<?xml version="1.0" encoding="UTF-8"?>
<rss version="2.0" xmlns:atom="http://www.w3.org/2005/Atom" xmlns:dc="http://purl.org/dc/elements/1.1/">
  <channel>
    <title>DEV Community: Frank Heijdenrijk</title>
    <description>The latest articles on DEV Community by Frank Heijdenrijk (@fheijdenrijk).</description>
    <link>https://dev.to/fheijdenrijk</link>
    <image>
      <url>https://media2.dev.to/dynamic/image/width=90,height=90,fit=cover,gravity=auto,format=auto/https:%2F%2Fdev-to-uploads.s3.us-east-2.amazonaws.com%2Fuploads%2Fuser%2Fprofile_image%2F4211052%2F5dda1de1-d18c-4c23-8012-4bb79cbaad04.png</url>
      <title>DEV Community: Frank Heijdenrijk</title>
      <link>https://dev.to/fheijdenrijk</link>
    </image>
    <atom:link rel="self" type="application/rss+xml" href="https://dev.to/feed/fheijdenrijk"/>
    <language>en</language>
    <item>
      <title>AI Clipping With Claude Code: How to Automate Clipping From Campaign Brief to Checked, Disclosed Clips</title>
      <dc:creator>Frank Heijdenrijk</dc:creator>
      <pubDate>Sun, 15 Nov 2026 11:58:35 +0000</pubDate>
      <link>https://dev.to/fheijdenrijk/ai-clipping-with-claude-code-how-to-automate-clipping-from-campaign-brief-to-checked-disclosed-3i7c</link>
      <guid>https://dev.to/fheijdenrijk/ai-clipping-with-claude-code-how-to-automate-clipping-from-campaign-brief-to-checked-disclosed-3i7c</guid>
      <description>&lt;p&gt;AI clipping usually means one thing: a tool finds moments in a long video and cuts them vertical. In a paid clipping campaign that is the easy part. Clips get rejected for the rest of the job: a missing hashtag, the watermark the brief asked for, the wrong length, an account that isn't verified, a disclosure that isn't there, a link submitted 40 minutes after posting.&lt;/p&gt;

&lt;p&gt;This post shows how to automate clipping as the whole job, with AI Clipping Skill, a free, MIT-licensed skill for Claude Code (it also runs in Codex, Cursor and OpenClaw). Your agent reads the campaign brief into a checklist, cuts one disclosed clip per account from the footage the campaign supplied, checks every clip against the brief in code, shows one plan page, and after one yes posts to TikTok, Reels and Shorts and hands you each link with the minute its submission window closes. Everything before posting runs on your own computer with Python and ffmpeg and works without signing up for anything.&lt;br&gt;
&lt;/p&gt;

&lt;div class="highlight js-code-highlight"&gt;
&lt;pre class="highlight shell"&gt;&lt;code&gt;npx skills add social-media-skills/ai-clipping-skill -g -a claude-code -y
&lt;/code&gt;&lt;/pre&gt;

&lt;/div&gt;



&lt;p&gt;Repo: &lt;a href="https://github.com/social-media-skills/ai-clipping-skill" rel="noopener noreferrer"&gt;github.com/social-media-skills/ai-clipping-skill&lt;/a&gt;&lt;/p&gt;

&lt;p&gt;&lt;a href="https://media2.dev.to/dynamic/image/width=800%2Cheight=%2Cfit=scale-down%2Cgravity=auto%2Cformat=auto/https%3A%2F%2Fdev-to-uploads.s3.us-east-2.amazonaws.com%2Fuploads%2Farticles%2Fslq5dwp8cattbihow3ly.jpg" class="article-body-image-wrapper"&gt;&lt;img src="https://media2.dev.to/dynamic/image/width=800%2Cheight=%2Cfit=scale-down%2Cgravity=auto%2Cformat=auto/https%3A%2F%2Fdev-to-uploads.s3.us-east-2.amazonaws.com%2Fuploads%2Farticles%2Fslq5dwp8cattbihow3ly.jpg" alt="One 9-minute NASA interview on the left, an arrow, and three vertical clips for TikTok, Reels and Shorts on the right, each with &amp;quot;EXAMPLE ONLY: not a real ad&amp;quot; at the top, its own hook and captions" width="800" height="280"&gt;&lt;/a&gt;&lt;/p&gt;

