When to Use
User is job hunting — searching for positions, applying to companies, or preparing for interviews. Agent handles opportunity tracking, company research, application materials, and interview prep.
Architecture
Memory lives in ~/job-search/. See memory-template.md for setup.
CODEBLOCK0
Quick Reference
| Topic | File |
|---|
| Memory setup | INLINECODE2 |
| Research patterns |
research.md |
| Interview prep |
interviews.md |
Core Rules
1. Quality Over Volume
- - 5 targeted applications beat 50 generic ones
- Each application needs company-specific customization
- Never spray-and-pray — it burns reputation
2. Verify "Remote" Claims
Before recommending remote positions, check:
- - Geographic restrictions ("US only", "EU timezone required")
- Tax/legal requirements in fine print
- Actual timezone overlap expectations
3. Detect Red Flags
| Signal | Likely Meaning |
|---|
| Posted 3+ months | Ghost job or high turnover |
| "Rockstar/ninja" language |
Overwork culture |
| Vague salary ("competitive") | Below market |
| "Young dynamic team" | Age bias risk |
| Recent mass layoffs | Instability |
4. Preserve User Voice
- - Materials must sound like the USER, not generic AI
- Ask for writing samples to match tone
- Never over-optimize with keywords at cost of authenticity
- Recruiters detect AI-written content — personalization matters
5. Track Application State
Maintain in ~/job-search/applications.md:
- - Company, role, date applied
- Current status, next action
- Contacts, interview dates
- Follow-up reminders
6. Research Before Applying
For each target company, gather:
- - Financial health (funding, revenue trends)
- Glassdoor sentiment (filter for recency)
- Recent news (layoffs, acquisitions)
- Hiring manager profile if findable
7. Adjust for User Context
| User Type | Priority |
|---|
| Senior (10+ yrs) | Network activation, discretion, salary negotiation |
| Junior/New grad |
Volume with quality, entry-level friendly companies |
| Career changer | Transferable skills narrative, bridge roles |
| Urgent need | Speed, temporary options, immediate income |
Common Traps
- - ATS optimization kills authenticity — keyword stuffing passes filters but humans reject robotic text
- Salary data goes stale fast — verify ranges are current, not 2-year-old estimates
- "Perfect match" overconfidence — 60% requirement fit still means likely rejection
- Networking advice without context — cold outreach fails without warm introduction strategy
- Long-term advice for urgent needs — "build your brand" doesn't pay rent this month
使用时机
用户正在求职——搜索职位、投递公司或准备面试。智能体负责机会追踪、公司调研、申请材料及面试准备。
架构
记忆文件存储在 ~/job-search/ 目录下。配置方法参见 memory-template.md。
~/job-search/
├── memory.md # 热点：偏好设置、目标标准
├── applications.md # 活跃申请流程
├── companies.md # 目标公司调研
├── materials/ # 简历版本、求职信
└── archive/ # 已关闭申请
快速参考
| 主题 | 文件 |
|---|
| 记忆配置 | memory-template.md |
| 调研模式 |
research.md |
| 面试准备 | interviews.md |
核心规则
1. 质量优先于数量
- - 5份精准投递胜过50份海投
- 每份申请需针对公司定制
- 切勿广撒网——这会损害声誉
2. 核实远程声明
推荐远程职位前，请核查：
- - 地域限制（仅限美国、需欧盟时区）
- 细则中的税务/法律要求
- 实际时区重叠预期
3. 识别危险信号
| 信号 | 可能含义 |
|---|
| 发布超过3个月 | 僵尸职位或高离职率 |
| 摇滚明星/忍者用语 |
过度工作文化 |
| 薪资模糊（有竞争力） | 低于市场水平 |
| 年轻活力团队 | 年龄歧视风险 |
| 近期大规模裁员 | 不稳定性 |
4. 保留用户风格
- - 材料必须像用户本人撰写，而非通用AI
- 要求提供写作样本以匹配语气
- 切勿为优化关键词而牺牲真实性
- 招聘人员能识别AI生成内容——个性化至关重要
5. 跟踪申请状态
在 ~/job-search/applications.md 中维护：
- - 公司、职位、申请日期
- 当前状态、下一步行动
- 联系人、面试日期
- 跟进提醒
6. 申请前调研
针对每个目标公司，收集：
- - 财务状况（融资、收入趋势）
- Glassdoor评价（筛选近期内容）
- 近期新闻（裁员、收购）
- 如能找到，招聘经理资料
7. 根据用户背景调整
| 用户类型 | 优先级 |
|---|
| 资深人士（10年以上） | 人脉激活、谨慎行事、薪资谈判 |
| 初级/应届毕业生 |
质量与数量并重、友好型入门公司 |
| 转行者 | 可迁移技能叙述、过渡岗位 |
| 急需就业者 | 速度优先、临时选项、即时收入 |
常见陷阱
- - ATS优化牺牲真实性——关键词堆砌能通过筛选，但人类会拒绝机械文本
- 薪资数据快速过时——需验证当前范围，而非两年前的估算
- 完美匹配过度自信——60%要求匹配度仍可能被拒
- 脱离上下文的社交建议——缺乏引荐策略的冷启动注定失败
- 对紧急需求提供长期建议——打造个人品牌无法支付本月房租
