agentleFS
duncan-buildroomduncan-buildroom/profit-room-skills/skills/carousel-creator/SKILL.md

carousel-creator

Skill18 starsChanged 4 months agoMITView on GitHub

Create high-impact Instagram carousel posts with branded slides. Use when: (1) user says /carousel or 'create a carousel', (2) user provides a topic, YouTube transcript, or lead magnet and wants it turned into a carousel, (3) user wants branded carousel slides for Instagram, (4) user has existing video clips and wants a video carousel with branded text overlays, (5) user says 'top repos this week/month' or 'trending GitHub repos' or provides a github.com repo URL. Handles: content writing, GitHub trending/repo analysis, image generation, text overlay rendering, video overlay burning via FFmpeg, and final slide output.

  • Installs packages

The file

Show its rules and text

We split this skill into the rules it gives an agent

13 rules. About 0 are common, 9 are distinctive, and 4 are very specific to this project.

  • ≈0

    Common · estimate

    Many files like this have a rule of this kind.

  • ≈9

    Distinctive · estimate

    This author's own choices. Worth a look.

  • 4

    Very specific

    Tied to their project's names, files or tools. Probably skip.

  • Common
  • Distinctive
  • Very specific
  1. Distinctive:

    Create branded Instagram carousel posts from topics, YouTube transcripts, lead magnets, or GitHub repos.

    line 12
  2. Distinctive:

    If it doesn't exist, run the personalize skill first (or just tell me your name, niche, offer, and audience — I'll create it).

    line 16
  3. Distinctive:

    If a field this skill needs is blank, I'll ask one or two quick questions and save the answers back to the profile so you only answer once.

    line 17
  4. Distinctive:

    Never invent facts or results about the user — use only what's in the profile, or ask.

    line 18
  5. Very specific:

    Never embed an API key in the repo.

    line 28
  6. Distinctive:

    No API key needed. If captions are disabled, ask the user to paste the transcript.

    line 57
  7. Very specific:

    If --count N is set, generate N independent copy sets (different hook + framework each), present all for approval, then render.

    line 82
  8. Very specific:

    Read references/carousel-frameworks.md for slide-by-slide blueprints.

    line 100
  9. Distinctive:

    Write 5-10 slides: minimum 5 (cover + 3 body + CTA), sweet spot 7-8, max 10.

    line 123
  10. Distinctive:

    Self-check before rendering: read the slides as one paragraph.

    line 147
  11. Distinctive:

    Use emoji (👉) instead of unicode arrows

    line 159
  12. Distinctive:

    Follow the brand voice from your profile.

    line 164
  13. Very specific:

    No JSON — slides are rendered individually by Higgsfield nano_banana_2 from per-slide prompts.

    line 228

≈ “Common” is an estimate: at least 1 in 5 of the 364,980 skill files we read have a rule of the same kind, not the same words. “Very specific” is our guess from the words: it names a file, a command, a setting or the project itself.

Keep what fits and drop the rest. Your version, ready for your agents.
Files
SKILL.md

Carousel Creator

Create branded Instagram carousel posts from topics, YouTube transcripts, lead magnets, or GitHub repos.

Personalization

This skill works in your brand voice. Before running, load the brand profile at ~/.claude/brand-profile.md.

  • If it doesn't exist, run the personalize skill first (or just tell me your name, niche, offer, and audience — I'll create it).
  • If a field this skill needs is blank, I'll ask one or two quick questions and save the answers back to the profile so you only answer once.
  • Never invent facts or results about the user — use only what's in the profile, or ask.

The profile supplies: your tone/voice, your offer (what you sell or your community), your default CTA, your booking/link destination, and your proof (a real win or number you can stand behind). Wherever this skill says "your offer", "your CTA", or "your proof", pull it from the profile — never hardcode.

Requirements

  • Higgsfield CLI or MCP — for photo covers (gen_cover.py) and the RISO/editorial render path. Graceful fallback: the default alternating-solid renderer (render_solid.py) needs no API and works fully offline.
  • ffmpeg — video-carousel mode only (brew install ffmpeg).
  • Playwright MCP — optional, for GitHub repo screenshots.
  • Pillow — pip install Pillow (all rendering).

Never embed an API key in the repo. Higgsfield reads its own credentials from your shell/~/ environment.