&lt;p&gt;&lt;em&gt;The example in the repo: a made-up campaign on a 9-minute NASA interview, three clips for three accounts. Where a real campaign's clip shows its paid-partnership line, the example shows "EXAMPLE ONLY: not a real ad". NASA and Bob Behnken endorse nothing here.&lt;/em&gt;&lt;/p&gt;

&lt;h2&gt;
  
  
  What a clipping campaign asks for
&lt;/h2&gt;

&lt;p&gt;On Whop and Content Rewards, a campaign pays per 1,000 views for short clips of footage the brand or creator supplies. The brief sets the rules, and Content Rewards' own rejection guide says most rejections come from a detail the clipper skimmed past. A typical brief asks for:&lt;/p&gt;

&lt;ul&gt;
&lt;li&gt;
&lt;strong&gt;Money:&lt;/strong&gt; a rate per platform, a minimum payout per clip (nothing below it), a cap per clip, a fee, a total budget that can run out.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;The caption:&lt;/strong&gt; a hashtag, an @mention, a link, nothing from the banned list.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;The screen:&lt;/strong&gt; a watermark or logo, a length range, 9:16.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;The footage:&lt;/strong&gt; only the supplied files, sometimes only streams from the last few days.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;The account:&lt;/strong&gt; verified in the marketplace, a link in the bio, sometimes a minimum share of viewers in the US and an engagement floor.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;After posting:&lt;/strong&gt; submit the link within 30 minutes, one post into one campaign once, stay public and unchanged for 30 days, keep analytics screenshots as proof.&lt;/li&gt;
&lt;/ul&gt;

&lt;p&gt;And every campaign post is an ad. Content Rewards requires the disclosure at the start of the video and at the top of the caption, Whop's clipper terms require following the FTC's rules, and TikTok takes undisclosed branded posts out of For You.&lt;/p&gt;

&lt;p&gt;That list is why "automate clipping" can't stop at cutting. Each of those items is a check a script can run before the post exists.&lt;/p&gt;

&lt;h2&gt;
  
  
  The prompt
&lt;/h2&gt;

&lt;blockquote&gt;
&lt;p&gt;I joined a clipping campaign. Brief is in brief.txt, footage is behnken.mp4. Accounts: TikTok @example.clipper (everyone, comments on, duets and stitches off), Reels @example.clipper.reels, Shorts @exampleclipper (public, not for kids), all verified in the marketplace. No AI-made people or voices.&lt;/p&gt;
&lt;/blockquote&gt;

&lt;p&gt;That answers what the agent would otherwise ask in its first message: the brief, the footage, the accounts, whether each is verified, and the settings only the account owner may choose. When posting isn't connected yet, the agent puts the signup link at the top of that first message, so the account gets made while the clips render.&lt;/p&gt;

&lt;h2&gt;
  
  
  Step 1: the brief becomes a checklist
&lt;/h2&gt;

&lt;p&gt;The agent copies the brief's own numbers and words into &lt;code&gt;campaign.json&lt;/code&gt;, then:&lt;br&gt;
&lt;/p&gt;

&lt;div class="highlight js-code-highlight"&gt;
&lt;pre class="highlight shell"&gt;&lt;code&gt;python3 &lt;span class="o"&gt;&amp;lt;&lt;/span&gt;skill&lt;span class="o"&gt;&amp;gt;&lt;/span&gt;/scripts/campaign.py campaign.json
&lt;/code&gt;&lt;/pre&gt;

&lt;/div&gt;





