deepak-skill

deepak-skill 🎬

   ██████╗ ███████╗███████╗██████╗  █████╗ ██╗  ██╗
   ██╔══██╗██╔════╝██╔════╝██╔══██╗██╔══██╗██║ ██╔╝
   ██║  ██║█████╗  █████╗  ██████╔╝███████║█████╔╝
   ██║  ██║██╔══╝  ██╔══╝  ██╔══██╗██╔══██║██╔═██╗
   ██████╔╝███████╗███████╗██║  ██║██║  ██║██║  ██╗
   ╚═════╝ ╚══════╝╚══════╝╚═╝  ╚═╝╚═╝  ╚═╝╚═╝  ╚═╝

🎬 deepak-skill — crafted by Deepak · deepak-skill on GitHub · MIT

Open-source agent skills for the full content-creator stack — short-form video (text-motion + asset reels), Google Flow / Veo cinematic scene prompts, viral hooks & storyboards, AI voiceover + sound design, native-image-model carousels, designed ebook lead magnets, spam-free HTML emails with high-CTR subject lines, long-form repurposing (podcast→shorts, YouTube packaging, SEO blog content), LinkedIn personal branding, platform-native social content plans (algorithm-reset calendars), vibe-coded web apps, a meta skill-builder that scaffolds new skills, and a mcp-agent-builder that scaffolds MCP servers + the AI agents that use them (full path: discovery → PRD → architecture → runnable server → agent wiring, in any IDE/CLI) — plus the business side: proper branding (branding-studio: idea → researched naming + validation + complete brand book with visual + voice), product positioning (positioning-studio: one-liner + messaging hierarchy + voice guide), a prompt library in your voice (prompt-engineering: framework + test loop), and AI workflow design (ai-automation: worthiness gate → steps → human checkpoints → cost → build handoff). Built for AI coding agents (Claude Code, Cursor, Codex, Gemini CLI, Antigravity CLI, Grok Build, and 70+ more via the skills ecosystem).

