Pinned Tweet
I spent 10+ years in QA automation. Then I left corporate IT to build with AI full time.
Since then:
- trained 1500+ people through Social Eagle AI
- 94K subscribers on my Tamil AI channel
- building CoDude.io and shipping something every week
Here I post the real stuff. Claude Code setups, n8n agents, what broke and what it cost.
Follow if you'd rather build than just read about AI.
11
1,550
A 26B open model that fits in about 20GB just beat every open model under 100B on DeepSWE. It didn't get bigger. It thinks in loops.
Thinking Machines released Inkling 2 Mini on 20 Nov. It's a looped transformer: the middle layers run three times for every token, so it reasons more without writing longer answers.
From their own table:
- DeepSWE v1.1: 48.2% vs 42.2% for Qwen3.8-27B
- SWE-bench Pro v2 Hard: 77.9% vs 73.8%
- MoE with 26.4B total, 3.4B active
- About 20GB memory. They ran it offline on a 48GB MacBook Pro
Who it helps: devs who want a local coding agent without sending code to the cloud.
Honest limits: scores are self-reported and setups vary by model. Qwen3.8-27B still wins on Terminal Bench 2.1 (73.0 vs 71.6). Looping also doesn't add knowledge, so factual recall is only in line with similar sizes. Pair it with search.
thinkingmachines.ai/news/ink…
1
1
12
Most "AI automation" still stops at the browser tab. The boring work lives in your desktop apps and local files.
Google just put computer use into the Gemini app for Mac (announced 19 Nov).
Before:
Open the PDF, copy weights and addresses, paste them box by box into a shipping quote form. Repeat for every order.
After:
Describe the task once. Gemini clicks through menus, types and fills forms in an isolated background window, while your own screen and cursor stay free.
The safety design is the real story:
• You choose which apps it can touch
• You watch every step in a live preview
• You can pause anytime, and it asks before key actions
Google's own examples: shipping quotes filled from local PDFs, a Slides deck built from live dashboards, resale listings staged from a folder of photos.
Who it helps: small business owners and ops folks stuck in copy-paste work.
The catch: English and US only for now, Apple Silicon Mac on macOS Sequoia 15+, and you need Google AI Plus, Pro or Ultra. No word yet on other countries.
blog.google/innovation-and-a…
Which repetitive task would you hand over first?
3
3
30
Your agent is not forgetful. You asked it to remember everything.
Anthropic published this on 6 Nov 2026 in "The components of agent memory." Their finding: good memory is a prediction of what the next session will need, not a dump of every log line.
They compared two Claude models playing Pokémon at about 10,000 steps.
Claude Sonnet 3.7 wrote 413 memory files across 225 folders and earned 1 badge. It stored battles, items and NPCs as they happened. Then it spent turns creating and deleting files instead of playing. The notes went stale because the game state keeps changing.
Claude Fable 5 wrote 9 files in 2 folders plus one global.md and earned 4 badges. It stored strategy, plans and tested rules. Things you can read off the screen, like money or map position, stayed off the disk.
The same pattern showed up on a real document pile. Anthropic used about 350,000 internal files from Stainless, roughly 420 million tokens, and 25 questions. Opus 5 agents built a linked wiki over 128 offline "dreaming" sessions, checking notes against the source files. After 2 sessions the wiki hit 13% of graded criteria. After 128 it hit 48%. Search alone scored 53%. Search plus the wiki rose from 54% to 61% as the wiki grew, and they report about 20,000 fewer tokens per question.
Dreaming is their word for an offline pass: the agent reads old transcripts, corrects wrong notes, and rebuilds a small map of what matters. In one Pokémon run, Claude wrote that Mt. Moon was already behind it. That note was wrong. After dreaming checked the transcript, it went back into the cave and reached the next city. Without the correction, the run never got past the mountain.
If you keep a second brain or a persistent agent, this is the method their post points to:
1. Keep the always-loaded index small. Anthropic says Claude Code auto memory only injects the first 200 lines or 25 KB of MEMORY.md.
2. Write rules you tested, not raw state you can look up again.
3. Run an offline pass that checks notes against original logs before the next session trusts them.
4. Tell the agent what memory is for. When they told models to "record only failures," stronger models lost much of the gain that memory had given them.
The catch, from the same post: dreaming costs extra compute. A wrong note promoted into every new session can poison the next run. And a wiki can grow larger than the documents it describes, which helps nobody.