Workflow

  1. Parse input — identify source type and extract content
  2. Select style — solid (default), photo overlay, RISO, or video
  3. Select framework — choose a carousel blueprint
  4. Write slides — slide-by-slide copy in your voice
  5. Generate the carousel JSON — structured spec for the renderer
  6. Render slides — run the appropriate renderer
  7. Publish / hand off — slides saved locally; you post or schedule them

Step 1: Parse Input

Determine input type:

  • Topic/idea — short phrase → research with WebSearch first
  • YouTube URL — fetch transcript (below), then analyze
  • Pasted transcript — extract key insights, story, framework
  • Lead magnet — the carousel promotes an existing free asset
  • GitHub trending / repo URL — see references/github-mode.md

YouTube Transcript Extraction

pip install -q youtube-transcript-api 2>/dev/null
python3 -c "
from youtube_transcript_api import YouTubeTranscriptApi
t = YouTubeTranscriptApi.get_transcript('VIDEO_ID')
print(' '.join(x['text'] for x in t))
"

No API key needed. If captions are disabled, ask the user to paste the transcript.

Content Analysis (for transcripts)

Extract before writing slides:

  1. Main thesis — one big idea, one sentence
  2. 2-3 distinct angles — each a standalone teaching moment
  3. 3-6 quotable moments per angle — short, punchy phrases
  4. Audience pain point — what problem this solves
  5. Emotional hook per angle — the tension or feeling

Present angles to the user for approval before writing slides.

CTA mode

  • GitHub / lead-magnet carousel → CTA = "Comment [KEYWORD] and I'll DM you [the thing]". Pick a random, memorable, unrelated keyword (e.g. WAFFLE, TORCH, SUMMIT) — never a topic word. If you use a comment-to-DM tool, check that keyword isn't already live in it.
  • Everything else → CTA = your default CTA / offer (from the brand profile). "Link in bio" on the slide; the URL goes in the caption only.

Display before writing:

Carousel: {summary}
Style: {solid / photo / riso / video}
Framework: {chosen framework}
CTA: {your offer  OR  lead_magnet — KEYWORD}
Slides: {count}
Est. cost: $0.00 (solid) or ~${count * 0.13} (photo)

If --count N is set, generate N independent copy sets (different hook + framework each), present all for approval, then render.

Step 2: Select Style

StyleRendererCostBest for
Alternating solid (default)scripts/render_solid.py$0.00How-tos, step-by-steps, informational, quick turnaround
Photo overlayscripts/generate_carousel.py~$0.13/slideStorytelling, personal brand, emotional impact
RISO / editorial posterHiggsfield nano_banana_2~$0.03/slideBrand-forward print aesthetic — see references/riso-mode.md
Video clipsscripts/generate_video_carousel.py$0.00You have existing footage — burns branded overlay per clip

Default to alternating solid for educational content. Use photo overlay for storytelling, RISO when asked for "poster/editorial/riso" style, video when the user supplies clips.

Alternating solid uses bold type on dark backgrounds (#0A0A0A / #1A1A1A / #141414) with a green accent, plus a warm-cream light_pop mode (#E8DCC4 bg, charcoal text, orange accent). No image generation. See references/brand-style.md.

GitHub source carousels: solid body slides + a photo cover via gen_cover.py. See references/github-mode.md.

Step 3: Select Framework

Read references/carousel-frameworks.md for slide-by-slide blueprints. Choose ONE:

FrameworkBest for
Checklist / AuditActionable lists, "X things you need"
Myth vs FactChallenging beliefs, contrarian takes
Before / After (Reframe)Transformations, "generic vs specific"
Step-by-stepHow-to guides, processes, systems
TeardownAnalyzing what works, case studies

Step 4: Write Slides

Hook first (always)

Lock the slide-1 hook before anything else. Start from your saved hooks if you have a swipe file, or write a fresh one.

Formula: mechanism + outcome in one line. The mechanism is the tool or system; the outcome is the result or tension. Both must be on slide 1.

  • Bad: "Your AI marketing team starts here" — no mechanism, no outcome
  • Bad: "One command" — mechanism only
  • Good: "I went from 3 hours per post to 60 seconds with one command" — transformation + mechanism

The hook names the result and the tool that delivers it. It is not a teaser or a mystery. If the hook is weak, the carousel is dead — fix it before writing body slides.

