Skip to main content

​
Overview

The Executive Search Agent automates candidate sourcing for executive and senior positions. It reads job descriptions, builds targeted search strategies, collects candidate data, and delivers AI-analyzed shortlists — all using the anysite CLI. The agent works step-by-step, confirming each stage with you in plain language.
Requires the Claude Code Skill and the /anysite-cli skill installed.

​
Workflow

​
Step 1: Analyze Input Materials

When you provide a job description (PDF, DOCX, or text), the agent extracts structured search criteria: Basic Information:
  • Position and level (IC / Manager / Director / VP / C-level)
  • Company and context (size, industry, stage)
  • Geography and timezone requirements
  • Compensation range (if specified)
Technical Profile:
  • Required technologies/skills
  • Preferred technologies/skills
  • Anti-patterns (what we’re NOT looking for)
Candidate Profile:
  • Target companies for sourcing
  • Target industries
  • Tenure pattern (stable / any)
  • Minimum years of experience
Culture Fit:
  • Positive markers
  • Red flags
The agent presents extracted criteria for your confirmation before proceeding: 
Here's what I understood from the document:

Position: VP of Engineering at Acme Corp
Location: Germany / UK (remote OK)
Level: VP / Director

Technical Profile:
- Looking for: distributed systems, Kubernetes, Go/Rust
- Avoiding: consulting-only backgrounds

Target Companies: Stripe, Shopify, GitHub, Intercom
Target Industries: SaaS, DevTools, Fintech

Culture Fit:
- Positive: open-source contributions, technical blog posts
- Red flags: very short tenures (< 1 year)

Is this correct? Anything to change or add?

​
Step 2: Find URNs for Filters

Before building the search strategy, the agent finds LinkedIn identifiers needed for precise filtering. Locations: 
anysite api /api/linkedin/search/locations name="Germany" count=5 --format table
anysite api /api/linkedin/search/locations name="United Kingdom" count=5 --format table
Companies: 
anysite api /api/linkedin/company company=intercom --fields "name,urn"
anysite api /api/linkedin/search/companies keywords="Shopify" count=5 --format table
Industries: 
anysite api /api/linkedin/search/industries name="Software" count=10 --format table
The agent shows found identifiers: 
Found identifiers for filters:

Locations:
- Germany: urn:li:geo:101282230
- United Kingdom: urn:li:geo:101165590

Companies:
- Intercom: urn:li:company:2491343
- Shopify: urn:li:company:784652
- GitHub: urn:li:company:1418841

Continue with search strategy?

​
Step 3: Build Search Strategy

Based on the criteria, the agent proposes a strategy using appropriate sources: Sales Navigator Search (precise filters, paid)
  • When: need precise filters by level, tenure, function
  • Endpoint: /api/linkedin/sn_search/users
Regular LinkedIn Search (keywords, free)
  • When: keyword search is sufficient
  • Endpoint: /api/linkedin/search/users
Company Employees Search
  • When: have a list of target companies
  • Endpoint: /api/linkedin/company/employees
The agent presents its plan: 
Proposed search strategy:

Source 1: Sales Navigator search — VP/Director Engineering in Germany/UK
- Filters: level, function, geography, past companies
- Expected result: ~100 profiles

Source 2: Company employees — target companies (Stripe, Shopify, GitHub)
- Filter: engineering department
- Expected result: ~150 profiles

Processing:
- Deduplication by URN
- Full profile enrichment
- LLM analysis for criteria matching

Agree with this approach?

​
Step 4: Create Pipeline

After strategy confirmation, the agent builds a dataset.yaml dynamically based on:
  • Selected search sources
  • Extracted criteria
  • Found URNs
name: vp-engineering-search

sources:
  # Sales Navigator search
  - id: sn_search
    endpoint: /api/linkedin/sn_search/users
    params:
      keywords: "VP Engineering"
      location: ["urn:li:geo:101282230", "urn:li:geo:101165590"]
      current_company: ["urn:li:company:2491343", "urn:li:company:784652"]
      seniority: ["VP", "Director"]
      count: 100
    on_error: skip

  # Company employees search
  - id: target_employees
    endpoint: /api/linkedin/company/employees
    from_file: target_companies.txt
    input_key: companies
    input_template:
      companies: [{ type: company, value: "{value}" }]
      count: 50
    parallel: 3
    on_error: skip

  # Combine and deduplicate
  - id: all_candidates
    type: union
    sources: [sn_search, target_employees]
    dedupe_by: urn.value

  # Enrich with full profiles
  - id: profiles
    endpoint: /api/linkedin/user
    dependency:
      from_source: all_candidates
      field: urn.value
      dedupe: true
    input_key: user
    params:
      with_experience: true
      with_skills: true
      with_education: true
    parallel: 5
    on_error: skip