What's inside

Skill	What it does
text-motion-reels	Create trending text-only motion graphic reels (kinetic typography) as HyperFrames HTML compositions — 11 trending formats (Word Pop · Highlighter · 3D Editorial · Card Listicle · Chat Thriller · SVG Ambient · Micro-Fiction · Quiz Trap · Day Counter · Notification Drop · Thread Court), design styles, viral formulas. Copy follows the top-5-creator playbook (Hormozi / MrBeast / Bartlett / Gadzhi / Jenny Hoyos rails — hook ≤ 8 words, one claim per beat, zero hashtags on screen AND in captions, one CTA, loop ending — built to chase millions of views). Starts with an interactive format wizard and renders at 4K — audit harness: audit-reel.mjs (format slug, GSAP timeline, determinism, output, copy rails, captions → reel-audit.md) + a fresh reel-auditor subagent signing PASS / FIX NEEDED
hook-storyboard-retention	Write scroll-stopping hooks, engineer watch-time, and build beat-by-beat storyboards with script ↔ video timeline in sync — audit harness: audit-storyboard.mjs (hook/beat/sync/retention/determinism → storyboard-audit.md) + a fresh storyboard-auditor subagent signing PASS / FIX NEEDED
voice-sfx-audio	Open-source TTS voiceovers (Kokoro, Piper, etc.), royalty-free SFX/music sources with license guidance, FFmpeg audio ducking, and a deep-voice recipe (Tier 1: Kokoro male voice + FFmpeg pitch/EQ — see voice-plan.md) — audit harness: audit-audio.mjs (license compliance, mix contract, loudness → audio-audit.md) + a fresh audio-auditor subagent signing PASS / FIX NEEDED
video-asset-reels	Build reels from your own video clips & images — understand the prompt, cut assets to beats, overlay kinetic text, sync a voiceover, render 4K (see PLAN.md) — audit harness: audit-asset-reel.mjs (storyboard, cuts, HTML determinism, output, captions → asset-reel-audit.md) + a fresh asset-reel-auditor subagent signing PASS / FIX NEEDED
video-product-pipeline	The viral-engineered gatekeeper for every video request: hunt trending topics (trend-hunt.mjs + web research) → brainstorm + score angles → analyze ANY prompt (sloppy or not) → write a video-product.md spec → get your approval → generate (text-motion / asset reels) → audit the frames with a script + dedicated auditor subagent (spelling, text overlap, safe zones, style, readability)
carousel-post-images	Scroll-stopping carousel posts (LinkedIn/Instagram) as image sets — one script, two CLI modes: browser render (slides.html → deterministic 4K PNGs, any CLI) or native image-model generation (same deck exports photoreal 4K prompts for Antigravity / Codex image_gen / Grok /imagine). Photorealistic real-life visuals, viral anti-fluff copywriting, 4 trending styles (Cinematic Real-Life default), per-platform caption.md — audit harness: audit-carousel.mjs (deck, copy limits, fluff, output, captions → carousel-audit.md) + a fresh carousel-auditor subagent signing PASS / FIX NEEDED
ebook-builder	Build beautiful lead-magnet ebooks (guides / checklists / frameworks) as print-ready A4 PDF + 4K page PNGs + cover.png — six layout presets (Editorial Classic · Modern Bold · Minimal Luxury · Playful Pop · Technical Dark · Nature Calm), 30 design palettes (5 per layout) + typography pairings + 6 cover styles (full-bleed / solid / split / pattern / scene-frame / duotone) + page-motif families (step-cards / timeline / checklist / scenario / quote / chapter-dividers / comparison / framework-map) + mood/texture — ONE of each locked via the design picker, story-first structure (cover opens a loop → chapters escalate → payoff → CTA), and two generation modes like carousel-post-images (Mode 1 browser: render-ebook.mjs → headless Chrome A4 PDF + page PNGs; Mode 2 image-model: the same deck exports photoreal cover + interior scene prompts). Cross-session author memory + taste — ebook-memory.md (identity, voice fingerprint, design defaults, past builds) read at Stage 0 and written back at Stage 8, so every ebook is built in the author's voice with a taste profile (tone, rhythm, pet phrases, banned words the audit FAILs on). Cover-image brief built in — audit harness: audit-ebook.mjs (pages, cover, layout + palette + motif consistency, copy limits, fluff, memory/taste, output → ebook-audit.md) + a fresh ebook-auditor subagent (scores copy punch + voice match) signing PASS / FIX NEEDED
email-marketing	High-converting, spam-free HTML emails for Brevo / MailerLite / Mailchimp / Klaviyo — bulletproof responsive code, anti-fluff copywriting, E-E-A-T trust signals (real author + credentials + proof), and a high-CTR subject line pack (33–50 char formulas + A/B plan). Validated by validate-email.mjs (spam-trigger scan, CAN-SPAM compliance, EEAT signals → validation-report.md) and previewed at desktop + mobile via preview-email.mjs before send
thumbnail-studio	Engineer click-worthy YouTube thumbnails — a CTR teardown of the niche first (what wins, what's saturated, the ONE idea the video sells) → 3–5 thumbnail variants (each = ONE idea + ≤ 5-word overlay + face/emotion/contrast + a 1280×720 photoreal image-gen prompt, self-verified by thumbnail-prompts.mjs) → A/B plan with niche CTR benchmarks (templates/ctr-benchmarks.md) — audit harness: audit-thumbs.mjs (teardown, variants, overlay caps, anti-cliché blocklist, A/B plan → thumbs-audit.md) + a fresh thumbs-auditor subagent signing PASS / FIX NEEDED
content-repurposing-hub	Turn ONE source (blog post / podcast / video / thread / idea) into a multi-platform content pack — per-platform NATIVE angles (X = thread, LinkedIn = story/lesson, newsletter = issue, carousel = slides, blog = SEO article, Shorts = clip — from templates/platform-angles.md), no copy-paste reposts (anti-repost hook-overlap check enforced by repurpose-writer.mjs, exit 1 on duplicates), a staggered cross-post calendar, one CTA per piece — audit harness: audit-repurpose.mjs (source, platform count, native angles, no duplicates, CTAs, calendar → repurpose-audit.md) + a fresh repurpose-auditor subagent signing PASS / FIX NEEDED. Delegates production to the producer skills
sponsorship-pipeline	Turn a creator's audience into brand-deal revenue — media kit (audience stats + engagement + the offer), rate card computed from niche CPM/RPM benchmarks by rate-card.mjs (honest ranges, no guarantees), personalized outreach + negotiation scripts (FTC disclosure built in, zero placeholders survive), and a deal-tracking sheet with the day 3/7/14 follow-up cadence — audit harness: audit-sponsor.mjs (kit, rate card honesty, outreach structure + disclosure + no placeholders, tracking → sponsor-audit.md) + a fresh sponsor-auditor subagent signing PASS / FIX NEEDED
newsletter-growth	Grow an editorial newsletter (Beehiiv / Substack / ConvertKit) with story-first issues people actually open and forward — subject-line packs from proven formulas (≤ 60 chars, one formula), a story-spine issue writer (open loop → rising tension → payoff → ONE CTA + growth plug, enforced by issue-writer.mjs), welcome-sequence + referral-loop + cross-promo growth tactics, and an analytics review loop — audit harness: audit-newsletter.mjs (subject, story spine, anti-fluff, CTA/growth, word window, growth plan → newsletter-audit.md) + a fresh newsletter-auditor subagent signing PASS / FIX NEEDED
podcast-to-shorts	Turn long-form podcast/video content into viral-ready vertical clips — clip-finder.mjs scores every transcript moment on a virality scorecard (hook/emotion/controversy/quotability), picks the standalone winners, and generates the exact FFmpeg 9:16 cut commands — plus a hook + caption per clip — audit harness: audit-clips.mjs (transcript, scored moments, cut commands, clips output, captions → clips-audit.md) + a fresh clips-auditor subagent signing PASS / FIX NEEDED
youtube-video-pipeline	Plan long-form YouTube videos end-to-end: researched video-brief.md → retention-engineered script (hook in 30s, open loops) → 10-variant title pack scored on CTR formulas + 60-char limit (title-pack.mjs) → thumbnail brief + image-gen prompt → description + chapters + tags — audit harness: audit-video-plan.mjs (brief, script hook, title pack, thumbnail, metadata → video-plan-audit.md) + a fresh video-plan-auditor subagent signing PASS / FIX NEEDED
blog-seo-content	SEO articles that rank and get cited by AI search (GEO) — keyword-outline.mjs builds the keyword cluster + intent scores + outline scaffold → approved seo-brief.md → E-E-A-T-rich article (named author, cited stats, quotable blocks) + meta title/description pack — audit harness: audit-blog.mjs (brief, headings, EEAT, citations, quotable blocks, anti-fluff, meta → blog-audit.md) + a fresh blog-auditor subagent signing PASS / FIX NEEDED
linkedin-personal-brand	Founder/creator LinkedIn presence that compounds — capture the user's real voice first (voice-capture.md), then headline + About rewrite, a weekly post calendar (story/teaching/contrarian/win/question roles, one CTA each), and a comment + connection strategy — audit harness: audit-brand.mjs (voice, bio, calendar, engagement, blocklists → brand-audit.md) + a fresh brand-auditor subagent signing PASS / FIX NEEDED
x-growth	Plan, write and grow an X/Twitter presence FAST — a scripted growth content plan (goal + KPIs + 3+ pillars + daily cadence + engagement schedule + day-7 review) with a week of drafted posts in the top-creator format (500–800 char micro-essays: hook line ≤ 100 chars → story → bullet points → payoff → CTA, hook/value/story/proof/cta/loop mix, enforced by post-writer.mjs with anti-fluff/bait blocklists) — and the ZERO-HASHTAG copy-first rail: a single # fails the plan (X growth is copy-first, tags are noise). Story-spine + addiction rails per post, the weekly series loop (hook → value → proof → CTA converts viewers into followers), and the growth rituals (reply-first hour, quote posts, DM follow-ups) — audit harness: audit-x.mjs (posts, 500–800 window, hook lines, format, zero-hashtag gate, story mix, pillars, plan sections → x-audit.md) + a fresh x-auditor subagent signing PASS / FIX NEEDED
social-media-content-plan	30-day algorithm-reset content plan across user-selected platforms (Instagram · X · LinkedIn · TikTok · YouTube (Shorts + long-form) · Facebook · Threads) — honest per-platform strategy (how each algorithm actually ranks in 2026, the 14-day re-training sprint, native formats, cadence, hashtag/keyword rules — grounded by platform-playbook.mjs), 3–4 content pillars + hook bank, a deterministic day-by-day calendar (build-calendar.mjs from plan.json), a first-60-minute engagement protocol, and a metrics/review loop (day-7/14/21/30) — audit harness: audit-content-plan.mjs (platform coverage, reset framing, pillars, calendar, engagement, metrics, fluff blocklist → content-plan-audit.md) + a fresh content-plan-auditor subagent signing PASS / FIX NEEDED
skill-builder (meta)	Scaffold new skills the deepak-skill way — scaffold-skill.mjs generates the folder, SKILL.md contract (quality bar + workflow + checklist), Deepak-branded scripts, templates and examples, then wires the skill into README/USAGE/prompt-examples — audit harness: audit-scaffold.mjs --docs checks the new skill against the repo contract (SKILL.md, scripts, templates/examples, docs wiring + counts → scaffold-audit.md) + a fresh scaffold-auditor subagent signing PASS / FIX NEEDED
mcp-agent-builder	Build MCP servers + the AI agents that use them end-to-end in any IDE/CLI — a discovery interview (connector, users, action surface, auth, deployment model) → mcp-prd.md (problem, users, tools/resources/prompts, auth, non-goals, metrics) → mcp-architecture.md (ONE deployment model: stdio / remote HTTP / MCPB — transport + auth flow must match) → runnable server (scaffold-server.mjs generates a real @modelcontextprotocol/sdk server from mcp-plan.json, one handler per tool, exit 1 on bad plans) → agent-design.md (system prompt + tool wiring + guardrails) → ide-cli-matrix.md (connect in Claude Code, Cursor, Codex CLI, Gemini CLI, OpenCode, Cline, Windsurf, VS Code, Zed) — audit harness: audit-mcp.mjs (PRD, architecture, scaffold, agent wiring, IDE matrix, secrets → mcp-audit.md) + a fresh mcp-auditor subagent signing PASS / FIX NEEDED
branding-studio	Turn any idea into an expert-grade brand — researched naming (3 names + winner + availability signals), validation scorecard (/35 → BUILD/ITERATE/PIVOT with kill criteria + naming viability), and a complete brand book (brand-book.md: one-liner ≤ 15w + messaging hierarchy + proof bank + taglines + visual direction — palette/hex + type + mood/brand board + logo brief — + voice guide in the creator's taste). The research searches the best sources (competitors, naming pools, visual trends, voice benchmarks) and the brand validates BEFORE it's built — one brand everywhere via brand-map.md. Audit harness: audit-branding.mjs (name pack + one-liner + hierarchy + proof + taglines + visual + voice + memory/taste + fluff → branding-audit.md) + a fresh branding-auditor subagent scoring /50 (≥ 35 = worth shipping) signing PASS / FIX NEEDED. Built-in example: "TIDE — AI journal for founders"
positioning-studio	Turn any product or brand into ONE positioning that makes every content piece convert — ONE one-liner (≤ 15 words), a messaging hierarchy (audience → problem → promise → proof), 3 taglines, a proof-point bank, and a voice guide written in the creator's own taste (memory + taste rails) — anti-hype: proof beats adjectives; no proof → shrink the promise. Optional message-map.md rephrases the SAME message per channel (landing page, email, social). Audit harness: audit-positioning.mjs (one-liner cap, hierarchy completeness, proof ≥ 3, taglines, fluff + taste banned-words → positioning-audit.md) + a fresh positioning-auditor subagent signing PASS / FIX NEEDED
prompt-engineering	Build a structured, versioned prompt library in the creator's voice — recurring use cases (content / research / code / email) each with the role → context → task → format → constraints framework, written through the taste profile (voice rules + banned words) so outputs sound like the author — plus a test/iterate loop (test-results.md: verdict per prompt, weak prompts rewritten) so only prompts that work ship. Audit harness: audit-prompts.mjs (framework sections on every prompt, no placeholder gaps, voice rules carry taste banned-words, test verdicts per prompt → prompts-audit.md) + a fresh prompt-auditor subagent signing PASS / FIX NEEDED
ai-automation	Design AI agent workflows BEFORE building them — an honest automation-worthiness gate first (automatable? worth it? says NO when it isn't), then automation-design.md: trigger → per-step tool/agent with input/output contracts → human-in-the-loop checkpoints on irreversible actions (send / publish / delete / charge / deploy) → error handling → cost estimate → build handoff that delegates to mcp-agent-builder / vibe-code-webapp / prompt-engineering. Audit harness: audit-automation.mjs (worthiness verdict, trigger, step contracts, checkpoints, cost → automation-audit.md) + a fresh automation-auditor subagent signing PASS / FIX NEEDED
veo-cinematic-reels	Reel scripts + rich long-form copy-paste video prompts (~150–250 words) that work in ANY generator — Google Flow / Veo 3.1, Kling, Luma, Runway, Hailuo, Vidu, Pika, PixVerse — manual copy-paste workflow with character consistency locked (reference-image ingredients + a verbatim character block in EVERY prompt, self-verified by scene-prompts.mjs), IMAX-level cinematic language with per-scene lens/tempo/lighting detail, locked color grading (film-stock tokens), native dialogue/SFX audio, a labeled negative prompt + locked seed per scene, per-tool upload instructions (Ingredients / Elements / image-to-video) — and an audit harness: audit-reels.mjs (automated per-scene token/verify/length/negative/seed/bridge checks → reels-audit.md) + a fresh reel-auditor subagent scoring reel-worthiness (/50) and signing PASS / FIX NEEDED
serial-story-reels	EPISODIC story video series for Google Flow / Veo 3.1 — comic, love story, action, thriller or fantasy (or the user's own characters via image uploads): a story bible (season arc + per-episode hooks + cliffhangers, validated by series-arc.mjs) → character sheets (verbatim blocks + reference-image prompts) → a self-verified copy-paste prompt pack for EVERY scene of EVERY episode (episode-prompts.mjs, exits 1 on drift) so the SAME characters stay consistent ACROSS episodes — plus cinematic action/motion language, native Veo dialogue/SFX, and a voiceover line sheet for a Kokoro post pass
photoshoot-studio	AI photoshoot prompt packs for people AND products (Google Flow / Nano Banana Pro, Midjourney, Flux) — manual copy-paste workflow with subject consistency locked (reference-image ingredients + a verbatim person/product block in EVERY prompt, self-verified by shot-prompts.mjs), professional photography language (camera body, lens, f-stop, lighting setup, film stock), locked grade per shoot, per-platform aspect ratios, short edit/inpaint prompts for re-posing / re-outfitting / re-lighting — and an audit harness: audit-shoot.mjs (per-shot token/verify/aspect/edit checks → shoot-audit.md) + a fresh shoot-auditor subagent scoring shoot-worthiness (/50) and signing PASS / FIX NEEDED
paid-ads-studio	The paid engine — end-to-end ad campaign production for Meta + Google: copy-paste Veo 3.1 video ad prompts + image ad prompts (Nano Banana Pro / Midjourney / Flux) with locked brand consistency (verbatim product block in EVERY prompt, self-verified by ad-prompts.mjs), hook-first ad copy inside every platform's char limits (anti-fluff enforced by ad-copy.mjs), campaign blueprints (Meta ODAX + Advantage+ audiences · Google Demand Gen + PMax + Search, audience signals ≠ targeting), 2026 AI-content compliance checklists, a cost-management plan (ramp / learning / kill / scale), a benchmark-driven forecast of expected results — impressions, clicks, conversions, CPA, ROAS across conservative/base/aggressive scenarios (forecast-ads.mjs) — and an audit harness: audit-ads.mjs (automated hook/char/consistency/compliance checks → ad-audit.md) + a fresh ads-auditor subagent scoring hook worthiness (/50) and signing PASS / FIX NEEDED before anything ships. Built-in example: "Brew & Co Tumbler — Paid Launch" (full launch-ready pack).
vibe-code-webapp	Build or extend production-ready vibe-coded web apps at 20-year-expert depth — cross-project taste loaded first (creator-portfolio.md: stack/design defaults + past builds, read at Stage 0, written back at Stage 7) → detailed idea interview (+ existing-project scan scan-project.mjs → project-scan.md) → expert research (research-playbook.md: signals + competitor teardown + TAM/SAM/SOM + channels + pricing + positioning) → SaaS validator (saas-validator.md + saas-score.mjs → validation.md: /35 verdict, kill criteria, unit economics, kill guardrail, validation moves) → Build Pack assembled from ONE validated JSON (pack-builder.mjs: pack-plan.json → PRD.md + stack-blueprint.md — with a design source of truth: Figma via the Figma Developer MCP, Google Stitch DESIGN.md, or the open-source pack (frontend-design.md), plus AI-feature rails when the PRD has AI (ai-logic.md) — + sitemap.md + TODO.md P0/P1/P2 via todo.mjs; --check-only exits 1 on any placeholder, --batch does N apps in one run) → you approve the pack AND the todo list → builds step by step from vibe-coder-instructions.md (BUILD.md, golden loop + design parity) filing detailed build-report.md reports, with a deterministic stage tracker (progress.mjs → output/progress.md, 0 Onboard → 7 Deliver) → production audit (--payments/--ai flags) → everything-auditor final review → scripted delivery (deploy-setup.mjs generates real host config + CI + filled deploy-runbook.md; package-deliverable.mjs → HANDOFF.md + manifest.json + one ZIP) → daily MEMORY.md keeps you and any AI in sync