These are Anthropic's own evaluations, on their models and harness, not an independent bake-off. Pokémon and an internal company corpus are not your client folder. Treat the scores as direction, not a promise for your stack.
anthropic.com/research/compo…
If you keep a memory folder for your agent, what do you store: raw logs, or rules you have actually tested?
7
2
273
Your client's launch video dies the day they change one line of copy.
Most editors reshoot or rebuild from scratch. There's a Claude Code and Codex skill that turns the video you already have into a Remotion project you can edit as React.
It's remotion-clone-video by crafter-station. 18 GitHub stars. Created 21 Jul 2026. Public repo, checked via GitHub API, README and SKILL.md this run.
Give your agent the file and say: clone this, pixel perfect.
What it does, in 8 steps:
1. Probe the source with ffprobe (size, fps, duration)
2. Extract a 10fps frame survey plus the audio track with ffmpeg
3. Storyboard every scene, element, animation and transition
4. Sample exact colors pixel by pixel
5. Crop only what can't be code (logos, mascots, photos) and give them real alpha. Everything else becomes React.
6. Build one Remotion component per scene from 15 recipes, including typewriter, blur reveals, gradient shimmer, draw-on strokes, card morphs, camera zooms, push transitions and particle dissolves
7. Render stills and compare them to the source frames until they match
8. Render the final mp4 and spot-check it
Install:
npx skills add crafter-station/remotion-clone-video -a claude-code
For Codex:
npx skills add crafter-station/remotion-clone-video -a codex
The repo is clear: use this on your own videos, or as a study tool. Don't clone someone else's brand film and sell it as yours.
Money play: take a SaaS founder's own product demo, rebuild it as code, and sell updates (new headline, new price, new CTA) without another shoot.
github.com/crafter-station/r…
Would you sell the first rebuild, or a monthly "change the copy, re-render" retainer?
Stop paying about $50 a month for video tools your Mac can run for free.
zero-dollar-studio is a free MIT repo of Claude Code skills that makes Shorts with local tools. No API keys, no credits, no watermark.
Before (prices from the repo's own table, checked 30 Oct 2026):
OpusClip for podcast to Shorts: $15/mo, free plan watermarked
ElevenLabs Starter for a voice clone: $6/mo
HeyGen for an AI presenter: $29/mo, free plan watermarked
After:
snipshorts: podcast link to vertical Shorts, speaker tracked, captions, posting copy
edit-footage: your raw talking-head take to a reel, best take of each line, ums and pauses cut
make-video: idea to script, Chatterbox voice, Whisper captions, music, Remotion render, QA check
cartoonsmith: voiced cartoon episodes with lip sync
Install in Claude Code:
/plugin marketplace add TheNoumanDev/zero-dollar-studio
/plugin install make-video@zero-dollar-studio
Codex: npx skills add TheNoumanDev/zero-dollar-studio -a codex
The catch: Apple Silicon Mac only, about 8 GB free disk, and Remotion needs a paid license for companies above 3 people. The AI face step is still on the roadmap.
Money play: run snipshorts on a local podcaster's episodes and sell a monthly no-watermark Shorts pack. Your tool bill stays at zero.
github.com/TheNoumanDev/zero…
Which would you sell first, podcast clips or cleaned-up talking-head reels?
1
17
741
Stop paying an AI video model to invent every frame of your story Shorts.
There's a free Claude Code skill that draws 55 to 70 second vertical cartoon stories on your own PC. No video generator. three.js inside Remotion.
Repo: github.com/mromidev/claude-a…
MIT. 2,238 GitHub stars. Created 2026-11-01 (checked via GitHub API + README + SKILL.md this run).
The skill is called cartoon-3d-short. 2D bean-head characters in a painted 3D town: ink outlines, lens blur, gliding cameras, a march score and synthesized sound effects.
Their proven example: The Bridge That Fell for a Song, a 67 second short about the 1831 Broughton bridge and 74 soldiers marching in step.
Playbook:
1. Install the skill
git clone github.com/mromidev/claude-a… ~/.claude/skills/cartoon-3d-short
2. Install Remotion with @remotion/three, plus Node, Python 3 and ffmpeg.
3. Point REMOTION_TMP at a folder on a big drive. Every Remotion bundle copies public/ (about 1.3 GB) into temp and does not clean it up.
4. In Claude Code, say:
Use the cartoon short skill and make a short about [one surprising true story].
