Razbakov Skills

Fire-and-forget agent dispatch. Use when user says '/inbox: <task>' or 'inbox <task>'. Spawns a Claude agent in a tmux session with git worktree isolation. Logs everything for retry. Does NOT wait or check on the agent.

Standards and process for building modern web projects with AI. Content-first, mobile-first, component-based methodology. Use when starting a new project, creating features, designing components, making architecture decisions, or when the user asks to build something for the web.

Check status of all dispatched agents. Use when user says '/scrum'. Lists all wf-* tmux sessions with their status: running, done, failed, needs-input. Reads agent logs and git status.

A content-first, mobile-first methodology for creating modern websites. Use this prompt when starting any new website project.

Conduct and document research with consistent file organization. Use when researching topics, gathering information, or when the user asks to investigate a subject. Saves findings to research/yyyy-mm-dd-title.md format.

Orchestrate complex tasks with planning, subagents, and verification. Use when starting multi-step tasks, managing todo items, coordinating work with subagents, or when the user mentions planning, verification, or task management.

Interact with Atlassian Jira and Confluence from the terminal. Search, create, and update Jira tickets; read, search, create, and update Confluence pages. Use when the user mentions Jira tickets, issues, work items, sprints, Confluence pages, documentation, or Atlassian in general.

Generate high-quality, professionally styled PDF documents and reports using Python-generated LaTeX compiled with Tectonic. Use when the user asks to create a PDF, generate a report, build a PDF document, produce a printable report, or needs a polished multi-page document with tables, color coding, or statistical formatting.

Add internationalization (i18n) to a Nuxt 3 project — install @nuxtjs/i18n, extract hardcoded strings from components into locale JSON files, configure routing strategy, create a language switcher component, and fix hosting config for SSR locale routes. Use this skill whenever the user wants to translate a Nuxt site, add multiple languages, add a language switcher, localize content, or migrate a monolingual Nuxt site to multilingual. Also triggers when adding specific languages (e.g., "add German to my site") or when the user mentions i18n, localization, or translation in a Nuxt context.

Design your organization's governance, roles, and coordination. Use when multiple people or AI agents need clear accountability and decision-making.

Estimate the effort of user stories, tasks, bugs, or raw requirements using story points. Use when the user asks to estimate, size, or score a ticket, story, feature, or piece of work.

Use when ordering groceries from REWE Lieferservice, adding items to REWE cart, or reordering from a previous Flink shopping list. Requires user logged into rewe.de in Chrome with delivery slot selected.

Generate visual storyboards for video ads, product demos, and marketing videos. Creates a grid of frames with captions using Gemini image generation.

Transform requirements into structured user stories with acceptance criteria using INVEST principles. Use when the user asks to write a user story, create a ticket, define acceptance criteria, or convert requirements into dev-ready stories.

Write BDD scenarios grounded in existing UI/UX implementation. Researches pages, components, types, and mock data before writing any scenario. Use when generating .feature files for a product that already has a working app or prototype.

Use when ordering groceries from Flink, adding items to Flink cart, or reordering from a previous shopping list. Requires user logged into goflink.com in Chrome.

Designs and refactors large SKILL.md agent skills with decision trees, routing maps, gotcha capture, and command orchestration. Use when users ask to create, improve, consolidate, or scale complex skills.

Meta skill for browser automation — routes to the right browser tool. Use when you need to automate a browser task and aren't sure which MCP toolset to use (Claude in Chrome vs chrome-devtools vs agent-browser CLI). Also use when writing prompts for agents that need browser access, or when deciding between browser automation approaches.

Level 10 Life assessment — score 10 life areas (1-10), compare to previous assessment, identify patterns, pick 3 focus areas, create action plan. Use when the user says /year-review, 'let's do an assessment', 'L10L', 'level 10', or it's been 4+ weeks since the last assessment.

