talent-sourcing

Compare original and translation side by side

🇺🇸

Original

English

🇨🇳

Translation

Chinese

Talent Sourcing

人才寻访

Candidate discovery powered by Nimble Web Search Agents.

User request: $ARGUMENTS

Before running any commands, read

references/nimble-playbook.md

for Claude Code constraints (no shell state, no

wait

, sub-agent permissions, communication style).

由Nimble Web Search Agents驱动的候选人发掘。

用户请求：$ARGUMENTS

运行任何命令前，请阅读

references/nimble-playbook.md

了解Claude Code的约束条件（无shell状态、禁止使用

wait

、子Agent权限、沟通风格）。

Instructions

操作说明

Step 0: Preflight

步骤0：预检

Run the preflight pattern from

references/nimble-playbook.md

(5 simultaneous Bash calls: date calc, today, CLI check, profile load, index.md load).

Also simultaneously:

mkdir -p ~/.nimble/memory/{reports,talent-sourcing}

From the results:

CLI missing or API key unset → read
```
references/profile-and-onboarding.md
```
, stop
Profile exists → note industry keywords if any; proceed to Step 1
No profile → fine, talent-sourcing doesn't require onboarding; proceed to Step 1

执行

references/nimble-playbook.md

中的预检流程（同时进行5次Bash调用：日期计算、今日日期、CLI检查、配置文件加载、index.md加载）。

同时执行：

mkdir -p ~/.nimble/memory/{reports,talent-sourcing}

根据结果处理：

CLI缺失或API密钥未设置 → 阅读
```
references/profile-and-onboarding.md
```
，停止操作
配置文件存在 → 记录行业关键词（如有）；进入步骤1
无配置文件 → 无需问题，人才寻访无需初始化；进入步骤1

Step 1: Parse Request & Confirm Search Parameters

步骤1：解析请求并确认搜索参数

Parse

$ARGUMENTS

for:

Role — job title or function (e.g. "Senior React Engineer", "Head of Sales")
Location — city, metro, region, or remote (e.g. "New York City", "remote US")
Skills / requirements — specific technologies, years of experience, domain expertise
Seniority — junior, mid, senior, staff, director, VP, C-level
Source preference — specific platforms (LinkedIn, GitHub, Indeed, etc.) or "all"

If a full job description was pasted, extract the above fields from it.

If role is missing or ambiguous, ask with

AskUserQuestion

"What role are you hiring for, and where? (e.g. 'Senior ML Engineer, remote US' or paste a job description)"

Once parameters are clear, confirm with the user using

AskUserQuestion

"Searching for: [Role] | Location: [Location] | Key skills: [Skills] | Seniority: [Seniority]

Platforms to search: LinkedIn, Indeed, GitHub (for technical roles), AngelList / Wellfound, and professional communities.

Start search

Adjust parameters first"

从

$ARGUMENTS

中解析以下信息：

岗位 — 职位名称或职能（例如“Senior React Engineer”、“Head of Sales”）
地点 — 城市、都市圈、地区或远程（例如“New York City”、“remote US”）
技能/要求 — 特定技术、工作年限、领域专业知识
资历 — 初级、中级、高级、资深、主管、副总裁、高管
来源偏好 — 指定平台（LinkedIn、GitHub、Indeed等）或“全部”

若用户粘贴了完整职位描述，从中提取上述字段。

若岗位信息缺失或模糊，使用

AskUserQuestion

询问：

"您要招聘什么岗位，地点在哪里？（例如'Senior ML Engineer, remote US'，或粘贴职位描述）"

参数明确后，使用

AskUserQuestion

与用户确认：

"搜索条件：[岗位] | 地点：[地点] | 核心技能：[技能] | 资历：[资历]

搜索平台：LinkedIn、Indeed、GitHub（技术岗位）、AngelList / Wellfound及专业社区。

开始搜索

先调整参数"

Step 2: WSA Discovery

步骤2：Web Search Agent（WSA）发现

Discover available Web Search Agents for candidate-sourcing platforms. Run simultaneously:

bash

nimble agent list --search "linkedin people" --limit 20
nimble agent list --search "indeed resume" --limit 20
nimble agent list --search "github profile" --limit 20
nimble agent list --search "wellfound talent" --limit 20

Filter results for

entity_type: SERP

entity_type: PDP

. Prefer

managed_by: "nimble"

. Validate promising agents with:

bash

nimble agent get --template-name {name}

Cache discovered WSA names and required params. If no WSAs found for a platform, fall back to

nimble search

for that platform.