5. Claude follows SKILL.md: sourced facts, then a script (hook, escalation, snap, explanation, callback), voice with word timings, a shot plan in story.ts copied from templates/bridge/, stills checked against the style reports, then the 2K render at 1440x2560.
The look lives in templates/bridge/. Story, characters and voice have to be yours. The reference frames are for style study only.
Money play: sell history or science pages a fixed-price "true story animated short" pack. You run it locally. They get a 55-70s vertical with no video-model invoice.
Would you pitch this first as a history short or a science explainer?
Stop uploading raw Simulator recordings as your App Store preview.
There's a free MIT Claude Code skill that records your real app, puts it in your own brand, and checks Apple's preview rules before you upload.
It's called product-video by kengomatsuo. One star on GitHub, so almost nobody has found it.
What it does:
- Reads your app's brand from its own files: accent, fonts, icon, screenshots and clips
- Records the real product moving, from the iOS Simulator or a Mac/browser window. Not generated UI.
- Renders that recording on a 3D iPhone, iPad or MacBook in Remotion
- You approve a storyboard and an animatic before any final render
- One storyboard, six presets: landscape 1920x1080, vertical 1080x1920, square 1080x1080, plus App Store cuts for iPhone (886x1920), iPad (1200x1600) and Mac (1920x1080)
- check.ts fails the file if length, fps, audio or bit rate break Apple's preview rules: 15 to 30 seconds, at most 30 fps, stereo audio
Install (use the full HTTPS URL; the short name clones over SSH and fails without a key):
claude plugin marketplace add github.com/kengomatsuo/produ… && claude plugin install product-video@product-video
Needs bun, ffmpeg and Xcode command line tools. Simulator capture and App Store upload only work in Claude Code on a Mac. Remotion is free for individuals and companies of up to 3 people.
Money angle: indie iOS and Mac developers still need store previews and a launch cut. Sell a fixed pack: one App Store preview plus the social cuts, all from recordings of the real app, in their brand.
github.com/kengomatsuo/produ…
Would you sell the App Store preview first, or the social launch cuts?
5
16
898
Your face-cam Reels look amateur because you are doing 5 jobs by hand.
ostack-video-editor is a free, MIT licensed skill pack for Claude Code and Codex that splits a vertical short into 5 skills.
Before:
Write the script yourself. Guess a hook. Cut gaps in CapCut. Hunt for pictures. Type subtitles. Make animation in another app.
After:
1. write-script: research to a fact table, hook options and one script.md with the storyboard
2. hook: opening frame built to work as the thumbnail and fit the profile grid crop
3. edit-video: raw clips to a new CapCut draft with gaps cut, grade, levelled audio and subtitles
4. find-pictures: stills from official sources or Wikimedia, with licence
5. animated-shots: Remotion animation checked by two judges, one running a mute test
Install in Claude Code:
claude plugin marketplace add FOwen123/ostack-video-editor
claude plugin install ostack-video-editor@ostack
Needs CapCut desktop 9.4, ffmpeg, Python, Node, whisper.cpp. macOS or Windows.
The money angle: coaches and shop owners shoot on phones and hate editing. Hand them a finished CapCut draft every week.
github.com/FOwen123/ostack-v…
Would you trust an agent to edit inside your CapCut file, or only on a copy?
1
2
223
Stop letting Claude invent your cut points.
Most Remotion skills write pretty compositions and guess the timing.
remotion-ffmpeg-skill does the opposite. Free MIT Agent Skill for Claude Code, Cursor and Codex. It sits on top of ffmpeg-skill.
How the loop works:
1. ffmpeg-skill probes your real recording: duration, fps, size, silences, scene cuts, audio peaks
2. Remotion draws from those measured numbers only (Jumpcut, Karaoke, PeakPulse, Captioned, and more)
3. render.py checks the output frame for frame against the timeline and exits non-zero on a mismatch
4. deliver.py ships YouTube / Reels / TikTok / Shorts / X with a loudness compliance table
Install:
npx ffmpeg-skill
npx remotion-ffmpeg-skill
Try asking:
"Cut the silences out of talk.mp4, jump-cut style, keep the subtitles in sync, deliver as a Reel."
Money use: coaches and local founders who shoot one long talking-head. You return a tight Reel with captions that stay on the words. Sell it as a monthly pack.
Repo: github.com/kajisho5/remotion…
Which composition would you try first on a real client clip, Jumpcut or Karaoke?
1
4
306
Stop paying a CapCut editor every time a coach wants another Instagram Reel.
andrebrito16/skill-reels-editor is a free MIT Claude Code skill (3 GitHub stars) that edits vertical Reels end to end with Remotion on your machine.