Generate poster images. Use when the user asks to create a poster, flyer, print material, social media graphic, or any visual marketing asset

Use when the user provides a GitHub issue URL, says "implement issue

Operate Google services from the terminal using gog. Use when requests involve Gmail, Calendar, Chat, Classroom, Drive, Docs, Slides, Contacts, Tasks, People, Sheets, Forms, Apps Script, Groups, or Keep workflows, including auth checks, data listing/search, file transfer, and command automation.

Convert raster images (JPEG, PNG) to SVG vector graphics using Potrace. Use when the user asks to vectorize an image, convert a logo to SVG, trace a bitmap, or create vector graphics from a raster image.

Design Sprint Facilitation — guide any idea through a structured 6-phase sprint. Use when the user says 'design sprint', 'sprint my idea', 'facilitate a sprint', 'run a design sprint', or wants structured ideation with decision checkpoints.

Connect a Telegram bot to any Claude Code agent for collaborative group chats. Use when the user wants to set up a Telegram chat where external people (co-founders, partners, clients) can interact with an AI agent. Triggers include "connect bot to agent", "set up group chat with agent", "add person to agent chat", "envoy connect", or when creating collaborative Telegram groups with AI agents.

Improve an existing Cursor agent skill using prompt engineering best practices. Analyzes a SKILL.md file and rewrites it for clarity, effectiveness, and proper structure. Use when the user asks to improve, refine, optimize, or review a skill.

Set up LiteLLM AI Gateway proxy with Docker Compose for Claude Code Max subscription. Use when the user wants to route Claude Code through LiteLLM for cost tracking, budget controls, or usage monitoring. Also trigger when user mentions "litellm", "AI gateway", "proxy for Claude Code", "track Claude usage", "Claude Code billing", or wants to set up a local proxy between Claude Code and Anthropic API.

Discover the right problem, validate the solution, then build with confidence. Use when starting something new, unsure what to build, or onboarding to an existing workspace.

Build shared understanding before building software: README → user journey → story map → architecture → AI context. Use when starting a new project, onboarding to an existing one, or when AI needs better context about what the project is and why it exists.

Transform video/image lighting, backgrounds, and environments using Beeble SwitchX API. Use this skill when the user wants to relight a video, replace a video background, improve lighting in footage, make a clip look more cinematic or professional, add studio lighting to a recording, or do any video-to-video visual effects transformation. Also trigger when the user mentions Beeble, SwitchX, or wants VFX on short clips.

Transform content into viral social media threads using proven formulas. Use when the user asks to create a thread, write a viral post, turn research into content, or promote findings on X/Twitter.

Saturday weekly review + planning. Reviews the past week (sessions, calendar, projects), updates PROJECTS.md, reflects on focus areas, then plans next week with calendar blocks. Use when the user says /weekly-review, 'weekly review', 'weekly planning', or it's Saturday.

Generate logo concepts using Gemini image generation. Creates a 3x3 grid of variations for review, then iterates on the chosen direction.

Use when auditing or setting up SEO for a Nuxt site. Triggers include missing page titles, no structured data, sitemap not including all pages, hreflang/canonical issues, pages not prerendered, trailing slash redirect loops, AI crawlers blocked in robots.txt, or preparing a site for search engine indexing.

Personal coaching session — assessment, check-in, unblock, build jam, journal, GROW, or open conversation. Use when the user says /personal-coach, 'coach me', 'let's do an assessment', 'I'm stuck', 'help me find direction', or wants to reflect/process something.

Build a complete press kit for an event, product launch, or campaign — in multiple languages — and publish it as a shareable Google Drive folder ready to send to journalists, partners, or a delegate. Produces press releases (typically DE/EN/ES, or configurable), uploads press photos and flyers, creates an Overview document for at-a-glance briefing, and creates a Handover document with pending tasks, contacts, risks, and decisions so press distribution can be delegated. Use when the user says 'I need a press release', 'create a press kit', 'press release in X languages', 'set up a Drive folder for press', 'handover doc for someone else to run press', or has an upcoming announcement that needs to be sent to media. Trigger generously: even partial requests (just a press release, just a flyer folder) typically evolve into the full kit.