寻找适用于候选人挖掘平台的Web Search Agent。同时执行：

bash

nimble agent list --search "linkedin people" --limit 20
nimble agent list --search "indeed resume" --limit 20
nimble agent list --search "github profile" --limit 20
nimble agent list --search "wellfound talent" --limit 20

筛选

entity_type: SERP

或

entity_type: PDP

的结果。优先选择

managed_by: "nimble"

的Agent。通过以下命令验证候选Agent：

bash

nimble agent get --template-name {name}

缓存已发现的WSA名称及所需参数。若某平台未找到WSA， fallback至

nimble search

进行搜索。

Step 3: Parallel Candidate Search (Sub-Agents)

步骤3：并行候选人搜索（子Agent）

Spawn

nimble-researcher

agents (

agents/nimble-researcher.md

) with

mode: "bypassPermissions"

, max 4 concurrent. Assign one agent per platform:

Agent 1 — LinkedIn

Search for people matching the role criteria. Use Boolean-style query construction:

bash

nimble search --query "site:linkedin.com/in [Role] [Location] [Key Skills]" \
  --max-results 15 --search-depth fast
nimble search --query "[Role] [Location] linkedin profile [Skill1] [Skill2]" \
  --max-results 10 --search-depth fast

If a LinkedIn WSA was discovered in Step 2, use it instead with the role title, location, and skill keywords as inputs.

Agent 2 — Indeed / Resumes

bash

nimble search --query "site:indeed.com resume [Role] [Location] [Key Skills]" \
  --max-results 10 --search-depth fast
nimble search --query "[Role] resume [Location] [Key Skills]" \
  --max-results 10 --search-depth fast

Agent 3 — GitHub (technical roles only)

Skip this agent for non-technical roles (e.g. Sales, Marketing, Operations).

bash

nimble search --query "site:github.com [Role] [Location] [Key Skills]" \
  --max-results 10 --search-depth fast
nimble search --query "github [Key Skills] developer [Location] open to work" \
  --max-results 10 --search-depth fast

Agent 4 — AngelList / Wellfound + Communities

bash

nimble search --query "site:wellfound.com [Role] [Location] [Key Skills]" \
  --max-results 10 --search-depth fast
nimble search --query "[Role] [Location] open to work OR seeking opportunities \
  [Key Skills]" --max-results 10 --search-depth fast

Each agent returns: candidate name (if available), profile URL, current title, location snippet, inferred skills, availability signals ("open to work", "seeking", "available") with event date (if available) and source URL.

启动

nimble-researcher

Agent（

agents/nimble-researcher.md

），设置

mode: "bypassPermissions"

，最多同时运行4个。为每个平台分配一个Agent：

Agent 1 — LinkedIn

搜索匹配岗位条件的人员。使用布尔式查询构造：

bash

nimble search --query "site:linkedin.com/in [Role] [Location] [Key Skills]" \\
  --max-results 15 --search-depth fast
nimble search --query "[Role] [Location] linkedin profile [Skill1] [Skill2]" \\
  --max-results 10 --search-depth fast

若步骤2中发现LinkedIn WSA，改用该Agent，输入岗位名称、地点和技能关键词。

Agent 2 — Indeed / 简历

bash

nimble search --query "site:indeed.com resume [Role] [Location] [Key Skills]" \\
  --max-results 10 --search-depth fast
nimble search --query "[Role] resume [Location] [Key Skills]" \\
  --max-results 10 --search-depth fast

Agent 3 — GitHub（仅限技术岗位）

非技术岗位（例如销售、营销、运营）跳过此Agent。

bash

nimble search --query "site:github.com [Role] [Location] [Key Skills]" \\
  --max-results 10 --search-depth fast
nimble search --query "github [Key Skills] developer [Location] open to work" \\
  --max-results 10 --search-depth fast

Agent 4 — AngelList / Wellfound + 社区

bash

nimble search --query "site:wellfound.com [Role] [Location] [Key Skills]" \\
  --max-results 10 --search-depth fast
nimble search --query "[Role] [Location] open to work OR seeking opportunities \\
  [Key Skills]" --max-results 10 --search-depth fast

每个Agent返回：候选人姓名（若可用）、资料URL、当前职位、地点片段、推断技能、求职意向信号（“open to work”、“seeking”、“available”）及事件日期（若可用）和来源URL。

Step 4: Deep Profile Extraction

步骤4：深度资料提取

For the top candidates identified in Step 3 (aim for 10–20 unique profiles across all platforms), extract full profile details. Run all extractions simultaneously:

bash

nimble extract --url "[profile-url]" --format markdown

From each extracted profile, pull:

Full name
Current role & company
Location
Skills / tech stack
Experience summary (years, notable employers)
Education
Availability signals (open to work, recent job change, posting activity)
Contact signals (email, personal site, GitHub handle)

For extraction failures, follow the fallback pattern in

references/nimble-playbook.md

. If a profile is behind a login wall and extraction fails, keep the search-snippet summary instead — do not skip the candidate.