The skills are a complete production pipeline: what's trending (research + brainstorm) → what to say (hooks) → how it looks (motion / assets / thumbnails) → how it sounds (voice/SFX) → approved & audited (video-product-pipeline) → multiplied (podcast→shorts, YouTube packaging, SEO blog, LinkedIn, X threads, editorial newsletters) → paid (Meta + Google ad campaigns with forecast-first economics via paid-ads-studio).

Every skill now ships the same audit-harness pattern: an audit-*.mjs script runs the automated checks on the delivered pack (consistency tokens, verify markers, char limits, blocklists, output files → *-audit.md, exit 0/1/2), then a FRESH <skill>-auditor subagent completes a /50 worthiness scorecard and signs PASS / FIX NEEDED — nothing ships until PASS. All 31 skills (29 Deepak + data-analysis + web-scraping) now have a harness: text-motion-reels, hook-storyboard-retention, voice-sfx-audio, video-asset-reels, carousel-post-images, veo-cinematic-reels, photoshoot-studio, podcast-to-shorts, youtube-video-pipeline, blog-seo-content, linkedin-personal-brand, social-media-content-plan, skill-builder, serial-story-reels, paid-ads-studio, vibe-code-webapp, video-product-pipeline, email-marketing (audit-email.mjs wraps validate-email.mjs), x-growth, newsletter-growth, mcp-agent-builder, ebook-builder, thumbnail-studio, content-repurposing-hub, sponsorship-pipeline, positioning-studio, branding-studio, prompt-engineering, ai-automation, data-analysis (audit-data.mjs), web-scraping (audit-scrape.mjs).

---

Quick start — install in your project (3 steps)

Want the DEEPAK banner in your terminal? Clone the repo and run ./install.sh — the branded installer prints the DEEPAK ASCII banner, lists all 29 skills, and installs them (Windows: use Git Bash). Full guide: install.md.

Using the claude.com website (browser, no terminal)? No install needed — copy-paste the skill instructions and Claude follows them in the chat. Step-by-step: CLAUDE-COM.md.

1. Prerequisites

• Node.js 18+ (installs npx automatically)
• Google Chrome — used to render frames
• FFmpeg — assembles the final MP4 (only needed for video output)

2. Install the skills

Run this inside your project folder:

npx skills add Deepak-ai-93/deepak-skill --all

That's it. It installs all 29 skills into .agents/skills/ — no heavy media, no demo files. Or run the branded installer (install.sh) for the DEEPAK terminal banner.

3. Use it with any CLI

Just ask your agent (Claude Code, Cursor, Codex, Gemini CLI, etc.):

"Using the text-motion skills, make a 20-second psychology-facts reel with a strong hook, beat-synced typography, and a Kokoro voiceover."

The agent will run the format wizard, build the composition, and render your 4K reel.

