Career Ops

Career-Ops -- AI Job Search Pipeline

Based on santifer/career-ops. Extended with hh.ru, LinkedIn, and Russian market support.

Что понадобится

Всё личное живёт вне навыка — в ~/career-ops/. Сам навык твоих данных не хранит, поэтому его можно обновлять поверх, не боясь потерять резюме.

Commands

Working Directory

All career-ops data lives in: ~/career-ops/

~/career-ops/
  cv.md                    # Canonical CV (source of truth)
  article-digest.md        # Proof points and case studies
  portals.yml              # Portal scanner config (hh.ru + LinkedIn + global)
  config/
    profile.yml            # Candidate identity, targets, salary
  data/
    applications.md        # Application tracker
    pipeline.md            # Inbox of pending URLs
    scan-history.tsv       # Scanner dedup history
  reports/                 # Evaluation reports
  output/                  # Generated PDFs (gitignored)
  jds/                     # Saved job descriptions
  interview-prep/
    story-bank.md          # STAR+R stories across evaluations
  batch/
    tracker-additions/     # TSV files for merge

First Run

If ~/career-ops/ doesn't exist or is missing key files, enter onboarding — пошагово он расписан в modes/setup.md, начинай оттуда (там же готовые cp для примеров конфигов и разбор, какой из двух portals.example.yml брать). Коротко:

1. Create directory structure
2. Ask for CV (paste, LinkedIn URL, or dictate)
3. Fill config/profile.yml (name, targets, salary, location) — из config/profile.example.yml
4. Configure portals.yml — из config/portals.example.yml (не из templates/, см. modes/setup.md)
5. Create empty tracker
6. Get to know the user for better evaluations

Portal Strategy (3 tiers)

Tier 1: hh.ru (Russian market)

API-based scanning via hh.ru public API:

• GET https://api.hh.ru/vacancies?text={query}&area={area_id}&salary={min}&only_with_salary=true
• Area IDs: 1 (Moscow), 2 (SPb), 113 (Russia), 1001 (Remote)
• Filters: experience, schedule (remote/flexible), salary
• Returns JSON with structured data (title, company, salary, requirements)

Playwright scanning for full JD extraction:

• Navigate to hh.ru/vacancy/{id} for complete description
• Extract: requirements, responsibilities, conditions, skills, salary

hh.ru specific fields in profile.yml (значения — твои, из config/profile.yml):

hh:
  area_ids: [113, 1001]  # Russia + Remote
  experience: "moreThan6"  # noExperience, between1And3, between3And6, moreThan6
  schedule: ["remote", "flexible"]
  search_queries:        # 3-5 формулировок ТВОЕЙ роли, как её пишут в вакансиях
    - "<твоя роль>"
    - "<синоним роли>"
    - "<роль на уровень выше — для стретч-вакансий>"
  salary_from: <нижняя граница вилки>   # в валюте ниже
  currency: "RUR"
  only_with_salary: false

Формулировки берутся не из головы: прогони /career-ops analyze-skills, он соберёт частотный словарь требований по твоей роли с hh.ru и покажет, какими словами её ищут.

Tier 2: LinkedIn (Global + Russia)

Scanning strategy:

1. WebSearch with site:linkedin.com/jobs + role keywords
2. Playwright for authenticated browsing (if logged in)
3. LinkedIn skill linkedin for profile enrichment and outreach

LinkedIn-specific features:

• Generate connection request messages (300 char limit)
• Find hiring managers and recruiters
• Profile optimization suggestions per JD
• Easy Apply form assistance

Tier 3: Global ATS (Ashby, Greenhouse, Lever, Wellfound, Workable)

Same as original career-ops. See modes/scan.md for details.

Evaluation Pipeline (Blocks A-F)

When user pastes a JD or URL:

1. Extract JD (Playwright > WebFetch > WebSearch > ask user)
2. Block A -- Role Summary (archetype, domain, function, seniority, remote, TL;DR)
3. Block B -- CV Match (requirements vs cv.md, gap analysis with mitigation)
4. Block C -- Level & Strategy (detected vs natural level, up/down-level plan)
5. Block D -- Comp & Demand (WebSearch for salary data, market trends)
6. Block E -- Personalization Plan (CV changes, LinkedIn changes)
7. Block F -- Interview Plan (6-10 STAR+R stories mapped to JD requirements)
8. Block G -- Draft Application Answers (if score >= 4.0)
9. Save report to reports/
10. Generate PDF (ATS-optimized, keyword-injected)
11. Update tracker

Language Rules

• JD in English -> evaluation in English, CV in English
• JD in Russian -> evaluation in Russian, CV in Russian (separate template)
• hh.ru -> always Russian evaluation, but English CV option if company is international
• LinkedIn -> follow JD language
• Code and technical terms: always English

Scoring

Score 1-5 based on:

• Role-archetype match (weight: 30%)
• Technical requirements match (weight: 25%)
• Seniority alignment (weight: 15%)
• Comp vs target (weight: 15%)
• Remote/location fit (weight: 10%)
• Growth potential (weight: 5%)

PDF Generation

Uses generate-pdf.mjs (Puppeteer HTML->PDF):

node ~/.claude/skills/career-ops/generate-pdf.mjs input.html output.pdf --format=a4

Fonts: Space Grotesk (headings) + DM Sans (body). Single-column ATS layout.

Ethical Rules

• NEVER submit without user review -- fill forms, draft answers, generate PDFs, but STOP before Submit
• Discourage low-fit applications (score < 3.5)
• Quality over quantity -- 5 targeted > 50 generic
• NEVER invent experience or metrics
• NEVER modify cv.md without explicit permission

Integration with Existing Skills

Этот навык — про твоё трудоустройство: ты кандидат. Обратная задача (искать кандидатов, читать чужие резюме) в пак не входит — там чужие персональные данные.

Files Reference