- Appropriate margins (0.5-1 inch)
- Bold/italic emphasis (company names, titles, dates)
- Page breaks if 2-page resume

See docx skill documentation for:
- Paragraph and TextRun structure
- Numbering configuration for bullets
- Heading levels and styles
- Spacing and margins

输出： {Name}_{Company}_{Role}_Resume.docx

4.3 PDF 生成（可选）：

如果用户要求 PDF：

OPTIONAL SUB-SKILL: Use document-skills:pdf

Convert DOCX to PDF OR generate directly
Ensure formatting preservation
Professional appearance for direct submission

输出： {Name}_{Company}_{Role}_Resume.pdf

4.4 生成摘要报告：

创建元数据文件：

# Resume Generation Report
**{Role} at {Company}**

**Date Generated:** {timestamp}

## Target Role Summary
- Company: {Company}
- Position: {Role}
- IC Level: {If known}
- Focus Areas: {Key areas}

## Success Profile Summary
- Key Requirements: {top 5}
- Cultural Fit Signals: {themes}
- Risk Factors Addressed: {mitigations}

## Content Mapping Summary
- Total bullets: {N}
- Direct matches: {N} ({percentage}%)
- Transferable: {N} ({percentage}%)
- Adjacent: {N} ({percentage}%)
- Gaps identified: {list}

## Reframing Applied
- {bullet}: {original} → {reframed} [Reason: {why}]
...

## Source Resumes Used
- {resume1}: {N} bullets
- {resume2}: {N} bullets
...

## Gaps Addressed

### Before Experience Discovery:
{Gap analysis showing initial state}

### After Experience Discovery:
{Gap analysis showing final state}

### Remaining Gaps:
{Any unresolved gaps with recommendations}

## Key Differentiators for This Role
{What makes user uniquely qualified}

## Recommendations for Interview Prep
- Stories to prepare
- Questions to expect
- Gaps to address

输出： {Name}_{Company}_{Role}_Resume_Report.md

向用户展示：

"Your tailored resume has been generated!

FILES CREATED:
- {Name}_{Company}_{Role}_Resume.md
- {Name}_{Company}_{Role}_Resume.docx
- {Name}_{Company}_{Role}_Resume_Report.md
{- {Name}_{Company}_{Role}_Resume.pdf (if requested)}

QUALITY METRICS:
- JD Coverage: {percentage}%
- Direct Matches: {percentage}%
- Newly Discovered: {N} experiences

Review the files and let me know:
1. Save to library (recommended)
2. Need revisions
3. Save but don't add to library"

阶段 5：更新库（有条件）

目标： 可选择将成功生成的简历添加到库中，以便将来使用

时机： 用户审阅并批准生成的简历后

检查点问题：

"Are you satisfied with this resume?

OPTIONS:
1. YES - Save to library
   → Adds resume to permanent location
   → Rebuilds library database
   → Makes new content available for future resumes

2. NO - Need revisions
   → What would you like to adjust?
   → Make changes and re-present

3. SAVE BUT DON'T ADD TO LIBRARY
   → Keep files in current location
   → Don't enrich database
   → Useful for experimental resumes

Which option?"

如果选择选项 1（是——保存到库中）：

流程：

1. 将简历移至库中：

   Source: {current_directory}/{Name}_{Company}_{Role}_Resume.md
   Destination: {resume_library}/{Name}_{Company}_{Role}_Resume.md
   
   Also move:
   - .docx file
   - .pdf file (if exists)
   - _Report.md file
   

2. 重建库数据库：

   Re-run Phase 0 library initialization
   Parse newly created resume
   Add bullets to experience database
   Update keyword/theme indices
   Tag with metadata:
     - target_company: {Company}
     - target_role: {Role}
     - generated_date: {timestamp}
     - jd_coverage: {percentage}
     - success_profile: {reference to profile}
   

3. 保留生成元数据：

   {
     "resume_id": "{Name}_{Company}_{Role}",
     "generated": "{timestamp}",
     "source_resumes": ["{resume1}", "{resume2}"],
     "reframings": [
       {
         "original": "{text}",
         "reframed": "{text}",
         "reason": "{why}"
       }
     ],
     "match_scores": {
       "bullet_1": 95,
       "bullet_2": 87,
       ...
     },
     "newly_discovered": [
       {
         "experience": "{description}",
         "bullet": "{text}",
         "addresses_gap": "{gap}"
       }
     ]
   }
   

4. 宣布完成：

   "Resume saved to library!
   
   Library updated:
   - Total resumes: {N}
   - New content variations: {N}
   - Newly discovered experiences added: {N}
   
   This resume and its new content are now available for future tailoring sessions."
   

如果选择选项 2（否——需要修改）：

"What would you like to adjust?"