---

Install options

What you want	Command
All skills into the current project	npx skills add Deepak-ai-93/deepak-skill --all
A single skill	npx skills add Deepak-ai-93/deepak-skill --skill text-motion-reels
video-asset-reels only (self-contained — SKILL.md + scripts/, works standalone in any project)	npx skills add Deepak-ai-93/deepak-skill --skill video-asset-reels
video-product-pipeline only (self-contained — SKILL.md + scripts/, works standalone in any project)	npx skills add Deepak-ai-93/deepak-skill --skill video-product-pipeline
carousel-post-images only (self-contained — SKILL.md + scripts/ + examples/, works standalone in any project)	npx skills add Deepak-ai-93/deepak-skill --skill carousel-post-images
ebook-builder only (self-contained — SKILL.md + scripts/ + templates/ + examples/, works standalone in any project)	npx skills add Deepak-ai-93/deepak-skill --skill ebook-builder
vibe-code-webapp only (self-contained — SKILL.md + scripts/ + templates/, works standalone in any project)	npx skills add Deepak-ai-93/deepak-skill --skill vibe-code-webapp
email-marketing only (self-contained — SKILL.md + scripts/ + templates/ + examples/, works standalone in any project)	npx skills add Deepak-ai-93/deepak-skill --skill email-marketing
thumbnail-studio only (self-contained — SKILL.md + scripts/ + templates/ + examples/, works standalone in any project)	npx skills add Deepak-ai-93/deepak-skill --skill thumbnail-studio
content-repurposing-hub only (self-contained — SKILL.md + scripts/ + templates/ + examples/, works standalone in any project)	npx skills add Deepak-ai-93/deepak-skill --skill content-repurposing-hub
sponsorship-pipeline only (self-contained — SKILL.md + scripts/ + templates/ + examples/, works standalone in any project)	npx skills add Deepak-ai-93/deepak-skill --skill sponsorship-pipeline
podcast-to-shorts only (self-contained — SKILL.md + scripts/ + templates/ + examples/, works standalone in any project)	npx skills add Deepak-ai-93/deepak-skill --skill podcast-to-shorts
youtube-video-pipeline only (self-contained — SKILL.md + scripts/ + examples/, works standalone in any project)	npx skills add Deepak-ai-93/deepak-skill --skill youtube-video-pipeline
blog-seo-content only (self-contained — SKILL.md + scripts/ + templates/ + examples/, works standalone in any project)	npx skills add Deepak-ai-93/deepak-skill --skill blog-seo-content
linkedin-personal-brand only (self-contained — SKILL.md + templates/ + examples/, works standalone in any project)	npx skills add Deepak-ai-93/deepak-skill --skill linkedin-personal-brand
x-growth only (self-contained — SKILL.md + scripts/ + templates/ + examples/, works standalone in any project)	npx skills add Deepak-ai-93/deepak-skill --skill x-growth
newsletter-growth only (self-contained — SKILL.md + scripts/ + templates/ + examples/, works standalone in any project)	npx skills add Deepak-ai-93/deepak-skill --skill newsletter-growth
social-media-content-plan only (self-contained — SKILL.md + scripts/ + templates/ + examples/, works standalone in any project)	npx skills add Deepak-ai-93/deepak-skill --skill social-media-content-plan
skill-builder only (self-contained — SKILL.md + scripts/, works standalone in any project)	npx skills add Deepak-ai-93/deepak-skill --skill skill-builder
mcp-agent-builder only (self-contained — SKILL.md + scripts/ + templates/ + examples/, works standalone in any project)	npx skills add Deepak-ai-93/deepak-skill --skill mcp-agent-builder
veo-cinematic-reels only (self-contained — SKILL.md + scripts/ + templates/ + examples/, works standalone in any project)	npx skills add Deepak-ai-93/deepak-skill --skill veo-cinematic-reels
serial-story-reels only (self-contained — SKILL.md + scripts/ + templates/ + examples/, works standalone in any project)	npx skills add Deepak-ai-93/deepak-skill --skill serial-story-reels
photoshoot-studio only (self-contained — SKILL.md + scripts/ + templates/ + examples/, works standalone in any project)	npx skills add Deepak-ai-93/deepak-skill --skill photoshoot-studio
paid-ads-studio only (self-contained — SKILL.md + scripts/ + templates/ + examples/, works standalone in any project)	npx skills add Deepak-ai-93/deepak-skill --skill paid-ads-studio
positioning-studio only (self-contained — SKILL.md + scripts/ + templates/ + examples/, works standalone in any project)	npx skills add Deepak-ai-93/deepak-skill --skill positioning-studio
branding-studio only (self-contained — SKILL.md + scripts/ + templates/ + examples/, works standalone in any project)	npx skills add Deepak-ai-93/deepak-skill --skill branding-studio
prompt-engineering only (self-contained — SKILL.md + scripts/ + templates/ + examples/, works standalone in any project)	npx skills add Deepak-ai-93/deepak-skill --skill prompt-engineering
ai-automation only (self-contained — SKILL.md + scripts/ + templates/ + examples/, works standalone in any project)	npx skills add Deepak-ai-93/deepak-skill --skill ai-automation
Install globally (any project, any machine)	npx skills add Deepak-ai-93/deepak-skill --all -g
Target specific agents	npx skills add Deepak-ai-93/deepak-skill --all -a claude-code -a cursor
Install from a local clone	npx skills add ./deepak-skill --all

Verify the install:

ls .agents/skills/
# blog-seo-content  carousel-post-images  content-repurposing-hub  ebook-builder  email-marketing  hook-storyboard-retention  newsletter-growth
# linkedin-personal-brand  mcp-agent-builder  paid-ads-studio  photoshoot-studio  podcast-to-shorts  serial-story-reels  skill-builder
# social-media-content-plan  sponsorship-pipeline  text-motion-reels  thumbnail-studio  veo-cinematic-reels  video-asset-reels  video-product-pipeline
# vibe-code-webapp  voice-sfx-audio  x-growth  youtube-video-pipeline  ai-automation  branding-studio  positioning-studio  prompt-engineering

# or check versions
npx skills list

Upgrade — already installed? Get the latest (including branding-studio)

Already have deepak-skill in your project? Re-running install upgrades in place (idempotent — it overwrites with the latest, including every new command + template from branding-studio and all 29 skills). No need to uninstall first.

# Inside your project — upgrade everything to the latest (recommended)
npx skills add Deepak-ai-93/deepak-skill --all

# Or upgrade/add just the new branding skill to an existing project
npx skills add Deepak-ai-93/deepak-skill --skill branding-studio

# Or bulk-update all skills from the registry
npx skills update

# Same via the branded installer (re-running upgrades)
./install.sh              # upgrades all 29 skills
./install.sh --skill branding-studio   # add/upgrade just branding-studio

# Verify you have the latest
ls .agents/skills/branding-studio/   # should exist
npx skills list | grep branding-studio

Tip: npx skills add is safe to re-run — it always pulls the latest SKILL.md + scripts + templates. Run it after every git pull or when you see a new skill announced.

Remove: npx skills remove <skill-name>

See the skills.sh docs for more options (--list, --copy, packs, etc.).

---

Usage

Format wizard + 4K rendering (text-motion-reels)

The text-motion-reels skill starts with an interactive format wizard — the agent asks you to pick one of 11 trending text-only formats before building anything:

1. Word Pop — Hormozi-style high-impact captions (business, hot takes)
2. Highlighter — Vox-style analytical explainer (psychology, facts)
3. 3D Editorial — luxury quiet minimalist (mindset, stoic)
4. Card Listicle — numbered save-bait grid (finance, productivity)
5. Chat Thriller — text-message storytime (reddit, drama)
6. SVG Ambient — animated SVG backgrounds (mindset, aesthetic, brand)
7. Micro-Fiction — word-by-word story reveal, cliffhanger endings (horror, sci-fi, drama)
8. Quiz Trap — countdown riddle / "only 2% can…" (trivia, psychology, watch-twice)
9. Day Counter — "Day 1 vs Day 90" metric timeline (finance, fitness, growth)
10. Notification Drop — simulated phone notification skits (humor, relatable)
11. Thread Court — Reddit-style post + verdict reveal (AITA, drama, comment bait)

After you pick a format, the agent builds the composition to that format's spec, generates a beat-synced voiceover (Kokoro), renders it at vertical 4K (2160x3840) into a proper folder, and writes a ready-to-post caption.md:

node render/render-frames.mjs --html word-pop_reel.html --name word-pop_money-rules_4k --duration 15 --fps 30 --scale 2 --audio assets/full_mix.m4a

Output → one folder per video:

output/word-pop_money-rules_4k/
├── word-pop_money-rules_4k.mp4   # 2160x3840, H.264, -14 LUFS mix
├── frames/                        # source frames
└── caption.md                     # YouTube Shorts, Instagram, X, LinkedIn, TikTok, Threads captions

Sample prompts

Goal	Prompt
Hook + storyboard only	"Write a scroll-stopping hook and 15-second storyboard for a finance reel."
Full text-only reel	"Make a 20-second psychology reel using the Word Pop format, hook: 'Your brain does this every time you scroll.'"
Add voiceover	"Generate a Kokoro voiceover for this reel and mix it with a CC0 ambient track."

Ready-to-paste prompts for SaaS, AI, web-dev life, finance, mindset & 10+ more niches → prompt-examples.md — every prompt runs trend-research → spec → approval → generate → audit.