Extraction budget: extract up to 15 profiles. If more than 15 candidates were found in Step 3, prioritize by relevance score (seniority match + skill overlap + location match) before extracting.

针对步骤3中识别出的顶尖候选人（所有平台共10–20个独特资料），提取完整详情。同时执行所有提取操作：

bash

nimble extract --url "[profile-url]" --format markdown

从每个提取的资料中获取：

全名
当前岗位及公司
地点
技能/技术栈
经验总结（工作年限、知名雇主）
教育背景
求职意向信号（开放求职、近期换工作、发布动态）
联系信号（邮箱、个人网站、GitHub账号）

提取失败时，遵循

references/nimble-playbook.md

中的 fallback流程。若资料需登录查看导致提取失败，保留搜索片段摘要——请勿跳过该候选人。

**提取预算：**最多提取15份资料。若步骤3中找到超过15位候选人，按相关性得分（资历匹配+技能重叠+地点匹配）优先提取。

Step 5: Score & Rank Candidates

步骤5：候选人评分与排名

Score each candidate (1–10) using these weighted signals:

Signal	Weight
Role / title match	30%
Skill overlap with requirements	30%
Location match	20%
Seniority match	10%
Availability signals	10%

Group candidates into tiers:

Tier 1 (Strong match, 7–10): All required signals present
Tier 2 (Partial match, 4–6): Most signals present, 1–2 gaps
Tier 3 (Stretch, 1–3): Worth reviewing if Tier 1/2 list is thin

使用以下加权信号为每位候选人评分（1–10分）：

信号	权重
岗位/职位匹配	30%
技能与要求匹配度	30%
地点匹配	20%
资历匹配	10%
求职意向信号	10%

将候选人分为三个层级：

**Tier 1（高度匹配，7–10分）：**满足所有核心信号
**Tier 2（部分匹配，4–6分）：**满足大部分信号，存在1–2个缺口
**Tier 3（潜力候选人，1–3分）：**若Tier 1/2候选人数量不足，值得考虑

Step 6: Output

步骤6：输出结果

Before presenting results, check

~/.nimble/memory/talent-sourcing/[role-slug].md

— if a candidate was surfaced in a prior run, mark them

(previously surfaced)

rather than re-presenting them as new.

Present a structured candidate report:

undefined

展示结果前，检查

~/.nimble/memory/talent-sourcing/[role-slug].md

——若某候选人在之前的搜索中出现过，标记为

(previously surfaced)

，而非作为新候选人展示。

呈现结构化候选人报告：

undefined

Candidate Report: [Role] in [Location]

候选人报告：[岗位] - [地点]

Searched: LinkedIn, Indeed, GitHub, Wellfound Found: [N] candidates | Tier 1: [N] | Tier 2: [N] | Tier 3: [N]

TL;DR: [2-3 sentence summary of the strongest candidates and any notable patterns]

搜索平台：LinkedIn、Indeed、GitHub、Wellfound 共找到：[N]位候选人 | Tier 1：[N] | Tier 2：[N] | Tier 3：[N]

摘要：[2-3句话总结顶尖候选人及显著规律]

Tier 1 — Strong Match

Tier 1 — 高度匹配

1. [Name] — [Score]/10

1. [姓名] — [得分]/10

Current role: [Title] at [Company]
Location: [Location]
Skills: [Skill1], [Skill2], [Skill3]
Experience: [X years, notable employers]
Availability: [signal] — [event date or "date unknown"] — [source URL]
Profile: [URL]
Contact signals: [email / personal site / GitHub]

...

当前岗位：[职位] @ [公司]
地点：[地点]
技能：[技能1]、[技能2]、[技能3]
经验：[X年工作经验，知名雇主]
求职意向：[信号] — [事件日期或“日期未知”] — [来源URL]
资料链接：[URL]
联系信息：[邮箱 / 个人网站 / GitHub]

...