Playbook:
1. Install the skill
npx skills add andrebrito16/skill-reels-editor
Or clone and check tools first:
git clone github.com/andrebrito16/skil…
cd skill-reels-editor
./install.sh --check
./install.sh
2. One-time Remotion project
npx create-video@latest
cd my-reels
npx remotion add @remotion/media
npx remotion add @remotion/google-fonts
npx remotion add @remotion/captions
npx remotion add @remotion/install-whisper-cpp
3. Drop the footage into Claude Code and run /reels-editor
What the skill does next (local):
- Whisper.cpp transcription, then silence cuts with a cut plan you approve before anything builds
- Sony S-Log3 / S-Gamut3.Cine to Rec.709 grade with skin-tone and teeth checks
- Voice EQ + light compression, normalised to -14 LUFS
- Remotion compose: 2-4 word captions, lower thirds, B-roll cutaways, SFX, music bed
- Instagram safe area respected on every graphic
- 540p preview, then H.264 High 2-pass VBR export for Instagram
Money angle: sell monthly Reels packs to coaches and local shops who already shoot talking-head clips. You run the skill, they get upload-ready Reels.
github.com/andrebrito16/skil…
Which step would you productize first for a client, the silence cuts or the S-Log3 grade?
Stop uploading YouTube videos blind. Most creators guess. This free repo measures.
utillayer/youtube-agent-skill: 12 Claude skills plus 9 small Python tools for the parts of YouTube that decide if anyone watches. MIT, no API key, nothing to connect. Works in Claude Code and Codex.
What it actually does:
- /yt-script scores 5 openings and keeps the best one before you record
- /yt-package checks 10 titles against the thumbnail (length, mobile cut-off, repeated words)
- /yt-edit turns a raw .srt into a cut list of silences, filler and false starts with timecodes
- /yt-retention reads your Studio export and shows the second viewers left, with what you were saying
- /yt-shorts pulls the moments that stand alone from a long video
- /yt-chapters drafts chapters YouTube will actually show
Install in Claude Code:
/plugin marketplace add utillayer/youtube-agent-skill
/plugin install youtube-agent-skill@utillayer
Money angle: sell small YouTubers a pre-upload check per video. Hook score, title and thumbnail fix, cut list, Shorts picks. You run the skills, they get a one-page report.
github.com/utillayer/youtube…
Which one would you run first on your last video, /yt-retention or /yt-package?
Stop paying someone to cut silences in CapCut by hand.
An agent can now do it inside your actual CapCut project, and every cut stays editable.
capcut-editing-skill by emirrcodes (github.com/emirrcodes/capcut…) is a free MIT skill for Claude Code, Codex, Cursor and Antigravity. Created 2026-10-04, checked via GitHub API + README this run.
You type one line like:
"project name" cut the non-speech parts and add subtitles
The agent reads your CapCut Desktop timeline and writes the edits back into it:
1. Silence cut: volume based, -30 dBFS threshold, 80 ms minimum gap.
2. Non-speech cut: local Silero VAD finds speech, checks Whisper word timings, keeps a 30 ms edge so words don't get clipped. Built to tell fan and traffic noise apart from speech.
3. Subtitles: meaning-based lines, usually 4 to 6 words, added to the existing timeline.
No screen or mouse control. It edits the project files locally, so the project just needs to be saved and closed while it writes. If it finds zero speech, it stops instead of deleting your whole video.
Needs CapCut Desktop, Python 3.10+, FFmpeg. macOS and Windows install paths (README says Windows open-and-save check isn't fully verified yet). README is in Turkish, the commands work through your agent.
Money play: offer creators a "rough cut + subtitles inside your own CapCut file" service. They keep full control, you skip the boring hour.
Would you trust an agent to cut your raw talking-head take, or only the subtitles?
Your raw phone clips are dying in your camera roll because you think you need an editor.
You don't. 3 free agent skills now do the boring edit for you.