Dispatch parallel agents to address review comments on all open PRs. Each agent fixes code, replies inline, resolves threads, and reports what needs human input. Use when user says '/review-all-prs', 'fix all PR reviews', 'address all reviews', or when /scrum shows PRs with pending reviews.

Use when reviewing user stories, auditing backlog quality, or checking story consistency before sprint planning. Use when docs/backlog.yaml and docs/issues/ exist and need quality validation.

Generate images from self-contained HTML/CSS designs. Use when the user wants a reproducible, code-first pipeline to produce print or social graphics (posters, flyers, ads) from HTML and screenshot automation.

Run a sprint planning session — select stories from backlog, estimate them, create GitHub issues with labels/milestone/project board, document the sprint in the repo, and open a PR. Use when the user says /sprint-planning, 'plan a sprint', 'sprint planning', 'what should we build next', or wants to select and organize work for a time-boxed iteration.

Track ticket sales for a live event (concert, festival, conference, workshop) with daily snapshots, generate a burndown chart comparing actual sales to ideal-linear targets and tier-cumulative milestones, and report whether the event is on pace. Use when the user asks how sales are going, wants to know if their event will sell out, asks for a daily sales report, wants to set up sales tracking for an upcoming event, or asks about ticket pace / velocity / projection. Trigger generously: phrases like 'how is concert sales going', 'burndown for my event', 'are we going to sell out', 'sales velocity', 'daily ticket chart', 'how many tickets do we need to sell', or any case where the user has a ticketed event with a fixed sales window and wants visibility on pacing.

Daily SEO health check for dancegods + razbakov.com. Checks indexing status, sitemap, meta tags, and structured data. Reports findings and flags issues for the daily review pipeline. Use when running the morning SEO check or when asked for an SEO status report.

Produce a print-ready Event Pack for a SPECIFIC upcoming event (has a date, venue, URL) using Gemini for the design and a real QR overlay. Builds 5+ assets — A6 flyer front/back, A4 4-up home-printable, DJ tent card, 1080×1920 Instagram story — by prompting Gemini to render each full asset (title, date, venue, price, dresscode, and QR *placeholder*) and then compositing a real scannable QR over the placeholder. Use when the user has a concrete event to promote (party, workshop, gig, pop-up, launch) and needs physical + digital promo together. For brand-only cards or single posters without an event, use brand-poster instead.

Generate images from LaTeX/TikZ compiled with Tectonic and converted to PNG via ImageMagick. Use when the user wants vector-quality graphics, thumbnails, diagrams, or visual assets with precise typography and geometric elements from code.

Generate OBS stream overlay assets (background, overlay, starting-soon, BRB screens) and configure OBS scenes via WebSocket. Use when the user asks to set up OBS, create stream overlays, design a live stream layout, make their stream look nicer, or mentions OBS scenes/sources. Also trigger when the user says "set up my stream", "create stream graphics", "OBS overlay", or wants branded live stream assets.

AI content and SEO agent that audits sites, fixes indexing issues, writes blog posts aligned with OKRs, generates YouTube metadata, and produces monthly analytics reports. Use when asked to run SEO audit, write a blog post, create content calendar, fix indexing, generate YouTube descriptions, or produce an analytics report.

Generate images using Google Gemini API. Use when the user asks to generate, create, or make an image, illustration, poster design, mockup, thumbnail, or any visual asset using AI. Supports uploading reference photos for poster/flyer generation with real people. Also trigger when the user mentions Gemini image generation or asks to visualize something as an image.

Generate QR codes with URLs and UTM tracking. Exports PNG/SVG with captions. Supports styled QR codes with custom module shapes (rounded, circle, gapped, bars), colors, gradients (radial, square, horizontal, vertical), and logo overlay. Use for single codes, batch generation, or marketing campaigns with tracking parameters.

