Video & Audio Production Skills

Mix a music / audio track onto an existing video via ffmpeg. Three modes: replace (drop original audio), overlay (mix both audible), duck (sidechain-compressor lowers music when speech is detected). Volume + fade-in + fade-out controls. No API calls — pure ffmpeg. Use when the user says 'mix this music onto the video', 'add background music to video', 'duck the music under voiceover', 'смешай музыку с видео', 'фоновая музыка для видео'.

Orchestrator skill — turn a user photo into N profile-pic / headshot / avatar variants in a consistent style. Wraps image-prompt --execute + the carousel style library + the runner's batch executor. Identity preserve is THE differentiator — defaults to nano-banana-pro for best face preservation. Multi-aspect output (square 1:1, square-tight 1:1 cropped, cover-banner 4:5, story 9:16). Modes: --photo --style auto| --variants N --aspects square,cover,story. Outputs: ./generated/avatar/ / -v .png + manifest.json. Use when the user says 'make me an avatar', 'profile picture from this photo', 'headshot variants', 'LinkedIn profile pic', 'аватарка', 'портрет', 'хедшот'.

Banner-ad / display-creative generator with standard-size presets — Google Display (728×90 leaderboard, 300×250 medium rectangle, 320×100 mobile banner, 160×600 wide skyscraper), LinkedIn ad (1200×627), OG image (1200×630), Twitter card (1500×500), Facebook ad (1200×628), generic header (1456×180). Dense text + CTA aesthetic, ad-feel composition. Wraps image-prompt --execute + the carousel style library with text-heavy bias. Default model ideogram-3-quality for clean embedded text. Outputs: ./generated/banner/ / .png + manifest.json. Use when the user says 'banner ad', 'display ad', 'Google Display creative', 'OG image', 'Twitter card', 'баннер для рекламы', 'OG-картинка'.

Remove backgrounds from images using FAL.ai's BiRefNet model. Use when users ask to remove background, make transparent PNG, extract subject from image, or create cutouts. Trigger phrases include "remove background", "transparent background", "cut out", "extract subject", or any background removal request.

Read-only story-bible auditor for fiction series with a documented canon. Cross-references character / artifact / location mentions in chapters against the bible and flags drift. Produces a drift report — never edits files. Use before opening or committing a chapter, or when reintroducing an established entity.

Orchestrator skill — turns a topic or research brief into an N-slide Instagram / LinkedIn / TikTok carousel with consistent visual style and ready-to-post captions. Wraps essay-write + viral-text (for content) + image-prompt --execute (for slides) + common style library (24 visual styles). Modes: --topic / --research; --style auto| |--style-ref ; --slides 3-12; --platform instagram|linkedin|tiktok; --aspect portrait|square|story; --text-mode embedded|overlay|none; --execute; --resume. Outputs: ./generated/carousel/ /slide-{1..N}.png + captions.md + manifest.json. Use when the user says 'make a carousel about X', 'turn this research into a post', '8 slides on Y', 'carousel for LinkedIn'.

Orchestrator skill — turn cover metadata (title / creator / subtitle / medium) into an album / book / podcast / report / deck / magazine cover. Wraps image-prompt --execute + the carousel style library (24 visual styles) + the runner's batch executor. Picks aspect by medium (album 1:1, book 2:3 portrait, podcast 1:1, magazine 2:3, report 1:√2 A4). Multi-variant output. Optional photo / artwork reference. Outputs: ./generated/cover/ / .png + manifest.json. Use when the user says 'album cover', 'book cover', 'podcast cover', 'report cover', 'обложка для альбома / книги / подкаста / отчёта'.

Write or edit non-fiction prose — essays, popular-science chapters, longreads. Wraps writer; adds source-backed claims, Manson-style ironic coda, mechanism over surface, sparing metaphors, V/H/P hypothesis markers. Use when drafting non-fiction that needs argumentative weight — longer than a viral post, shorter than a doctoral chapter.

