DOCSEARCHbyAlgolia
Submit Docs

Polyakov Claude Skills

repository·main·Indexed 41 days ago

https://github.com/artwist-polyakov/polyakov-claude-skills

A collection of specialized skills and plugins for Claude Code, including the Codex Review plugin for cross-agent reviews using GPT models, the genome-analizer skill, and crawl4ai-seo for web scraping and SEO workflows.

Tokens
208.8K
Snippets
483
Records
881
Agent score
52%

What's inside Polyakov Claude Skills

  1. Overview of ZoomKit features and benefits

    main

    ZoomKit is an automation service for advertising management. Key capabilities include:

    • Yandex.Direct Bid Management: Automatically updates bids every 30 minutes.
    • Landing Page Monitoring: Daily checks of landing pages; automatically stops ads if an error is detected and resumes them once fixed.
    • Budget Forecasting: Predicts when advertising account funds will run out and provides balance top-up reminders.
    • Bulk Operations: Enables mass changes to settings and scheduled changes to advertisements.
    • Multi-account Management: Allows working with multiple advertising accounts simultaneously.
    • Data Integration: Uploads advertising expenses to Yandex.Metrica to help combine them with conversion data.

    When explaining value to users, focus on specific use cases: unified management for multiple accounts, risk reduction via automatic error detection, and reduced manual reporting through Metrica integration.

  2. Overview of the Yandex Direct Skill

    main

    The yandex-direct skill enables interaction with the Yandex Direct advertising platform via its official API. It allows for analyzing statistics, conversions, and search queries, performing audits, and managing advertising accounts (creating/modifying campaigns, ads, phrases, bids, retargeting conditions, and settings). It also supports data export and previews.

    Core Workflow Pattern: Regardless of whether you are working with a single ad, a campaign, or an entire account, follow this sequence: Task → Required Data → Calculation & Explanation → Action → Result Verification.

    Execution Requirements:

    • Use uv run scripts/... or python3 scripts/... (requires Python 3.11+).
    • No external libraries are required.
    • Command-specific parameters can be found using the --help flag.

  5. Use the Scrape.do Web Scraper skill

    main

    The scrapedo-web-scraper skill provides web scraping capabilities via the Scrape.do API. It is designed to bypass blocks, CAPTCHAs, and Cloudflare protections.

    When to use: Automatically use this skill when standard WebFetch operations fail with the following errors:

    • 403 (Forbidden)
    • 401 (Unauthorized)
    • 429 (Too Many Requests)
    • timeout
    • access denied
    • Cloudflare block

  6. Use the genome-analyzer skill

    main

    The genome-analyzer skill is designed to analyze a user's genetic data from a VCF file. It should be used when a user asks about their genetics, hereditary traits, predispositions, metabolism (e.g., caffeine, alcohol, medications), athletic abilities, disease risks, or gene-based nutrition.

    To use this skill, ensure a .vcf file is present in the current working directory.

    --- 
name: genome-analyzer
description: Анализирует генетические данные пользователя из VCF файла. Используй когда пользователь спрашивает о своей генетике, наследственных признаках, предрасположенностях, метаболизме веществ (кофеин, алкоголь, лекарства), спортивных способностях, рисках заболеваний, питании на основе генов.
---

  7. Understand the repository structure

    main

    The repository is organized into a marketplace configuration and a collection of individual plugins (skills).

    • .claude-plugin/marketplace.json: Contains the configuration for the Claude marketplace.
    • plugins/: A directory containing various specialized skills, such as web scrapers, API connectors (Yandex, Perplexity, ZoomKit), and publishing tools (Telegraph, GitHub Pages).

  8. What is the knowledge-compiler skill?

    main

    The knowledge-compiler is a specialized skill designed to transform personal books, PDFs, EPUBs, TXTs, or Markdown files into an applied Claude Code skill. Instead of creating a simple summary, it builds a functional knowledge map containing:

    • Applied Ideas: Heuristics, decision rules, playbooks, and anti-patterns.
    • Source Mapping: Every significant thesis is linked to its specific location in the source (chapter, heading, or character range).
    • Structured References: A glossary, concept list, and decision rules.

    It is intended for creating personal advisors, reference guides, or reusable skills from technical manuals, management books, or internal reports. It focuses on compilation (extracting utility) rather than republication (copying text).

  9. Reporting and Attribution for Experiments

    main

    When evaluating experiments, ensure you compare metrics using identical goals and attribution models.

    • Default Reporting: Use AUTO for reports by default.
    • Strategy Models: Read the current strategy model from the campaign itself (or from the package if using a batch strategy). If there is a discrepancy, suggest the user choose a reporting model or compare both.
    • Attribution Preference: Preferring AUTO allows for suggesting a separate campaign setting change (refer to PLAYBOOK.md#автоматическая-модель-атрибуции for details).
    • Master of Campaigns (Мастер кампаний): Automatic attribution is mandatory for this tool.

    Interpreting Results

    Show both the specific result of the experiment and the overall campaign/account outcome.

    Caution on 'Before and After' comparisons: A simple comparison shows change but does not prove causation, as demand, auctions, and neighboring campaigns change simultaneously. A stable result could mean either a redistribution of leads or a new result compensating for a drop in overall demand. For more reliable conclusions, a simultaneous control group (a comparable part of demand without changes) is ideal, though not always possible to organize. Stopping a campaign to compare periods does not create a valid control group.

  10. Select a fal-ai image backend model

    main

    The fal-ai-image skill supports two backend families via a single FAL_KEY. You can choose between nano-banana/google/gemini for cost-effective runs and strong text rendering, or gpt/openai for high photorealism and prompt adherence.

    Alias familyResolved endpointBest for
    nano-banana, google, geminifal-ai/nano-banana-procheaper default runs, infographics, strong text rendering
    gpt, openaiopenai/gpt-image-2stronger prompt adherence, photorealism, precise editing

  11. Distinguish between 'Any Visit' and 'Last Visit' in segments

    main

    When using ym:s:date, the condition checks if a user had at least one visit matching the criteria. It does not define the date of their last visit.

    Example Scenario: If a user visited 5 days ago (matching a range) and also visited today, they will still be included in a segment targeting '2 to 7 days ago'.

    Warning: Do not use ym:s:daysSincePreviousVisit==1 as a substitute for 'last visit was yesterday'. That grouping specifically means exactly one day passed between the current and the previous visit, which is a different logic.

  12. Understand Read-Only and Restricted Fields in Yandex Direct Ads

    main

    Certain fields in the Yandex Direct API cannot be modified after creation or are restricted to read-only access.

    Key Restrictions:

    • Immutable Fields: Type (set at creation), AdCategories (cannot be changed via API), and FeedId for ShoppingAd/ListingAd (cannot be changed after add).
    • Deprecated Fields: TurboPageId and VCardId (values will not be saved), Mobile and PreferVCardOverBusiness (always treated as NO).
    • Status Restrictions: Ads with status ARCHIVED or ads within an archived campaign cannot be edited.
    • Age Labels: AgeLabel values can be changed, but you cannot add or remove a label via the API (requires support).