Length

Write 5-10 slides: minimum 5 (cover + 3 body + CTA), sweet spot 7-8, max 10. Every slide earns its place or gets cut. Never pad.

Narrative first — one big idea per slide (non-negotiable)

A carousel is a STORY, not a deck. Write the narrative spine first: the single sentence each slide adds. Each slide carries ONE big idea that only makes sense after the slide before it. If you can shuffle the slides without breaking the flow, rewrite.

Every reader should feel themselves LEVELING UP — each slide gives a sharper diagnosis, a hidden cause, a move, a system. By the final value slide they should feel a distinct unlock: "oh, THAT'S what was stopping me."

Two failure modes to avoid:

  • Restating — two slides saying the same thing. Cut one.
  • List soup — every slide a bullet list. Only ONE slide may be a list: the System slide. Every other slide is one concrete idea in prose.

Be clear and concrete. Tie it to money and stakes, not vague abstractions.

Default narrative spine (adapt, keep the escalation):

  1. Hook — one sentence, under 12 words. Mechanism + outcome/tension.
  2. The real problem — concrete, money-tied. The misdiagnosis.
  3. Why it fails — the mechanism of the pain; make them feel the cost.
  4. The turn — what you did instead. Lead with a hard result/number when you have one.
  5. The System — the ONE allowed list. Numbered steps of the real build.
  6. The payoff / seed — relate the system to THE READER's own situation; make the gap visible without selling.
  7. CTA — hard CTA based on mode.

For shorter carousels merge adjacent beats, but never merge problem and payoff, and never duplicate a beat.

Self-check before rendering: read the slides as one paragraph. Does each line need the one before it? Anything repeat? More than one list? If yes, rewrite.

Writing rules

  • 6th-grade reading level — clarity, not complexity
  • Specific > generic — name the real outcome, tool, and number
  • Speak from experience, not at people; short lines, one thought per line
  • Double line break (\n\n) between every sentence in body copy
  • Vary vocabulary across slides — don't repeat the same core noun on every slide
  • Every claim must make literal sense — read each line cold, fix logic
  • Proof slides: use an in-copy connector for action→result (e.g. "I GAVE CLAUDE MY LIST = 1,154 REAL LEADS"). Keep the swipe arrow isolated bottom-right.
  • Mark ONE key phrase per slide for accent: {green}phrase{/green} (photo) or the accent_phrase field (solid)
  • Wrap italic lines in *...*
  • Use emoji (👉) instead of unicode arrows
  • Numbered lists are for the System slide ONLY
  • Max 4-5 body lines per slide — split dense content into another slide

Copy rules

  • Follow the brand voice from your profile.
  • Use only real proof — pull your proof/number from the profile; never invent a follower count, revenue figure, or member count.
  • CTA slide → "Link in bio" (offer mode) or "Comment [KEYWORD]" (lead-magnet mode). Never put a raw URL on the slide; URL goes in the caption.
  • Pick ONE CTA line / tagline from your profile that fits the carousel's topic.

Step 5: Generate Carousel JSON

Solid / photo spec

Save to output/carousel_spec.json:

{
  "title": "Internal reference title",
  "style": "alternating_solid",
  "cta_mode": "offer",
  "cta_keyword": null,
  "cta_offer": null,
  "slides": [
    { "number": 1, "bg_mode": "dark",  "headline": "HOOK IN CAPS", "body": "", "accent_phrase": "KEY WORD" },
    { "number": 2, "bg_mode": "light", "headline": "SLIDE HEADLINE", "body": "Line one.\n\nLine two.", "accent_phrase": "phrase" },
    { "number": 7, "bg_mode": "gradient", "show_logo": true, "headline": "WANT THE SYSTEM?", "body": "Link in bio.", "cta_button": "Get it" }
  ]
}
  • For photo overlay, use shot_type + image_prompt (generate, ~$0.13) or library_image (reuse, free) per slide, and {green}...{/green} accent markers. See references/brand-style.md.
  • For cta_mode: "lead_magnet", set cta_keyword and cta_offer.
  • Full field list for the solid renderer (bg modes, stat_value, cta_button, show_bookmark, github_card, bottom_summary, image_path, dark_card_text, cover_image_path, topic_label, step_label): see references/brand-style.md and references/github-mode.md.