1. claude-shorts by Ansub
Raw vertical phone clip in, finished Short out. Cuts dead air, retakes and filler, word-by-word captions with spelling fixed, music ducked under your voice, -14 LUFS. Runs locally on ffmpeg + whisper.cpp. No API keys, nothing uploaded.
github.com/Ansub/claude-shor…
2. shorts-video-studio by mohdkhatab
Script to a finished 1080x1920 MP4 + SRT: kinetic captions, 8 transitions, synthesized music bed, QA report. Scripts in Hinglish, Hindi or English. No GPU, no video editor.
github.com/mohdkhatab/shorts…
3. ChrisHKAI-Video
Paste a creator's X link, get a short film starring them as a 3D vinyl toy, built in code with Blender. Made for Claude Code, Codex support is still experimental.
github.com/ChrisHKAI/ChrisHK…
Money play: coaches and shop owners shoot on their phones and never edit. Sell them a monthly Reels pack. Pitch creators a custom intro film on top.
Which one are you testing first on your own clips?
You're still pasting an article into CapCut and hoping the captions stick.
That's why your explainer Shorts take half a day and still look generic.
short-video-maker by chemny is a free MIT Agent Skill for Claude Code, Codex and OpenClaw. Drop an article, topic or opinion brief. Get a full narrated vertical short package: script, storyboard, TTS-synced captions, Remotion render, cover image and publish copy.
Defaults from the README:
- 1080x1920
- 90-130 seconds
- first-frame preview before full render
- Edge TTS by default (no paid key for basic runs)
- Node + FFmpeg + Remotion, local
Install:
git clone github.com/chemny/short-vide…
# put the folder in your agent's skills dir so SKILL.md is at the root
npm install
cd remotion && npm install && cd ..
npm run doctor
Copy-paste starter (from their README):
Use the short-video-maker skill to create a 60-second Chinese vertical explainer from examples/input.md. Generate the script, storyboard, video plan, captions, and first-frame preview. Do not render the full video until I confirm.
Swap the language in that brief for Tamil or Tanglish when you sell it.
Money play: monthly narrated explainer Shorts packs for coaches and local brands. Same pipeline, new topic each week, client gets video + cover + publish copy.
Repo: github.com/chemny/short-vide…
Would you sell this as a fixed-price pack per topic, or a monthly retainer?
You're still building every Short from a blank CapCut timeline.
That's why your faceless channel never ships on a schedule.
Before:
Open CapCut. Hunt stock. Guess the hook. Export once. No script file. No rights trail. No way to rerun the same lane next week.
After with content-machine (github.com/45ck/content-mach…):
A free MIT skill pack for Claude Code, Codex CLI and Cursor. 19 GitHub stars. Created 2026-01-06 (GitHub API + README + MIT license checked this run).
You ask the agent for a video outcome. It installs into the repo under .content-machine/, picks a lane, and writes inspectable artifacts: script, audio, timestamps, captions, render metadata, source rights, review notes. Then a normal MP4.
Lanes that ship:
- Reddit / confession stories over gameplay
- SaaS problem-solution shorts
- Longform cut into vertical Shorts
- Motion-card lessons and coded graphics
- 3D / procedural gameplay backgrounds
It won't call an MP4 ready until publish-prep passes.
Money play: sell niche creators a monthly faceless Shorts pack. Same skill pack, new topics every week, artifacts you can show the client.
Install prompt for your agent is in the README Agent Harness Install guide.
Which lane would you sell first: Reddit-over-gameplay, SaaS shorts, or longform-to-Shorts?
114
Stop trusting Claude when it grades its own video.
In the make-short repo, Claude scored its own render 6.5/10. A fresh critic agent scored the same render 4.2.
make-short by nocodehack is a free, MIT Claude Code skill that builds animated YouTube Shorts: cartoon characters in a 3D Blender world, voice, captions and sound, all made with code.
How it works:
1. Breaks a Short you love into numbers (shots, colour, brightness)
2. Writes a hook, escalation, payoff, twist story and fact-checks it. You approve
3. Your own voice (free) or ElevenLabs, every word timed
4. Opus manages 4 Sonnet workers to build look, characters, world, shots
5. A fresh critic scores the render with the exact second of every lost point
6. Fix only what the critic confirmed, render again
Their example went from critic 4.2 to 7.9 after fix rounds.
Nothing gets posted unless you type go.
Money angle: sell niche channels a monthly pack of animated history or explainer Shorts.
github.com/nocodehack/make-s…
Would you trust a critic agent over your own eye on the final cut?
Stop paying for After Effects templates just to sell Reels.