  # LLM analysis
  - id: analyzed
    type: llm
    dependency:
      from_source: profiles
      field: name
    llm:
      # Extract structured attributes
      - type: enrich
        add:
          - "distributed_systems_years:number"
          - "management_experience_years:number"
          - "tenure_avg_years:number"
          - "has_open_source:boolean"
          - "has_technical_blog:boolean"
        fields: [name, headline, description, experience, skills]

      # Classify fit
      - type: classify
        categories: "strong_fit,good_fit,maybe,not_fit"
        output_column: fit_score
        fields: [headline, experience, skills]

      # Detailed analysis
      - type: generate
        prompt: |
          You are an expert executive recruiter. Analyze this candidate for:

          ROLE: VP of Engineering
          COMPANY: Acme Corp (Series B SaaS, 200 employees)
          REQUIREMENTS: distributed systems, Kubernetes, Go/Rust, team leadership 50+
          CULTURE: open-source, technical depth, stable tenure

          CANDIDATE:
          Name: {name}
          Headline: {headline}
          Location: {location}
          Experience: {experience}
          Skills: {skills}

          Provide:
          ## SCORE: [1-10]
          ## FIT ASSESSMENT (2-3 sentences)
          ## STRENGTHS
          ## CONCERNS
          ## RED FLAGS
          ## RECOMMENDATION: [STRONG PROCEED / PROCEED / MAYBE / PASS]
          ## OUTREACH ANGLE (personalized hook for initial contact)
        output_column: analysis
        temperature: 0.3
        fields: [name, headline, location, experience, skills]

    export:
      - type: file
        path: ./output/candidates_{{date}}.csv
        format: csv

storage:
  format: parquet
  path: ./data/
The LLM analysis prompt is dynamically generated based on the job description. The example above shows the structure — actual content adapts to each search.

​
Step 5: Run Collection

The agent always does a dry-run first: 
# Preview execution plan
anysite dataset collect dataset.yaml --dry-run

# Full collection after confirmation
anysite dataset collect dataset.yaml
And reports results: 
Collection complete!

Results:
- Sales Navigator search: 87 profiles
- Target company employees: 142 profiles
- After deduplication: 201 unique candidates
- Full profiles enriched: 198 (3 errors skipped)
- LLM analysis complete: 198 records

Show fit_score distribution?

​
Step 6: Analysis and Reporting

The agent presents a summary: 
Candidate Distribution:

| Fit Score    | Count | %   |
|--------------|-------|-----|
| strong_fit   | 12    | 6%  |
| good_fit     | 34    | 17% |
| maybe        | 68    | 34% |
| not_fit      | 84    | 43% |

Top Candidates (strong_fit):

1. Jane Smith — VP Platform Engineering at Stripe, Berlin
   Score: 9/10 — 8 years distributed systems, Kubernetes expert, manages 60+ engineers

2. Alex Chen — Director of Engineering at GitHub, London
   Score: 8/10 — Active open-source contributor, Go/Rust background, led 40-person team

3. Maria Garcia — VP Engineering at Shopify, Remote (UK)
   Score: 8/10 — Built platform team from 10 to 50, strong K8s expertise

What's next?
- Show detailed analysis of top candidates?
- Export to CSV?
- Apply additional filters?
- Set up weekly incremental search?

​
Step 7: Incremental Updates

For repeated runs (e.g., weekly candidate refresh): 
anysite dataset collect dataset.yaml --incremental

Incremental mode:
- Previously collected: 198 candidates
- New found: 23
- New enriched and analyzed: 23
- Total now: 221

Show only new candidates?

​
Reference Endpoints

​
Finding Identifiers

# Locations
anysite api /api/linkedin/search/locations name="Germany" count=5

# Companies
anysite api /api/linkedin/company company={slug} --fields "name,urn"
anysite api /api/linkedin/search/companies keywords="..." count=10

# Industries
anysite api /api/linkedin/search/industries name="Software" count=10

# Sales Navigator (precise filters)
anysite describe /api/linkedin/sn_search/users

# Regular search (keywords)
anysite describe /api/linkedin/search/users

# Company employees
anysite describe /api/linkedin/company/employees

​
Profiles

anysite describe /api/linkedin/user

​
Key Principles

  1. Always confirm before executing — especially Sales Navigator searches (expensive) and LLM analysis (token costs)
  2. Adapt the pipeline to each specific role — search sources, filters, LLM extraction fields, and analysis prompts are all dynamic
  3. Use --dry-run before actual collection runs
  4. Use --incremental for repeated runs to avoid re-collecting existing candidates
  5. Present results in plain language at every step
  6. Show intermediate results in readable format — tables for quick scanning, detailed analysis for top candidates