Prompt → spec → approve → generate → audit (video-product-pipeline)

The premium default for every video request — engineered to reach millions of views, and a sloppy prompt can't silently produce a sloppy video:

1. Trend research + brainstorm — node render/trend-hunt.mjs --niche "{topic}" --subreddits "{r1},{r2}" --geo US (Reddit top-of-day + Google Trends, no API key) + web research → ≥5 angles scored on the viral scorecard → winner locked in trend-brief.md
2. Analyze the prompt (asks ≤3 clarifying questions if vague — never guesses silently)
3. Write video-product.md — full spec: style, hook, beat-by-beat text + timings, safe-zone map, voice, audio, output name
4. Stop and wait for your approval — nothing is generated before you say "approve" (or "edit")
5. Generate by delegating to text-motion-reels / video-asset-reels / voice-sfx-audio
6. Audit — node render/audit-composition.mjs --html reel.html captures a keyframe per beat and auto-checks safe zones (x 8–92%, y 15–85%), text overlap, word caps & determinism, then a dedicated auditor subagent reviews spelling, style and readability and signs audit-report.md PASS before delivery
7. Deliver — MP4 + caption.md + video-product.md + trend-brief.md + audit-report.md

"Make a 20-second psychology reel — hook: 'Your brain does this every time you scroll.'" → spec file first → approve → generate → audit → deliver

Carousel posts (carousel-post-images)

Scroll-stopping LinkedIn/Instagram carousels as image sets at 4K — one script, two CLI modes on a shared slides.html deck. Every deck clears a fixed quality bar: 4K resolution (≥ 4000px long edge), photorealistic real-life scenario visuals, and viral anti-fluff copywriting (hook formulas + a fluff blocklist — no "unlock", no "game-changer"). The agent always presents the two-mode choice before generating:

• Mode 1 — Browser render (any CLI): node render/render-carousel.mjs --html slides.html --out carousel/ --4k → pixel-perfect 4320×5400 (4:5) / 4320×4320 (1:1) PNGs, text never garbled
• Mode 2 — Image-model generation (photoreal): node render/render-carousel.mjs --html slides.html --out carousel/ --mode model → exports per-slide 4K prompts (prompts.md) with the deck's exact copy → dispatch to the CLI's native image tool (Codex image_gen / Grok /imagine / Antigravity artifacts) at 4K; text is verified — LLM models garble on-image text, so any wrong character gets regenerated or falls back to Mode 1

Then: audit (text accuracy, copy punch, scene authenticity, contrast, style consistency, 4K size) → caption.md (500–900 chars per platform, no hashtags, hook first, one CTA).

Built-in examples — "Day in the life of an AI developer" (examples/day-in-the-life-dev/, browser-mode deck) and "3 Money Rules Nobody Told You" (examples/real-life-money/, image-model mode: hand-written per-slide 4K prompts + fallback deck) — both with full caption packs.

"Make a carousel: 3 money rules nobody told you. Real scenes, image-model mode, 8 slides, 4K, LinkedIn." → choose mode → prompts.md → 4K images → caption.md → auditor check → deliver

Ebook lead magnets (ebook-builder)

Build designed ebooks (guides / checklists / frameworks) as print-ready A4 PDF + 4K page PNGs + cover.png — it starts from author memory + taste (ebook-memory.md: identity, voice fingerprint, design defaults, past builds — read at Stage 0, written back at Stage 8, so every book is written in the author's voice with zero banned words), then runs the design picker: pick ONE of six layouts (Editorial Classic · Modern Bold · Minimal Luxury · Playful Pop · Technical Dark · Nature Calm) + ONE of 30 palettes + accent/type pairing + ONE of 6 cover styles + ONE motif family + mood/texture, then generate in two modes like the carousel:

cd skills/ebook-builder
node scripts/render-ebook.mjs --html ebook.html --out ebook/ --pdf         # Mode 1: A4 PDF + cover.png + page PNGs (headless Chrome)
node scripts/render-ebook.mjs --html ebook.html --out ebook/ --mode model # Mode 2: photoreal cover + interior scene prompts (prompts.md)
node scripts/audit-ebook.mjs --pack <ebook-folder> --out ebook-audit.md   # automated audit → spawn ebook-auditor subagent

Deliverables: ebook.html (single source of truth, one .page per book page, data-layout + data-palette + data-motif on every page, data-cover-style on the cover) → ebook.pdf + cover.png + pages/*.png (Mode 1, 2× = 2160×3056) or prompts.md (Mode 2 — design-token prompts dispatched to Nano Banana / Midjourney / Flux at 4K) → ebook-audit.md. Story-first structure enforced (cover opens a loop → chapters escalate → payoff → CTA) + author memory (templates/memory-profile.md — the durable ebook-memory.md every build reads + writes) + taste profile (templates/taste-profile.md — voice capture card) + the design picker card (templates/design-picker.md) + cover-image brief (templates/cover-brief.md — one idea, title ≤ 6 words, thumbnail-readable). Built-in examples: "The 5-Page Pricing Fix" (examples/pricing-guide/, Modern Bold · Electric Blue) and "The Shipping Playbook" (examples/dev-shipping-playbook/, Technical Dark · Terminal Green) — decks + generated prompts, each with its author's ebook-memory.md.

"Build me a lead-magnet ebook: the 5-Page SaaS Pricing Fix — Modern Bold design, ~7 pages, killer cover, A4 PDF." → memory + taste (author voice loaded) → design picker (layout + palette + cover style + motif) → ebook.html → render (PDF + cover + pages) → auditor check → deliver

YouTube thumbnails (thumbnail-studio)

The visual CTR engine for youtube-video-pipeline — a CTR teardown of the niche first, then 3–5 thumbnail variants with photoreal image-gen prompts + an A/B plan:

cd skills/thumbnail-studio
node scripts/thumbnail-prompts.mjs --plan thumbnail-plan.json --out thumbnails.md   # builds + self-verifies (overlay ≤ 5 words, anti-cliché, exit 1 on drift)
node scripts/audit-thumbs.mjs --pack <thumbnail-folder> --out thumbs-audit.md       # automated audit → spawn thumbs-auditor subagent

Deliverables: thumbnail-teardown.md (niche patterns + the ONE idea) → thumbnails.md (3–5 variants, each = ONE idea + ≤ 5-word overlay + 1280×720 photoreal prompt for Nano Banana / Midjourney / Flux) → ab-test.md (2 variants, identical title, window, niche CTR benchmarks) → thumbs-audit.md. Built-in example: "3 Pricing Mistakes" (examples/pricing-video/).

"Make 3 thumbnails for my SaaS pricing video — teardown my niche first, give me prompts + an A/B plan." → CTR teardown → 3–5 variants (thumbnail-prompts.mjs) → ab-test.md → auditor check → deliver

Repurposing hub (content-repurposing-hub)

Turn ONE source into platform-native pieces — the orchestrator that hands each piece to its producer skill:

cd skills/content-repurposing-hub
node scripts/repurpose-writer.mjs --plan repurpose-plan.json --out hub-plan.md   # builds + validates (≥ 2 platforms, no near-duplicate hooks, exit 1 on bad plans)
node scripts/audit-repurpose.mjs --pack <hub-folder> --out repurpose-audit.md    # automated audit → spawn repurpose-auditor subagent

Deliverables: hub-plan.md (angle bank + per-platform cards: format, native angle, unique hook, producer skill, one CTA) + calendar.md (staggered cross-post order — no same-day self-competing) + repurpose-audit.md. The anti-repost rule is script-enforced: two pieces opening with the same line fail the build. Built-in example: "The 5-Page Pricing Fix hub" (examples/pricing-guide-hub/, X + LinkedIn + newsletter + carousel).

"Repurpose my pricing-guide blog post — X thread, LinkedIn post, newsletter issue and a carousel, all natively, no copy-paste." → angle bank → hub-plan.md (unique hooks per platform) → calendar.md → auditor check → deliver to each producer skill

Sponsorship pipeline (sponsorship-pipeline)

The monetization layer — turn the audience into brand deals:

cd skills/sponsorship-pipeline
node scripts/rate-card.mjs --niche saas --followers 48000 --platform youtube --views 210000   # → rate-card.md from niche CPM/RPM benchmarks
node scripts/audit-sponsor.mjs --pack <sponsor-folder> --out sponsor-audit.md                   # automated audit → spawn sponsor-auditor subagent

Deliverables: media-kit.md (audience + engagement + the offer — the sponsor's problem first) → rate-card.md (benchmark-grounded ranges, no guarantees) → outreach.md (≥ 2 personalized pitches with problem/proof/offer/ask + negotiation scripts + FTC disclosure) → tracking.md (pipeline + day 3/7/14 follow-ups) → sponsor-audit.md. Built-in example: "Alex Rivera — SaaS creator deals" (examples/saas-creator-deals/).

"Build my media kit + rate card + outreach for brand deals. Niche: SaaS, 48k YouTube, 12k newsletter." → interview → rate-card.mjs (benchmarks) → media kit → 3 personalized pitches → tracking → auditor check → deliver

Emails (email-marketing)

High-converting, spam-free HTML emails for Brevo / MailerLite / Mailchimp / Klaviyo and any ESP — one folder per campaign with five files: a high-CTR subject line pack (subject-lines.md — 3–5 variants from proven formulas, winner marked, A/B plan), a bulletproof responsive email.html (tables + inline CSS, dark-mode aware, preheader, compliance footer), a plain.txt fallback, a validation-report.md from validate-email.mjs (spam-trigger scan, ALL-CAPS/!!! check, unsubscribe + physical-address compliance, alt text, text:image ratio, E-E-A-T signals — real author, proof, social proof, human reply-to), and preview/ screenshots (desktop 600px + mobile 320px) for the visual audit.