Orchestrator skill — turns event details (title / date / location / CTA) plus an optional photo into a poster/flyer/social-event-graphic with embedded text in a chosen visual style. Wraps image-prompt --execute + the carousel style library (24 visual styles) + the runner's batch executor. Modes: --title --date --location [--cta ] [--photo ] [--style auto| ] [--aspects portrait,square,story,landscape,a4]. Picks a text-friendly + multi-ref-capable model (gpt-image-2 / ideogram-3-quality / nano-banana-pro). Outputs: ./generated/flyer/ / .png + manifest.json + prompts.md. Use when the user says 'make a flyer for X', 'event poster', 'workshop flyer', 'meetup poster', 'афиша', 'плакат', 'флайер для мероприятия'.

Short looping GIF utility. Two modes: (A) convert an existing MP4 to GIF with 2-pass palette optimization; (B) generate a 1-3 second clip via a video provider (Veo / Sora / Kling / Runway / fal-video) then convert. Aspect crop presets (1:1 social, 9:16 story, 16:9 banner, 2:1 wide). ffmpeg-driven (palette generation + paletteuse for quality). Outputs: ./generated/gif/ .gif. Use when the user says 'gif', 'looping animation', 'short loop', 'make a gif', 'сделай гифку', 'короткая зацикленная анимация'.

Write marketing copy — landing page sections (hero/features/pricing/FAQ), SEO meta (title+description+OG+Twitter), ad copy (Google/Facebook/LinkedIn/X). Julian Shapiro hero formula, char limits per platform. Wraps writer. Use when user says 'landing page copy', 'SEO meta', 'Google Ad', 'hero section'.

Brand mark / wordmark / logo generator. Defaults to ideogram-3-quality for cleanest embedded text (other text-strong fallbacks: gpt-image-2). Single-image output (no aspect multiplexing); N variants per call differ only in stochastic interpretation. Six style presets (wordmark / minimal / illustrated / typographic / geometric / emblem) + optional palette hint. Outputs PNG (typically isolated on white via prompt hint; pipe through bg-remover for transparent BG). Use when the user says 'logo', 'brand mark', 'wordmark', 'логотип', 'эмблема', 'фирменный знак'.

Meme-format graphic generator — top text + bottom text + optional centerpiece image. Wraps image-prompt --execute with a meme aesthetic anchor (Impact-style bold typography with stroke, classic internet meme template feel). Supports 5 popular templates as composition hints (drake, distracted-boyfriend, expanding-brain, two-buttons, custom). Optional --base-photo to use a user image as the centerpiece. Default model gpt-image-2 (good at illustration + text + visual integration). Outputs: ./generated/meme/ /meme-v .png + manifest.json. Use when the user says 'meme', 'meme template', 'meme image', 'мем', 'сделай мем'.

Write prompts for 10+ frontier AI music generators (Suno v5.5, Udio v4, Google Lyria 3 Pro, ElevenLabs Music, Stable Audio 2.5, MusicGen, Tencent SongGeneration, Sonauto v2, Riffusion, Mubert). 2026 canonical meta-tag taxonomy (Structure / Vocal delivery / Vocal effects / Instrumental / Mix-production / Energy-dynamics / Era-genre / FX), | stacking, two-box Style+Lyrics workflow (Suno), exclude-styles (Eleven), field-driven (Lyria). Lyrics conventions: ad-libs, repetition, section headers, language switching. Use when the user says 'prompt for Suno / Udio / Lyria / Eleven Music', 'song about X', 'AI music prompt', 'meta tags for this chorus', 'cover song style', 'orchestral cue'.

Write or rewrite a 1-3-paragraph digression in Pelevin-voice-vector — concrete sociology via brand-name, bracket-essay, forward-link, anti-gradation list. Wraps prose-edit (fiction) or essay-write (non-fic) as final cleanup. Use when the user asks for a digression in this specific voice. Vector, not impersonation.

Fiction-prose rewrite + style pass. Wraps writer; adds voice vector (Pelevin/Manson — not impersonation), no-business-editing rules, artistic rewrite over compression, 5-trigger structural-synthesis detector, depth-pass checklist. Use when rewriting a fiction chapter / scene / passage. Pairs with canon-check.

