Aiming for X, it blew up on Instagram: how a talking-head editing skill got carried by "comment EDIT for the link" repost accounts

Ghost Editor is a Claude Code skill: you drop in raw phone talking-head footage, the kind where you redo one sentence a dozen times, and it cuts a 30-second reel with subtitles, motion graphics, and sound effects. It sat on GitHub for half a month at around eighty stars, then after Oct 28 climbed to 225 stars over 9 days. What carried it wasn't programmers; it was a batch of Instagram posts saying "5 skills that let Claude edit your video, comment EDIT and I'll DM you the link," the same list translated into a dozen-plus languages, with Ghost Editor at #1. Anyone who wants star growth for an open source project should study how it got onto that list; building another talking-head editor isn't worth much.

An ending cut from 14 takes

In the first GIF of the README, on the left is the author's raw footage, fast-forwarded 10x, with a counter in the corner ticking up, ending on the last line counted to "take 14." On the right is that same line cut into five styles. Below, a flowchart reads: 2 minutes 57 seconds of raw footage, 33 takes of speech; the discarded parts turn red and drop out, and the best take of each sentence slides into a 0:30 finished video.

The first line of the README is a before/after pitch:

「Three minutes of you stumbling through a script in. A 30-second reel that looks paid-for out.」(进去的是你磕磕绊绊念了三分钟稿子,出来的是一条看着像花钱请人剪的 30 秒 reel。)

The author, Magomed Kurbaitaev, is the founder of sports startup Gameplan and posts reels himself in Russian. On Sep 24 he posted a launch tweet on X, roughly: "Now that Opus 5.5 has gotten so strong, everyone is starting an AI editing business, so I've open sourced my editing skill," with 195 likes and 27k views; the same day he also posted a "paid editor's cut vs. LLM cut" side-by-side that got 29k views. These two posts brought the first 70 stars, and then growth stalled: from Oct 14 to 27 it averaged roughly half a star a day (GitHub daily star counts).

What happened on Oct 28

On Oct 28, several Instagram posts with nearly identical content appeared at the same time:

"5 skills that turn Claude into your video editor: 1. Ghost Editor (raw footage in, finished video out, picks the best take, subtitles that don't cover your face) 2. FFmpeg Skill 3. Higgsfield 4. Claude Shorts 5. OpenMontage… comment EDIT and I'll DM you all the links."

The Italian @ciak.sid.gira post got 44k plays, and its caption included kurbaitaev/ghost-editor directly; the Uzbek @akbarshokh_blog post got 65k plays and 2,122 comments; the English @david_knowsai posted "Comment 'Edit' and I'll send you the setup" the same day, with 267k plays and 10k+ comments, though whether the video actually covered this list can't be confirmed. Over the next week, Russian, Portuguese, German, Vietnamese, Hebrew, Persian, and Indonesian versions appeared one after another, at least 20 accounts posting; it later expanded into a "10 GitHub repos" version, and Ghost Editor was still at #1 (Instagram and TikTok search, 11-06).

These accounts are all AI-tool repost accounts running the "keyword in comments, auto-DM the link" playbook to farm comments and grow followers. The list is a content template they copy from one another. I found no evidence the author paid for promotion or had any relationship with these accounts. He himself wrote on X on Nov 2: "14.4K people saved my GitHub repo on Instagram; I only found out by chance while scrolling reels."

The star curve lines up with this (GitHub daily star counts):

The other two repos on the same list were also climbing at the time: kajisho5/ffmpeg-skill went from 10-24 to 2,060 stars, reaching 11-06 stars with 4,594, and AgriciDaniel/claude-shorts went from 10-28 to 242, reaching 318 (GitHub daily star counts).

Millions of impressions for 150 stars

The conversion here is astonishingly low: by the author's own account, millions of impressions, 14k saves, and over 6 thousand clones produced roughly 150 new stars (all of these figures are self-reported by the author and unverified). The reason is easy to guess: the people scrolling reels are creators, not GitHub users. They paste the install command into Claude Code as the post says, use it, and never come back to star it.

So for tools aimed at creators, star counts badly underestimate how many people are actually using them. If you want to build something similar, set your expectations right: Instagram gets you users, and stars are only a small part of that.