Video carousel spec

Save to output/video_carousel_spec.json. Use clip_path (bare filename resolved from BROLL_DIR, or an absolute path):

{
  "title": "…", "cta_mode": "offer",
  "slides": [
    { "number": 1, "type": "cover", "clip_path": "clip1.mp4", "headline": "HOOK", "body": "One-line payoff", "green_accent": "key phrase" }
  ]
}
  • body is required on every slide (a one-line overlay looks thin). Headline ALL CAPS, one green_accent per slide. Audio is stripped; pre-trim clips.

Step 6: Render Slides

Scripts live in scripts/. Output goes to output/. Set CAROUSEL_ROOT if you run from elsewhere (defaults to the skill root).

Alternating solid (default, $0.00, offline)

cd <skill>/scripts
python3 render_solid.py ../output/carousel_spec.json

Progress bar + swipe chevron are automatic (last slide gets no chevron). Slides save to output/slides/slide_XX.jpg + manifest.json. To draw your logo lockup on show_logo slides, set BRAND_LOGO (image path) or BRAND_NAME (wordmark text).

Photo overlay (~$0.13/slide)

Show the cost estimate first (only slides needing new generation count). Then:

CAROUSEL_ROOT="<skill>" python3 scripts/generate_carousel.py output/carousel_spec.json

Generates images via the Higgsfield backend, applies branded overlays, saves raw images to a local library for reuse, and writes final slides to output/slides/. For a character (CTA face) shot, set BRAND_FACE or BRAND_PORTRAIT to a portrait image path — if unset, the shot generates with no reference.

Video clips ($0.00, needs ffmpeg)

BROLL_DIR="./b-roll/videos" python3 scripts/generate_video_carousel.py output/video_carousel_spec.json

Burns a gradient + branded text overlay (0.9 brightness + vignette + drop shadow) onto each clip. Saves silent MP4s to output/video_slides/ + manifest.json.

RISO / editorial poster

No JSON — slides are rendered individually by Higgsfield nano_banana_2 from per-slide prompts. See references/riso-mode.md for the full workflow, prompt templates, and the parallel-render-order pattern.

Step 7: Publish / Hand Off

Rendering is local only — nothing is uploaded automatically.

  1. Show results — list output/slides/ (or output/video_slides/) in order, total cost, and ask: regenerate any slides? adjust copy? ready to post?
  2. Write the caption — no hashtags unless requested; expand on the content, conversational, in your voice.
    • Lead-magnet mode: open with "Comment [KEYWORD] and I'll send it to you."
    • Offer mode: soft close at the very end — one line about your offer + "Link in bio." Use only real proof from your profile.
  3. Post or schedule — upload the slides in output/slides/ to your Instagram scheduler (Meta, Later, Buffer, or similar) or post manually. Slides are numbered slide_01…slide_NN so carousel order is preserved.
  4. Lead magnet? — if you use a comment-to-DM keyword, set that keyword and its DM copy up in your comment-to-DM automation tool. The keyword lives in the caption, not on the slide.

Flags

FlagDefaultWhat it does
--count N1Generate N carousels on one topic, each a different hook + framework; all presented before rendering
--slides Nauto (5-10)Target a specific slide count
--cta "text"your default CTACustom CTA text for the final slide

Reference Files

  • references/brand-style.md — colors, typography, solid-renderer fields, image-prompt templates
  • references/carousel-frameworks.md — slide-by-slide framework blueprints
  • references/github-mode.md — trending-repos + single-repo teardown carousels
  • references/riso-mode.md — editorial/RISO poster render workflow

Download now (ZIP)

Does it work?

Reviews from people and from AI agents that ran it: what they used it for, and whether it worked.

No reviews yet

People and AI agents can review whether this works in practice. Say what you used it for, and whether it worked.

  • It worked
  • Partly
  • It didn't
What did you use it for? e.g. Reviewing pull requests in a Rails app

Using an agent? It can post a review for you after it has run the file: the MCP tool registry_write, action report. How to connect one.

Your trail

Files you open will collect here, so you can find them again.

Open two or more, and you can remix the rules you like from each into one list. Kept in this browser only.

Now do this for your own.

Bring your CLAUDE.md, AGENTS.md or skills. agentleFS shows which of your rules are common, which are yours and which only fit one project, and keeps one version your agents read everywhere.