Quote card / aphorism graphic generator — bold short text + attribution + minimal visual. Output: text-dominant composition where typography IS the image (1080×1080 square, 1080×1350 portrait, 1080×1920 story). Wraps image-prompt --execute + the carousel style library (24 visual styles) with a text-friendly bias. Default model: ideogram-3-quality (cleanest embedded text). Multi-aspect for Twitter / Instagram / LinkedIn quote posts. Outputs: ./generated/quote/ / .png + manifest.json. Use when the user says 'quote card', 'aphorism graphic', 'pull quote post', 'цитата для соцсетей', 'афоризм на картинке', 'постер с цитатой'.

Orchestrator skill — turns a topic / research brief / script into a 9:16 vertical reel: 1-4 video shots + matched background music + ffmpeg-stitched final.mp4 with optional burned-in captions. Wraps viral-text (for script) + video-prompt --execute + music-prompt --execute + common video/music style library + ffmpeg. Modes: --topic / --research / --script-file; --shots 1-5; --shot-duration ; --style auto| ; --music-style auto| ; --aspect vertical|square|horizontal; --captions on|off; --execute; --resume. Outputs: ./generated/reel/ /final.mp4 + shots/ + music.mp3 + script.md + manifest.json. Use when the user says 'make a reel about X', 'short video on Y', 'TikTok / Reels / Shorts about Z'.

Produce a structured research brief on any topic — TL;DR, key facts with citations, notable quotes, suggested angles, open questions. 3-15 queries by depth, multi-source (WebSearch + WebFetch + optional Firecrawl/Exa MCP). Output is a markdown file ready for downstream consumption by carousel-builder, reel-builder, viral-text, essay-write, landing-copy. Use when the user says 'research X', 'gather facts on Y', 'brief me on Z', 'what's known about W', 'prepare research for a post / carousel / reel about Q'.

Write engineer-facing design documents — RFCs, ADRs (Architecture Decision Records), Tech Specs, Design Docs. Structure: context / problem / proposal / alternatives / consequences / decision / open questions. RFC 2119 keywords (MUST/SHOULD/MAY). Wraps writer. Use when the user says 'write an RFC for...', 'design doc for...', 'ADR for...', 'tech spec for...'.

Manage API keys for the runner's --execute layer. CRUD on ~/.skills.env (chmod 600): list / add / update / remove / enable / disable gate flags / verify (ping vendor APIs) / export (eval-ready). Single source of truth for OPENAIAPIKEY, GEMINIAPIKEY, BFLAPIKEY, FALKEY, REPLICATEAPITOKEN, RUNWAYAPIKEY, KLINGACCESSKEYID/SECRET, SUNOAPIKEY, ELEVENLABSAPIKEY, IDEOGRAMAPIKEY, ANTHROPICAPIKEY, S3 + gate flags (LYRIAAPIENABLED, SUNOAPIENABLED, OPENAISORAAPIENABLED). Explicit shell exports always win over file entries. Use when the user says 'add my OpenAI key', 'rotate the Suno key', 'check which keys are set', 'verify my Gemini key works', 'where do I put my keys'.

Manage the local style library — bundled styles (24 carousel + 12 video director + 12 music genre) plus user overrides at ~/.claude/style-library/ / .md. CRUD on user styles (add / edit / remove / show / list / diff / validate / path) and build upstream-PR submission packages (submit). Frontmatter + body schema validation per modality. Templates ship with the library. Use when the user says 'add a custom carousel style', 'create a directorial style', 'genre preset for k-pop', 'list available styles', 'check my style is valid', 'submit my style upstream', 'show what's in kinfolk-minimal', 'edit the soft-watercolor style'.

Read-only pre-commit lint over writer/prose-edit/essay-write. Takes a staged diff (or a specified file / commit range) and flags neuro-slop, synthetic structures, voice drift in prose. Read-only — never edits. Use before git commit, as a post-commit check, or as a pre-commit hook.

Apply an artistic style to an existing image. Default provider Flux Kontext (best for natural-language style transfer). 12 style presets (watercolor / oil-painting / sketch / line-art / ink-wash / cyberpunk / studio-ghibli / pixar-3d / manga / art-deco / low-poly / vaporwave) + custom mode with --prompt-mod. Single image in, stylized image out. Use when the user says 'style transfer', 'make this look like X', 'turn this photo into a watercolor', 'стилизуй фото', 'переведи в стиль X', 'переведи в акварель / манга / гибли'.