Built-in example — "Meet Replays" product launch (examples/product-launch/): full email + subject pack + copy breakdown (why every line is there) + plain text + the ESP merge-tag swap sheet.

"Write a product-launch email for our SaaS — audience: existing customers, goal: demo clicks. Make it spam-free with high-CTR subject lines and strong E-E-A-T." → subject line pack → email.html + plain.txt → validate-email.mjs report → preview screenshots → auditor sign-off → deliver

Editorial newsletters (newsletter-growth)

Grow an editorial newsletter (Beehiiv / Substack / ConvertKit) with story-first issues — subject-line pack (≤ 60 chars, one proven formula) → story-spine issue (open loop → stakes → payoff → value receipts → ONE CTA + ONE growth plug):

cd skills/newsletter-growth
node scripts/issue-writer.mjs --plan issue-plan.json --out issue.md          # assembles + validates (exit 1 on bad plans)
node scripts/audit-newsletter.mjs --pack <newsletter-folder> --out newsletter-audit.md   # automated audit → spawn newsletter-auditor subagent

Deliverables: issue.md (subject + story-first body, word window 350–900, no fluff) + growth-plan.md (welcome sequence, referral loop, cross-promos, analytics review) + newsletter-audit.md. Distinct from email-marketing — that's transactional ESP campaigns; this is editorial + growth. Built-in example: "Founder Welcome" (examples/founder-welcome/).

"Write my newsletter issue this week — story-first. Niche: SaaS pricing for founders. Goal: replies + forwards." → subject formula → story plan → issue-writer.mjs → growth-plan.md → auditor check → deliver

Podcast → Shorts (podcast-to-shorts)

Long-form content → viral-ready vertical clips. The pipeline scores, picks, and cuts:

cd skills/podcast-to-shorts
node scripts/clip-finder.mjs --transcript transcript.txt --clips 5 --out clip-plan.md   # score moments
node scripts/clip-finder.mjs --cuts clip-plan.md --input episode.mp4 --run               # cut 1080x1920 clips

The transcript is auto-scored on hook/emotion/controversy/quotability; only standalone moments above the cutoff become clips. Each clip gets a hook + platform caption → captions.md. Built-in example: examples/transcript-sample.txt + examples/clip-plan.md.

"Turn this 1-hour podcast episode into 5 shorts — clip the most viral moments." → transcript → score → clip-plan → FFmpeg cuts → captions → auditor check → deliver

YouTube pipeline (youtube-video-pipeline)

Plan long-form YouTube videos end-to-end: researched angle → retention-engineered script (hook in the first 30s) → a 10-variant title pack (scored on CTR formulas + the 60-char mobile limit):

cd skills/youtube-video-pipeline
node scripts/title-pack.mjs --topic "saas pricing" --angle "3 pricing mistakes founders make" --out titles.md

Then the agent writes the thumbnail brief (+ image-gen prompt), description with chapters, and tags — all matching the script's promise (no clickbait mismatch). Built-in example: examples/titles.md.

"Plan a YouTube video: 3 SaaS pricing mistakes founders make — script, 10 titles, thumbnail brief, description." → brief → approve → script → title pack → thumbnail → metadata → auditor check → deliver

SEO blog content (blog-seo-content)

Articles that rank on search and get cited by AI search engines:

cd skills/blog-seo-content
node scripts/keyword-outline.mjs --seed "saas onboarding" --out seo-brief.md

Keyword cluster + intent scores + outline scaffold → your approval → E-E-A-T-rich article (named author, every stat cited, quotable "Bottom line" block) + meta title/description/slug pack.

"Write a blog post about saas onboarding that ranks and gets cited by ChatGPT — EEAT-heavy, one keyword, meta pack." → seo-brief → approve → article.md → meta.md → auditor check → deliver

LinkedIn personal brand (linkedin-personal-brand)

Build a founder/creator presence that compounds — your real voice captured first (never template-speak), then a headline + About rewrite, a weekly post calendar (varied roles, one CTA each), and a comment + connection strategy. Ships templates/voice-capture.md, templates/post-formulas.md, and a week-one calendar example.

"Help me build my LinkedIn — capture my voice, rewrite my bio, and give me a week of posts." → voice capture → voice-profile.md → bio.md → calendar.md → engagement.md → auditor check → deliver

X growth — fast content planning + posting (x-growth)

Plan, write and grow an X presence fast — growth goal + KPIs + 3+ pillars + a week of drafted single posts (hook → value → story/proof → CTA/loop mix, zero hashtags — a single # fails the plan):

cd skills/x-growth
node scripts/post-writer.mjs --plan content-plan.json --out content-plan.md   # assembles content-plan.md + posts.md (exit 1 on bad plans)
node scripts/audit-x.mjs --pack <plan-folder> --out x-audit.md                # automated audit → spawn x-auditor subagent

Deliverables: content-plan.md (goal + KPIs + 3+ pillars + cadence table + engagement schedule + day-7 review) + posts.md (the week of drafted top-creator posts — 500–800 chars each, hook line ≤ 100, role-annotated story mix, zero hashtags, no fluff/bait-spam) + x-audit.md. Built-in example: "Founder Growth Plan" (examples/founder-growth-plan/).

"Plan a fast-growth week on X for my solo SaaS — goal: 2k → 4k followers in 30 days, voice: first-person founder. NO hashtags." → goal + niche → content plan → post-writer.mjs → posts.md → auditor check → deliver

Social media content plans (social-media-content-plan)

Reset a stalled algorithm with a platform-native 30-day content plan — you pick the platforms (Instagram · X · LinkedIn · TikTok · YouTube (Shorts + long-form) · Facebook · Threads):

cd skills/social-media-content-plan
node scripts/platform-playbook.mjs --platform instagram,x,linkedin   # grounded 2026 algorithm facts for strategy.md
node scripts/build-calendar.mjs --plan plan.json --out calendar.md   # deterministic 30-day calendar (validated, exit 1 on bad plans)
node scripts/check-skills.mjs --out companion-skills.md              # companion gate: which producer skills are installed + install commands
node scripts/audit-content-plan.mjs --pack <plan-folder> --platforms instagram,x --out content-plan-audit.md   # automated audit → spawn content-plan-auditor subagent

Deliverables: strategy.md (per-platform algorithm strategy + honest 14-day re-training sprint framing — no "hacks", no guaranteed view counts) → pillars.md (3–4 narrow pillars + ≥ 8 hook bank) → calendar.md (day-by-day posts: pillar / format / hook / one CTA / metric-to-watch) → engagement.md (the first-60-minute protocol — the window that decides tier expansion) → metrics.md (dwell / completion / save-share / follower-conversion + day-7/14/21/30 review loop) → companion-skills.md (the companion gate — check-skills.mjs shows which producer skills are installed — text-motion-reels / veo-cinematic-reels / carousel-post-images / linkedin-personal-brand / youtube-video-pipeline / voice-sfx-audio — and installs the missing ones so the calendar is actually executable). Built-in example: "Dan — fitness coach reset" (Instagram + X + LinkedIn + YouTube, full pack in examples/fitness-coach-reset/).

"My Instagram and X are dead — build me a 30-day content plan to reset the algorithm. Niche: fat loss for busy professionals." → Platform Wizard (3 questions) → playbook → strategy → pillars → 30-day calendar → engagement + metrics → auditor check → deliver

Build new skills (skill-builder)

Scaffold the next skill in the repo's conventions — quality-bar SKILL.md, Deepak-branded scripts, templates, examples:

cd skills/skill-builder
node scripts/scaffold-skill.mjs --name my-new-skill --desc "one-liner" --scripts tool-a,tool-b --templates guide --example 1
node scripts/scaffold-skill.mjs --list   # show every skill in skills/

"Add a new skill for {topic} to the repo." → scaffold → fill the placeholders → wire into README/USAGE/prompt-examples → validate → audit

MCP servers + AI agents (mcp-agent-builder)

Build an MCP server and the agent that uses it end to end — the full path from idea to a running server connected in any IDE/CLI. Discovery first (never scaffold blind), then PRD, architecture, scaffold, agent wiring:

cd skills/mcp-agent-builder
node scripts/scaffold-server.mjs --plan mcp-plan.json --out server    # generates a real @modelcontextprotocol/sdk server (exit 1 on bad plans)
node scripts/audit-mcp.mjs --pack <mcp-folder> --out mcp-audit.md    # automated audit → spawn mcp-auditor subagent

Deliverables: mcp-discovery.md (interview + Decisions) → mcp-prd.md (problem/users/tools/auth/non-goals/metrics) → mcp-architecture.md (ONE deployment model — stdio / remote HTTP / MCPB — with matching transport + auth flow) → server/ (runnable scaffold: SDK + zod, a handler per tool in mock mode, README, per-IDE config/ snippets, .env.example — no secrets) → agent-design.md (system prompt + tool wiring + guardrails) → ide-cli-matrix.md (Claude Code claude mcp add, Cursor .cursor/mcp.json, Codex config.toml, Gemini CLI, OpenCode, Cline, Windsurf, VS Code, Zed) → mcp-audit.md. Built-in example: "social-poster" (an Instagram + X posting/analytics MCP server, full pack in examples/social-poster/).