3 free MIT agent skills worth bookmarking if you sell video editing:
1. synchronicity-labs/skills
Official sync. labs skills for Claude Code, Codex and Cursor. Lipsync a clip to new audio, make a photo talk, dub a video into another language with matching lips, or change a few spoken words in the speaker's own voice. Needs a sync. account.
npx skills add synchronicity-labs/skills
2. claude-edits-skill
Claude edits Reels inside the Instagram Edits Mac app through its built-in automation server: overlays, sound effects, titles, auto captions, word-level transcripts, export.
github.com/iamjeeth/claude-e…
3. rudywizz-motion
CLI plus agent skills for motion graphics drawn by code: kinetic type, 3D titles, map flights, lower thirds, captions. 36 styles, transparent MOV for your editor, no AI video model.
github.com/Rudywizz/rudywizz…
Money angle: take a coach's best Reels, dub them into another language, sell it as a per-video add-on.
Would you trust lip-synced dubs on a client's face, or stick to captions?
3
9
487
Stop paying a motion studio for a 60-second product promo.
showreel by revfactory is a free, MIT licensed Claude Code skill that turns one short request into a delivery-ready motion graphics video. 69 GitHub stars. Created 2026-10-08 (checked via GitHub API + README + LICENSE this run).
The playbook:
1. Install the skill
git clone github.com/revfactory/showre… ~/.claude/skills/showreel
2. Install deps (macOS shown in the README)
brew install ffmpeg node
python3 -m pip install numpy pillow
cd ~/.claude/skills/showreel/scripts && npm install
3. Add your ElevenLabs key (needed for narration)
echo 'export ELEVENLABS_API_KEY="sk_..."' >> ~/.zshrc && source ~/.zshrc
4. Run the preflight
python3 ~/.claude/skills/showreel/scripts/preflight.py
5. Open Claude Code in an empty folder and ask something like
"Make a 60-second promo for our product launch, cinematic, with narration."
What you get back: master MP4, share MP4, poster JPG, SRT, plus a production note with sources and regenerate commands.
Under the hood: concept candidates scored first, research with source URLs, Korean-first typography and narration rules (engine is language-neutral), HTML/Canvas/WebGL/three.js render, critique and rebuild weak segments, then QA for loudness, flashes and frozen frames.
Heads up: macOS is the tested platform. Music on ElevenLabs needs a paid plan. Don't invent facts. The skill wants a source URL for every claim on screen.
Money play: sell SaaS founders and local brands a fixed-price launch or teaser pack. You bring taste and the brief. Claude Code runs the pipeline.
Would you pitch this as a Korean-market promo first, or try an English launch teaser with the same engine?
2
161
You're still delivering the same CapCut look on every client Reel.
That's why coaches stop renewing after month one.
heditor by PycodeBR (github.com/pycodebr/heditor) is a free MIT Agent Skill for Claude Code and Codex. 108 GitHub stars. Created 2026-10-13 (checked via GitHub API + README + LICENSE this run).
It turns talking-head footage into vertical Reels or long-form YouTube with named styles, not one recycled template.
Forty-nine styles ship in the box: cinema, podcast, game, thriller, trailer, vox, documentario, keynote, plus paid-ad formats like lapidacao and copy-noir. Creative mode invents a visual world for that specific speech. Creative + reference builds a licence-safe homage (README example: "in the style of GTA 6").
Local pipeline: word-timed transcript, frame-accurate cuts, HyperFrames + GSAP motion graphics, voice-first mix at -14 LUFS, then lint, vision preview and loudness checks.
Install:
curl -fsSL raw.githubusercontent.com/py… | bash
Or:
npx skills add pycodebr/heditor -g
Needs ffmpeg, Node 22+, Python 3.10-3.13.
Money play: sell coaches a monthly named-style Reels pack. Same raw takes, different looks each week. CapCut sameness is what you're selling against.
Which style would you pitch first on a coaching Reel: podcast, cinema, or trailer?
You're still pasting Remotion docs into Claude and praying the animation lands.
That's why your motion graphics look like week-one React homework.
Remotion's own team ships free Agent Skills for Claude Code and Codex.
6,496 GitHub stars. One install:
npx skills add remotion-dev/skills
Then paste this:
/remotion-best-practices
Make a 20-second vertical promo for a local coaching brand.
Hook in the first 2 seconds, kinetic title, word-synced captions,
soft CTA end card, 1080x1920 for Instagram Reels.
Follow /remotion-markup rules. Preview in Studio, then /remotion-render.
Also in the pack: /remotion-create, /remotion-captions, /remotion-maps, /remotion-saas.
Sell monthly Remotion Reels packs to coaches without guessing the Remotion API again.
Which skill from remotion-dev/skills are you loading first?
1
3
285