What This Means

结果解读

[1-2 sentences on hiring outlook: supply/demand signal, speed recommendation, any standout sourcing channel]


Omit fields where data is unavailable. Do not fabricate details — use "unknown"
for missing fields. Add a one-sentence **"Why this candidate"** note for each
Tier 1 result.

[1-2句话分析招聘前景：供需信号、推荐推进速度、突出的招聘渠道]


省略无数据的字段。请勿编造信息——缺失字段使用“未知”。为每位Tier 1候选人添加一句**“推荐理由”**。

Step 7: Save to Memory

步骤7：保存至内存

Make all Write calls simultaneously:

Report →

~/.nimble/memory/reports/talent-sourcing-{YYYY-MM-DD}.md

(full candidate report with all tiers)

Per-role →

~/.nimble/memory/talent-sourcing/[role-slug].md

(candidate list; write or update)

Profile → update

last_runs.talent-sourcing

~/.nimble/business-profile.json

using the python3 snippet in

references/profile-and-onboarding.md

. Skip if the file does not exist.

Update

~/.nimble/memory/talent-sourcing/index.md

with a row for this search. Follow the wiki update pattern from

references/memory-and-distribution.md

同时执行所有写入操作：

报告 →
```
~/.nimble/memory/reports/talent-sourcing-{YYYY-MM-DD}.md
```
（包含所有层级的完整候选人报告）
按岗位分类 →
```
~/.nimble/memory/talent-sourcing/[role-slug].md
```
（候选人列表；写入或更新）
配置文件 → 使用
```
references/profile-and-onboarding.md
```
中的python3代码片段，更新
```
~/.nimble/business-profile.json
```
中的
```
last_runs.talent-sourcing
```
。若文件不存在则跳过。

更新

~/.nimble/memory/talent-sourcing/index.md

，添加本次搜索记录。遵循

references/memory-and-distribution.md

中的 wiki 更新流程。

Step 8: Share & Distribute

步骤8：分享与分发

Always offer distribution — do not skip this step. Follow

references/memory-and-distribution.md

for connector detection, sharing flow, and source links enforcement.

**必须提供分发选项——请勿跳过此步骤。**遵循

references/memory-and-distribution.md

中的连接器检测、分享流程及来源链接要求。

Step 9: Follow-ups

步骤9：后续操作

Offer next steps using

AskUserQuestion

What's next?

Go deeper on a candidate — extract full profile + find contact info

Expand search — broaden location, relax seniority, try more platforms

Narrow search — add a required skill or tighten location

Export list — save as CSV or formatted doc

Done

Sibling skill suggestions:

Run
company-deep-dive
on a candidate's current employer for deal context
Run
meeting-prep
before reaching out to a Tier 1 candidate

使用

AskUserQuestion

提供后续选项：

下一步操作？

深入了解某候选人 — 提取完整资料+查找联系方式

扩大搜索范围 — 放宽地点限制、降低资历要求、尝试更多平台

缩小搜索范围 — 添加必备技能或收紧地点限制

导出列表 — 保存为CSV或格式化文档

完成

关联技能推荐：

对候选人当前雇主运行
company-deep-dive
获取背景信息
联系Tier 1候选人前运行
meeting-prep
做准备

Error Handling

错误处理

See

references/nimble-playbook.md

for the standard error table. Skill-specific handling:

Profile behind login wall: Keep search-snippet summary; note "full profile unavailable — LinkedIn/Indeed login required" in the candidate entry.
< 5 total candidates found: Notify the user, suggest broadening location to remote or relaxing seniority, then ask whether to re-run with adjusted params.
Search 500 on a platform: Retry once with a simplified query; if still failing, skip that platform and note it in the report header.
GitHub agent skipped for non-technical role: Note "GitHub not searched for this role type" in the report header.

标准错误处理请见

references/nimble-playbook.md

中的错误表。本技能专属处理规则：

**资料需登录查看：**保留搜索片段摘要；在候选人条目中标注“完整资料不可用——需登录LinkedIn/Indeed查看”。
**找到的候选人不足5位：**通知用户，建议放宽地点至远程或降低资历要求，询问是否调整参数后重新搜索。
**某平台搜索返回500错误：**使用简化查询重试一次；若仍失败，跳过该平台并在报告头部注明。
**非技术岗位跳过GitHub Agent：**在报告头部注明“此岗位类型未搜索GitHub”。",