[Collect user feedback]
[Make requested changes]
[Re-run relevant phases]
[Re-present for approval]

[Repeat until satisfied or user cancels]

如果选择选项 3（保存但不添加到库中）：

"Resume files saved to current directory:
- {Name}_{Company}_{Role}_Resume.md
- {Name}_{Company}_{Role}_Resume.docx
- {Name}_{Company}_{Role}_Resume_Report.md

Not added to library - you can manually move later if desired."

更新库的好处：

• 每份成功生成的简历都会扩充简历库
• 可供使用的要点变体增加
• 可重复使用有效的重构方式
• 永久记录已发现的经历
• 后续会话可以使用更丰富的简历库
• 系统会随着时间不断自我改进

输出： 已更新的库数据库 + 元数据保留（如果选择选项 1）

错误处理与边缘情况

边缘情况 1：简历库不足

SCENARIO: User has only 1-2 resumes, limited content

HANDLING:
"⚠️ Limited resume library detected ({N} resumes).

This may result in:
- Fewer matching options
- More gaps in coverage
- Less variety in bullet phrasing

RECOMMENDATIONS:
- Proceed with available content (I'll do my best!)
- Consider adding more resumes after this generation
- Experience Discovery phase will be especially valuable

Continue? (Y/N)"

边缘情况 2：没有合适的匹配项（关键要求的置信度 <60%）

SCENARIO: Template slot requires experience user doesn't have

HANDLING:
"❌ GAP: {Requirement}

No matches found with confidence >60%

OPTIONS:
1. Run Experience Discovery - might surface undocumented work
2. Reframe best available ({score}%) - I'll show you the reframing
3. Omit bullet slot - reduce template allocation
4. Note for cover letter - emphasize learning ability

Which approach?"

[Don't force matches - be transparent about gaps]

边缘情况 3：研究阶段失败

SCENARIO: WebSearch fails, LinkedIn unavailable, company info sparse

HANDLING:
"⚠️ Limited company research available.

What I found:
- {Available info}

What's missing:
- {Missing areas}

Falling back to job description-only analysis.
Result: Template will be optimized for JD but may miss cultural nuances.

Do you have additional context about:
- Company culture?
- Team structure?
- Technologies used?

(Optional - you can share or skip)"

[Proceed with best-effort approach]

边缘情况 4：职位描述质量问题

SCENARIO: Vague JD, missing requirements, poorly written

HANDLING:
"⚠️ Job description lacks detail in areas:
- {Missing area 1}
- {Missing area 2}

This may limit template optimization.

Do you have additional context about:
- {Question 1}
- {Question 2}

(Optional - I'll work with what's available)"

[Extract what's possible, proceed]

边缘情况 5：职位合并存在歧义

SCENARIO: Unclear whether to merge roles or keep separate

HANDLING:
"🤔 Ambiguous consolidation decision for {Company}

OPTION A (Consolidated): {Rationale}
OPTION B (Separate): {Rationale}

Both are defensible. Which do you prefer?

(This becomes your preference for similar situations)"

[Remember preference for future]

边缘情况 6：简历长度限制

SCENARIO: Too much good content, exceeds 2 pages

HANDLING:
"⚠️ Content exceeds 2 pages (current: {N} bullets, ~{pages} pages)

PRUNING SUGGESTIONS (ranked by relevance):
Remove:
- {Bullet X}: {score}% match (lowest)
- {Bullet Y}: {score}% match
...

Keep all bullets and accept >2 pages?
OR
Remove {N} bullets to fit 2 pages?

Your preference?"

[User decides priority]

错误恢复：

• 所有检查点都允许返回上一阶段
• 用户可以在任何检查点请求调整
• 生成失败（DOCX/PDF）时，回退为仅提供 Markdown
• 各阶段之间会保存进度（中断后可以继续）

优雅降级：

• 研究受限 → 回退为仅分析职位描述
• 库内容较少 → 使用现有内容，并着重进行信息发现
• 匹配较弱 → 透明地指出差距
• 生成失败 → 提供 Markdown 和错误详情

使用示例

示例 1：内部职位（同一家公司）

USER: "I want to apply for Principal PM role in 1ES team at Microsoft.
      Here's the JD: {paste}"

SKILL:
1. Library Build: Finds 29 resumes
2. Research: Microsoft 1ES team, internal culture, role benchmarking
3. Template: Features PM2 Azure Eng Systems role (most relevant)
4. Discovery: Surfaces VS Code extension, Bhavana AI side project
5. Assembly: 92% JD coverage, 75% direct matches
6. Generate: MD + DOCX + Report
7. User approves → Library updated with new resume + 6 discovered experiences

RESULT: Highly competitive application leveraging internal experience

示例 2：职业转型（不同领域）

USER: "I'm a TPM trying to transition to ecology PM role. JD: {paste}"