"Build me an MCP server that posts my drafts to Instagram and X — full path from idea to connected server in Claude Code." → discovery → PRD → architecture → scaffold (scaffold-server.mjs) → agent design + IDE matrix → audit (audit-mcp.mjs + mcp-auditor) → deliver

Positioning (positioning-studio)

Turn any product or brand into ONE message every piece of content lifts — one-liner, messaging hierarchy, taglines, proof bank, voice guide — written in the author's taste (memory + taste rails):

cd skills/positioning-studio
node scripts/audit-positioning.mjs --pack <positioning-folder> --out positioning-audit.md   # automated audit → spawn positioning-auditor subagent

Deliverables: positioning.md (one-liner ≤ 15 words → hierarchy Audience/Problem/Promise/Proof → ≥ 3 proof points → ≥ 3 taglines → voice guide with banned words) + optional message-map.md (SAME message rephrased per channel) + positioning-audit.md + positioning-memory.md (author identity + taste + past builds, read at Stage 0 / written at the end). Anti-hype rail: proof beats adjectives — no proof, shrink the promise. Built-in example: "ClipDeck positioning" (examples/saas-tool-positioning/ — async video feedback tool, one-liner + message map + memory).

"Give my product a positioning — one-liner, taglines, and a voice guide that sounds like me." → memory + taste (Stage 0) → positioning card → positioning.md (+ message map) → audit (audit-positioning.mjs + positioning-auditor) → deliver

Prompt libraries (prompt-engineering)

Build a structured, versioned prompt library in your voice — framework prompts (role → context → task → format → constraints) + voice rules + a test loop that records a verdict per prompt:

cd skills/prompt-engineering
node scripts/audit-prompts.mjs --pack <library-folder> --out prompts-audit.md   # automated audit → spawn prompt-auditor subagent

Deliverables: prompt-library.md (## Voice rules from the taste profile + one section per prompt) + test-results.md (tool, test input, output quality, verdict per prompt) + prompts-audit.md + prompt-memory.md. Built-in example: "Creator prompt library" (examples/creator-prompt-library/ — hooks, newsletter intro, competitor teardown, carousel repurpose — each framework-written and tested).

"Build me a prompt library for my content workflow — prompts that make AI sound like me." → memory + taste (Stage 0) → voice rules → framework prompts → test loop → audit (audit-prompts.mjs + prompt-auditor) → deliver

AI workflow design (ai-automation)

Design AI agent workflows before building them — an honest worthiness gate, then trigger → steps with contracts → human checkpoints → cost → build handoff to mcp-agent-builder / vibe-code-webapp / prompt-engineering:

cd skills/ai-automation
node scripts/audit-automation.mjs --pack <design-folder> --out automation-audit.md   # automated audit → spawn automation-auditor subagent

Deliverables: automation-design.md (## Automation-worthiness verdict → ## Trigger → ## Workflow steps with Tool/Input/Output/Human-checkpoint/Error-handling → ## Human checkpoints → ## Cost + risk → ## Build handoff) + automation-audit.md + automation-memory.md (past verdicts — the NOs matter most). Built-in example: "Weekly content pipeline agent" (examples/content-pipeline-agent/ — collect → draft → author review → publish, with checkpoints and a per-run cost estimate).

"Design an AI workflow that turns my week's posts into a newsletter — and tell me honestly if it's worth automating." → memory (Stage 0) → worthiness gate → workflow design → cost + handoff → audit (audit-automation.mjs + automation-auditor) → deliver

Google Flow / Veo reels (veo-cinematic-reels)

Reel scripts + rich, long-form copy-paste video prompts (~150–250 words) that work in any video generator — Google Flow / Veo 3.1, Kling, Luma, Runway, Hailuo, Vidu, Pika, PixVerse — one prompt per scene, engineered so the character never changes between scenes (reference-image ingredients uploaded once per tool + a verbatim character block in EVERY prompt, self-verified word-by-word):

cd skills/veo-cinematic-reels
node scripts/scene-prompts.mjs --plan scene-plan.json --out prompts.md   # builds + verifies the prompt pack

Deliverables: character-sheet.md (verbatim character/world/grade blocks + 2–3 reference-image prompts for Nano Banana Pro) → scene-script.md (hook → agitate → payoff → CTA/loop with continuity written across boundaries) → prompts.md (IMAX-scale camera language, per-scene lens/tempo/lighting, locked film-grade token, native dialogue/SFX, a labeled negative prompt + locked seed per scene, and a header telling you exactly where to upload reference images for YOUR tool). Built-in example — "The Last Courier" cinematic action reel (full character sheet + scene plan + generated prompt pack).

"Make a cinematic action reel with scenes and video prompts — same character in every scene, IMAX look, great color grading. I generate in Kling." → character sheet → scene script → prompt pack (self-verified) → paste into your tool → audit → deliver

Episodic story series (serial-story-reels)

The serialized sibling of veo-cinematic-reels — multi-episode story video series (comic, love story, action, thriller, fantasy — or the user's own character photos) where the SAME characters stay consistent ACROSS every episode:

cd skills/serial-story-reels
node scripts/series-arc.mjs --plan series-plan.json --bible story-bible.md   # validates the arc + writes the story bible
node scripts/episode-prompts.mjs --plan series-plan.json --out prompts.md --vo voiceover.md   # builds + self-verifies the prompt pack + VO sheet

Deliverables: story-bible.md (season arc, per-episode hooks + cliffhangers, locked grade/world/cinematic tokens, cast blocks) → character-sheet.md (verbatim character blocks + 2–3 reference-image prompts per character → upload to Flow's Ingredients ONCE, reuse for the whole series) → prompts.md (ONE cinematic Veo 3.1 prompt per scene, grouped by episode, every prompt carrying the verbatim character blocks + grade + cinematic tokens, self-verified word-by-word, first/last-frame bridging 🔗 across scenes AND episodes) → voiceover.md (every line per episode with delivery direction — native Veo audio or a Kokoro post pass). Built-in example: "Neon Hearts" — a 3-episode comic love-story action series (2 characters, 12 scenes, full pack in examples/neon-hearts/).

"Make me an episodic comic love story — 3 episodes, same 2 characters in every episode, cinematic action and proper voiceover, Google Flow prompts." → interview (≤3 questions) → story bible (arc validated) → character sheets → episode prompt pack (self-verified) → voiceover sheet → audit harness → paste into Flow → deliver

The audit harness ("is it good to go?" before you generate): after the pack is built, run the automated check then spawn a fresh auditor:

cd skills/serial-story-reels
node scripts/audit-series.mjs --pack examples/neon-hearts --out series-audit.md   # automated: arc re-validation, per-scene token consistency, verify/bridge counts, voiceover coverage, cinematic language, sheet + bible
# then spawn a fresh series-auditor subagent (see SKILL.md Stage 7 / templates/series-auditor-brief.md)
# → completes the consistency-worthiness scorecard (/50, ≥35 = good to go) → signs PASS or FIX NEEDED

AI photoshoots (photoshoot-studio)

Copy-paste image prompts for person AND product photoshoots — Google Flow / Nano Banana Pro, Midjourney, Flux — engineered so the same person or the same product never changes between shots (reference-image ingredients uploaded once + a verbatim subject block in EVERY prompt, self-verified word-by-word):

cd skills/photoshoot-studio
node scripts/shot-prompts.mjs --plan shoot-plan.json --out prompts.md   # builds + verifies the prompt pack

Deliverables: subject-sheet.md (verbatim person OR product block + 2–3 reference-image prompts) → shot-list.md (hero → detail → lifestyle → closing) → prompts.md (professional photography language — camera body, lens, f-stop, lighting setup, film stock; locked grade token; per-platform aspect ratios; localized Edit: prompts for re-posing / re-outfitting / re-lighting). Built-in examples — "Ava — Founder Editorial Shoot" (person, warm Kodak Portra grade) and "Brew & Co Tumbler — E-commerce Shoot" (product, high-key commercial grade).