Summarize one or more YouTube videos from their links. Use this whenever the user pastes a youtube.com or youtu.be URL (or several) and wants to know what it's about — phrasings like "summarize to telegram", "tldr these videos", "what do these say", "summary of this talk", or just dropping links with no instruction at all. Fetches each video's real transcript via yt-dlp (not the page text, which never contains the transcript), cleans the captions, and writes a per-video summary. Default delivery is Telegram; honor any other surface the user names ("to my notes", "just here in chat", "email it"). Trigger even when the user only pastes bare links — bare YouTube links almost always mean "tell me what's in these".

Use when the user says "develop PP-XYZ" or asks to work on, develop, or implement a ticket, issue, or story referenced by ID

Search a freelance platform for jobs matching user's skills, verify each listing is live with actual budget, and create a project with verified opportunities and draft proposals.

Produce AI avatar YouTube videos using the Julia McCoy method: write scripts with Claude, generate avatar with HeyGen, clone voice with ElevenLabs. Use when the user wants to create YouTube content at scale using AI avatars, plan an avatar video pipeline, write video scripts, or audit readiness for AI-powered video production. Triggers: 'create avatar video', 'julia mccoy method', 'AI YouTube pipeline', 'avatar content plan', 'write video script for avatar', 'scale YouTube with AI'.

Process a livestream transcript into publishable content: blog post (all languages), YouTube metadata with chapters, thumbnail, hero image, and publish. TRIGGER when: user shares a livestream transcript, mentions processing a livestream, says "process stream", references a YouTube live stream recording they want to publish, or pastes a transcript with timestamps. DO NOT TRIGGER for: pre-recorded video editing, podcast episodes without video, or general YouTube video uploads without a transcript.

AI sales and business development agent that manages lead pipelines, prepares outreach messages, tracks deals, updates CRM data, and produces weekly pipeline reports. Use when asked to find leads, prepare outreach, update pipeline, close a deal, track sales, onboard a partner, or run a sales sprint.

Scan Gmail for subscription and purchase receipts, extract amounts, categorize spending, and produce a markdown tracker with optional Notion database sync. Use this skill whenever the user wants to: audit subscriptions, review recurring charges, check purchase history, analyze spending, find what they're paying for, cancel unused services, track monthly expenses, check burn rate, review SaaS costs, look at email receipts/invoices, update a subscription tracker, or do any kind of expense review based on email data. Even if the user just says something casual like "what am i paying for" or "check my spending" or "how much do my subscriptions cost" — this is the right skill. Also use when the user wants to sync subscription data to Notion or create a spending breakdown by category.

Export and share AI chat sessions from Claude Code, Cursor, or Conductor. Lists sessions across providers, exports to markdown, publishes as a GitHub Gist. Use when the user asks to export, share, backup, or review chat history.

Use when a PR has review comments to address - fetches inline comments from GitHub, fixes code issues, replies to each comment, and resolves threads. Trigger on "address the review", "fix PR comments", "handle review feedback on PR

Apply storytelling tactics to craft narratives, recommend story structures, and transform content (threads, pitches, posts). Based on Pip Decks Storyteller Tactics by Steve Rawling. Use when the user asks to tell a story, write a pitch, craft a narrative, improve content with storytelling, create a compelling post, or mentions story arcs, storytelling, or Pip Decks.

Translate one English blog post into multiple target languages via parallel sub-agents, preserving frontmatter conventions, hero image, and brand voice. Use when the user shares a published English post URL or markdown path and says 'translate it', 'add other languages', 'publish in DE/ES/RU/UK', 'translate to 5 languages', or asks for localized versions of a specific post.

Drive two running UIs through matching states, capture side-by-side screenshots at a locked viewport, and produce a markdown + PDF report with per-screen diff notes. Use whenever the user asks to compare a prototype against a port, staging vs production, before vs after a refactor, "does the live site match the design", pixel-fidelity sign-off, visual regression check, or 1:1 parity audit — even if they don't say "audit". Strongly prefer this over ad-hoc screenshotting whenever two versions of the same UI need visual comparison.