SKILL:
1. Library Build: Finds existing TPM resumes
2. Research: Ecology sector, sustainability focus, cross-domain transfers
3. Template: Reframes "Technical Program Manager" → "Program Manager,
             Environmental Systems" emphasizing systems thinking
4. Discovery: Surfaces volunteer conservation work, graduate research in
             environmental modeling
5. Assembly: 65% JD coverage - flags gaps in domain-specific knowledge
6. Generate: Resume + gap analysis with cover letter recommendations

RESULT: Bridges technical skills with environmental domain

示例 3：处理职业空档期

USER: "I have a 2-year gap while starting a company. JD: {paste}"

SKILL:
1. Library Build: Finds pre-gap resumes
2. Research: Standard analysis
3. Template: Includes startup as legitimate role
4. Discovery: Surfaces skills developed during startup (fundraising,
             product development, team building)
5. Assembly: Frames gap as entrepreneurial experience
6. Generate: Resume presenting gap as valuable experience

RESULT: Gap becomes strength showing initiative and diverse skills

示例 4：多职位批处理（3 个相似职位）

USER: "I want to apply for these 3 TPM roles:
      1. Microsoft 1ES Principal PM
      2. Google Cloud Senior TPM
      3. AWS Container Services Senior PM
      Here are the JDs: {paste 3 JDs}"

SKILL:
1. Multi-job detection: Triggered (3 JDs detected)
2. Intake: Collects all 3 JDs, initializes batch
3. Library Build: Finds 29 resumes (once)
4. Gap Analysis: Identifies 14 gaps, 8 unique after deduplication
5. Shared Discovery: 30-minute session surfaces 5 new experiences
   - Kubernetes CI/CD for nonprofits
   - Azure migration for university lab
   - Cross-functional team leadership examples
   - Recent hackathon project
   - Open source contributions
6. Per-Job Processing (×3):
   - Job 1 (Microsoft): 85% coverage, emphasizes Azure/1ES alignment
   - Job 2 (Google): 88% coverage, emphasizes technical depth
   - Job 3 (AWS): 78% coverage, addresses AWS gap in cover letter recs
7. Batch Finalization: All 3 resumes reviewed, approved, added to library

RESULT: 3 high-quality resumes in 40 minutes vs 45 minutes sequential
        5 new experiences captured, available for future applications
        Average coverage: 84%, all critical gaps resolved

示例 5：增量添加批次

WEEK 1:
USER: "I want to apply for 3 jobs: {Microsoft, Google, AWS}"
SKILL: [Processes batch as above, completes in 40 min]

WEEK 2:
USER: "I found 2 more jobs: Stripe and Meta. Add them to my batch?"
SKILL:
1. Load existing batch (includes 5 previously discovered experiences)
2. Intake: Adds Job 4 (Stripe), Job 5 (Meta)
3. Incremental Gap Analysis: Only 3 new gaps (vs 14 original)
   - Payment systems (Stripe-specific)
   - Social networking (Meta-specific)
   - React/frontend (both)
4. Incremental Discovery: 10-minute session for new gaps only
   - Surfaces payment processing side project
   - React work from bootcamp
   - Large-scale system design course
5. Per-Job Processing (×2): Jobs 4, 5 processed independently
6. Updated Batch Summary: Now 5 jobs total, 8 experiences discovered

RESULT: 2 additional resumes in 20 minutes (vs 30 min if starting from scratch)
        Time saved by not re-asking 8 previous gaps: ~20 minutes

测试指南

手动测试清单：

测试 1：理想路径

- Provide JD with clear requirements
- Library with 10+ resumes
- Run all phases without skipping
- Verify generated files
- Check library update
PASS CRITERIA:
- All files generated correctly
- JD coverage >70%
- No errors in any phase

测试 2：最小化简历库

- Provide only 2 resumes
- Run through workflow
- Verify gap handling
PASS CRITERIA:
- Graceful warning about limited library
- Still produces reasonable output
- Gaps clearly identified

测试 3：研究失败

- Use obscure company with minimal online presence
- Verify fallback to JD-only
PASS CRITERIA:
- Warning about limited research
- Proceeds with JD analysis
- Template still reasonable

测试 4：经历挖掘价值

- Run with deliberate gaps in library
- Conduct experience discovery
- Verify new experiences integrated
PASS CRITERIA:
- Discovers genuine undocumented experiences
- Integrates into final resume
- Improves JD coverage

测试 5：职位名称重新表述

- Test various role transitions
- Verify title reframing suggestions
PASS CRITERIA:
- Multiple options provided
- Truthfulness maintained
- Rationales clear

测试 6：多格式生成

- Generate MD, DOCX, PDF, Report
- Verify formatting consistency
PASS CRITERIA:
- All formats readable
- Formatting professional
- Content identical across formats

回归测试：

After any SKILL.md changes:
1. Re-run Test 1 (happy path)
2. Verify no functionality broken
3. Commit only if passes