"Give me AI photoshoot prompts — same person in every shot, different outfits and backgrounds, editorial look. Also a product photoshoot for my tumbler store." → subject sheet → shot list → prompt pack (self-verified) → paste into Flow / Midjourney → audit → deliver

Paid ad campaigns (paid-ads-studio)

The paid engine for Meta + Google — a launch-ready campaign pack from one brief, with a forecast of expected results before a dollar is spent. Every "I want to create ads for X" prompt starts with the Ads Wizard — exactly 3 questions (1. platform: Meta / Google / Both · 2. goal: Sales / Leads / Traffic · 3. budget + AOV, with defaults if skipped) before anything is built:

cd skills/paid-ads-studio
node scripts/forecast-ads.mjs --platform meta --objective sales --daily-budget 50 --aov 40 --niche ecommerce --days 30 --margin 40   # → forecast.md (conservative/base/aggressive)
node scripts/ad-prompts.mjs --plan ads-plan.json --out prompts.md        # Veo 3.1 video + image ad prompts, self-verified
node scripts/ad-copy.mjs --brief copy-brief.json --out copy.md          # hook-first copy, char limits + anti-fluff enforced

Deliverables: campaign-brief.md → forecast.md (shown BEFORE creatives — if the economics don't clear, adjust the brief) → prompts.md (copy-paste Veo 3.1 video ads + Nano Banana Pro / Midjourney / Flux image ads, verbatim product block in every prompt) → copy.md (Meta ≤125/≤40/≤30, Google ≤30/≤90, no fluff) → campaign-blueprint.md (Meta ODAX + Advantage+ audiences · Google Demand Gen + PMax + Search) → cost-plan.md (ramp, learning phase, kill/scale rules) → guidelines-checklist.md (2026 AI-content compliance) → launch-checklist.md (manual copy-paste launch order) → ad-audit.md (the audit harness output). Built-in example: "Brew & Co Tumbler — Paid Launch" (full pack in examples/brew-co-launch/).

The audit harness ("is it worthy?" before you spend): after the pack is built, run the automated check then spawn a fresh auditor:

cd skills/paid-ads-studio
node scripts/audit-ads.mjs --pack examples/brew-co-launch --out ad-audit.md   # automated: hooks, char limits, fluff, consistency, forecast, compliance
# then spawn a fresh ads-auditor subagent (see SKILL.md Stage 8 / templates/ads-auditor-brief.md)
# → completes the hook-worthiness scorecard (/50, ≥35 = worth running) → signs PASS or FIX NEEDED

"I want to create ads for my tumbler store" → Ads Wizard (3 questions: platform → goal → budget/AOV) → campaign-brief.md → forecast → approve → creative pack (self-verified) → copy → blueprint + cost plan → audit-ads.mjs → ads-auditor subagent (hook-worthiness scorecard, PASS/FIX NEEDED) → deliver

Vibe-code web apps (vibe-code-webapp)

Build or extend production-ready vibe-coded web apps at 20-year-expert depth — "I want to build something like X with Next.js" or "add subscriptions to my existing app" — on any CLI or web builder (Lovable, Bolt, v0):

0. Onboard — a detailed idea interview (templates/idea-interview.md) extracts the idea, users, scope, stack, business, timeline, success metrics — and design + AI intent (section 9) — (each question: answer or "skip — you decide"). In existing projects, a structure scan runs first — node scripts/scan-project.mjs --dir . --name app → project-scan.md (stack, routes, data/auth/payments, gaps) — so the plan extends what's already there. MEMORY.md is initialized as the daily shared memory
1. Expert research — node scripts/research-idea.mjs --niche "saas for freelancers" --subreddits "freelance,Entrepreneur" --geo US → idea-brief.md (Reddit pain + Google Trends, no API key), then the deep layer from templates/research-playbook.md: competitor teardown, TAM/SAM/SOM, channels, pricing anchors, and a one-line positioning statement
2. SaaS validator — scored on the /35 scorecard with evidence, verdict computed by node scripts/saas-score.mjs --scores <7 numbers> → BUILD / ITERATE / PIVOT, then kill-criteria check + unit economics + kill guardrail + ≥1 validation move → validation.md. Any kill criterion → stop and talk
3. Build Pack + TODO — PRD.md + stack-blueprint.md (locked design source of truth — Figma via the Figma Developer MCP, Google Stitch DESIGN.md, or the open-source pack (templates/frontend-design.md) — + backend architecture + AI-feature rails when the PRD has AI (templates/ai-logic.md) + paste-ready data model + numbered build order + handoff prompts) + sitemap.md (the whole app on one page: full sitemap with every route, every frontend page block, the backend architecture and all workflows) + TODO.md — the build order as P0/P1/P2 tasks managed by node scripts/todo.mjs add|priority|done|list|confirm → your approval — nothing is coded before you approve the pack AND the todo list, and you can add/re-prioritize tasks anytime
4. Build anywhere, step by step — copy templates/vibe-coder-instructions.md → BUILD.md (the vibe coder's operating manual: flowchart + golden loop — pick highest-priority task → implement → run → verify design parity per screen → done → commit → report). Works in any CLI (runnable at every step, one task at a time, done only when verified) or pasted into Lovable / Bolt / v0 via the handoff prompt. Every session files a detailed build-report.md (what/where/evidence)
5. Audit — node scripts/audit-webapp.mjs --dir . --name app --payments --ai on whatever the builder produced (--payments if monetized, --ai if AI features) → auditor subagent signs PASS
6. Final review — the everything-auditor — a subagent analyzes the app, the plan, the instructions, the memory and the reports (including design parity + AI rails); then a fix loop applies hardening fixes, adds test-harness gaps, brainstorms next angles into NEXT.md, and feeds skill improvements back (with your approval) → signs off PASS → deliver

Daily memory (always-on, separate from the stages) — the AI reads MEMORY.md at session start and appends a Did / Decided / Blocked / Next entry at the end; you can add notes anytime, and any tool picks up exactly where the last session left off

Ships templates/prompts.md (copy-paste + handoff prompts), templates/idea-interview.md, templates/research-playbook.md, templates/saas-validator.md (→ validation.md via scripts/saas-score.mjs), templates/frontend-design.md (Figma MCP + Google Stitch + design QA), templates/ai-logic.md (AI-feature rails), templates/design-system.md, templates/backend-architecture.md, templates/sitemap-pages.md (→ sitemap.md — full sitemap + pages + backend architecture + workflows), templates/stack-blueprint-template.md, templates/prd-template.md, templates/todo.md, templates/memory.md, templates/vibe-coder-instructions.md (BUILD.md manual), templates/build-report.md, and templates/production-checklist.md.

"Add a subscription tier to my existing app. Remember where we left off." → scan → interview → research → evaluate → pack + TODO → approve → build (CLI / Lovable / Bolt) → audit → deliver + memory

---

Render tooling (in this repo)

The render/ folder contains the pipeline scripts the skills use to turn an HTML composition into a 4K MP4:

Script	Purpose
render-frames.mjs	Seekable frame renderer → deterministic 4K MP4 (CLI: --html --name --duration --fps --scale --audio; auto-seeks <video data-start> clips for asset reels)
generate-voice.mjs	Kokoro-82M TTS voiceover lines (deep male default: am_fenrir), auto-fit to beat windows, cap 1.15×
enhance-voice.mjs	Deep-voice enhancer (Tier 1): duration-preserving pitch shift (−2 st) + 120 Hz warmth EQ via FFmpeg — CPU-only, no new deps (see voice-plan.md)
mix-audio.sh	FFmpeg sidechain-ducking mix to -14 LUFS
cut-assets.mjs	FFmpeg cutter: pre-cuts video/image assets into per-beat 1080x1920 clips from a storyboard.json manifest
generate-caption.mjs	Writes caption.md from storyboard beats; auto-checks every section into the 500–900 char window
audit-composition.mjs	Post-generation audit: captures one keyframe per beat, auto-checks the 9:16 safe zone, text overlap, word caps, timeline coherence + determinism lint; writes audit-report.md for the auditor subagent
trend-hunt.mjs	Pre-generation viral research (no API key): Reddit top-of-day posts from niche subreddits + Google Trends "Trending now" RSS → trend-brief.md scaffold for the brainstorm + viral scorecard
render-carousel.mjs	Carousel-post renderer (carousel-post-images) — two modes on one deck: browser mode renders .slide elements → per-slide PNGs via headless Chrome (--4k = 4320×5400 / 4320×4320); --mode model exports the deck as per-slide native image-model prompts (prompts.md)

cd render
npm install          # playwright (headless Chrome driver)

# Render a composition to vertical 4K into output/my-reel_4k/
node render-frames.mjs --html my-reel.html --name my-reel_4k --duration 15 --fps 30 --scale 2

# With a mixed audio track
node render-frames.mjs --html my-reel.html --name my-reel_4k --audio assets/full_mix.m4a

Requires Chrome and FFmpeg on your PATH. All paths are relative to the script, so the pipeline works from any clone location.

---

License

MIT — free to use, modify, and monetize. See LICENSE.