Why this one came out on top

There are plenty of libraries in the same niche: mariagorskikh/talking-head-reel, ranahaani/i-hate-editing, tenfoldmarc/video-edit-skill, kevinbadi/open-edits, all doing "talking-head raw footage → reel," launched around the same time or even earlier, with stars stuck between 58 and 121. None of them made the list; this one did. My inference is three reasons:

• It owns the first slot in the workflow. The list is built around "one skill does one job," ordered by workflow. The people who read these posts film their own reels, and the most painful step is turning botched raw footage into something publishable, so the tool that handles that step naturally ranks first.
• The README can serve directly as post copy. "Three minutes of raw footage in, 30-second finished video out" says it in one line; the "I paid an editor once" section is a side-by-side of a human edit and its edit; and there's a wall of GIFs playing 34 styles at once. Repost accounts can drop them straight into their own videos.
• The install is written for people who don't code. Installing means "paste this sentence into Claude Code," and usage is "Edit ~/Downloads/my-video.mov into a reel in Clean OS" or "Make my video look like this reel." Creators can follow it.

The product itself has things too: it auto-picks the best take for each sentence; 19 named styles (Clean OS, Seoul list, Kinetic doc…) plus 7 presets you can switch with one word; before delivery it checks pauses, volume, subtitle size, B-roll usage, and whether anything covers the face. The author says a recent batch took 3 hours 40 minutes on a Mac mini to cut 21 videos, and he caught 2 errors himself. But that's not why it's gotten more stars than similar tools; the features aren't much different from its peers.

The requirements are also in the README: a Claude paid subscription, an Apple silicon Mac, about 4 GB installed, and a 60-second reel takes 15 to 25 minutes on an M3 MacBook Air.

Who's competing for this topic

"Have a coding agent edit video" got crowded in the second half of 2026; a single awesome-claude-video-skills list alone collected 427 libraries.

I haven't seen anyone copy the Ghost Editor product itself. What got copied is that promotional list.

If it were you

The takeaway: Don't build another talking-head editor. Study how it got picked up by repost accounts. For creator-facing skills, the biggest distribution channel is the "comment for the link" accounts on Instagram, and few developers notice this path (inferred). That's the part you can copy. The trade-off is that star conversion is very low; you'd probably end up with a few hundred stars, not thousands.

Who it's for: Developers who make content themselves and have been stuck on a specific editing or production step; or people who already have a creator-facing skill that can be explained as a before/after in one sentence but nobody knows about it.

Who it's not for: Anyone hoping to earn stars just from the "raw talking-head footage → reel" topic; that slot already has five or six libraries, and someone on the list already holds it. Anyone wanting thousands of stars or trending: this channel's audience doesn't click stars. Anyone building developer tools: Instagram repost accounts won't pick you.

How to do it:

1. Use the list's pattern to find the open slots. Right now the list covers talking-head editing, FFmpeg toolboxes, motion graphics, long-to-short, and recreating a reel; adjacent slots nobody has claimed include cover images and thumbnails, multilingual dubbing and subtitle translation, turning a reel into a carousel, and batch-generating a week of topics and scripts (inferred; search Instagram before publishing).
2. Put a before/after sentence on the first README line, use left/right before-and-after GIFs as the first screen, then add a real-person edit next to your edit as a comparison video. These assets should be easy for repost accounts to lift directly.
3. Write the install as "paste this sentence into Claude Code," and write usage as plain instructions so readers never have to open a terminal.
4. End with a line like "Star it if it saved you an edit"; that's how Ghost Editor ends, and it helps recover a few more stars.

How to get distribution: Don't wait for them to find you. Search Instagram for "Claude skills," "comment EDIT," and "GitHub repos," list the accounts that post these roundups, and prioritize smaller non-English accounts (Italian, Russian, Portuguese, Uzbek, with tens of thousands of views). DM them your before/after video and a one-line pitch. They're constantly short on material, and the list gets copied and translated by other accounts, so you only need to get into one version. You can also post before/after videos on X yourself, but don't count on it: Ghost Editor's X launch only brought in 70 stars.

When to move: The "5 Claude editing skills" template 10-28 only appeared recently and is still being translated and expanded into 10- and 20-item versions. Once the list grows to twenty or thirty entries, new additions stop getting noticed, so it's best to get your project out before December (inferred).

Signs you've misread this: If you DM a dozen repost accounts and none of them add you to a list within two weeks, this channel only accepts names already on the list. Or if you do get listed and it gets tens of thousands of views but stars stay in the double digits, then this path brings users but not stars. If you want stars, go back to X and developer communities.