&lt;div class="highlight js-code-highlight"&gt;
&lt;pre class="highlight plaintext"&gt;&lt;code&gt;1. **Platforms and rates:** TikTok $1.00 per 1,000 views, Instagram Reels $0.80 per 1,000 views, YouTube Shorts $0.50 per 1,000 views.
2. **Payout:** minimum payout per clip $5.00; cap per clip $150.00; marketplace fee 10%; total budget $3,000.00; $2,140.00 left when checked 16 Nov 13:30 UTC. When the budget runs out, views stop paying.
3. **Pays from:** TikTok 5,000, Instagram Reels 6,250, YouTube Shorts 10,000 views per clip; under that a clip earns nothing.
4. **Every caption:** starts with **#ad**; hashtags: #exampleorbit; mentions: @example.orbit; links: example.com/orbit.
...
16. **Submitting:** within 30 minutes of posting; one post goes into one campaign, once.
...
## Never followed
- "Please don't make it feel like an ad, we want it to look natural."

0 error(s), 1 brief sentence(s) flagged -&amp;gt; checklist.md
&lt;/code&gt;&lt;/pre&gt;

&lt;/div&gt;



&lt;p&gt;The example brief has 18 rules, and line 3 is the one most clippers never work out: at $1 per 1,000 views with a $5 minimum, a TikTok clip earns nothing until 5,000 views. The agent says that once, in plain words, before any cutting.&lt;/p&gt;

&lt;p&gt;The flagged sentence is the other point of this step. The brief asks for clips that don't feel like ads. The skill keeps the disclosure on every clip, tells you why (the marketplace, advertising law, and TikTok's reach rules all require it), and never follows that sentence. A brief that forbids any disclosure on screen is listed as a conflict: you ask the brand in writing or skip the campaign. You confirm the checklist before anything is cut.&lt;/p&gt;

&lt;h2&gt;
  
  
  Step 2: one edit per account
&lt;/h2&gt;



&lt;div class="highlight js-code-highlight"&gt;
&lt;pre class="highlight shell"&gt;&lt;code&gt;python3 &lt;span class="o"&gt;&amp;lt;&lt;/span&gt;skill&lt;span class="o"&gt;&amp;gt;&lt;/span&gt;/scripts/transcribe.py behnken.mp4 --engine &lt;span class="kd"&gt;local&lt;/span&gt; --out work
python3 &lt;span class="o"&gt;&amp;lt;&lt;/span&gt;skill&lt;span class="o"&gt;&amp;gt;&lt;/span&gt;/scripts/make_clips.py clips.json --transcript work/transcript.json --out clips
&lt;/code&gt;&lt;/pre&gt;

&lt;/div&gt;



&lt;p&gt;&lt;code&gt;transcribe.py&lt;/code&gt; gives a time for every word, with faster-whisper on your own computer, your own Groq or OpenAI key, or captions the campaign supplied. For a stream it also lists the loudest moments. The agent scores candidate moments for a hook, a payoff and whether they stand alone without the rest of the video, then picks one per account from across the footage and writes a hook for each, 3 to 8 words and true to the clip.&lt;/p&gt;

&lt;p&gt;&lt;code&gt;make_clips.py&lt;/code&gt; cuts each one to 9:16 with captions, the hook, the disclosure at the top for the first seconds (below TikTok's top tabs), and the brief's watermark text or logo. For a 16:9 stream with a facecam, &lt;code&gt;"reframe": "stack"&lt;/code&gt; puts the facecam over the gameplay.&lt;/p&gt;

&lt;p&gt;&lt;a href="https://media2.dev.to/dynamic/image/width=800%2Cheight=%2Cfit=scale-down%2Cgravity=auto%2Cformat=auto/https%3A%2F%2Fdev-to-uploads.s3.us-east-2.amazonaws.com%2Fuploads%2Farticles%2Fsxhjkf9i8r66nvceb095.jpg" class="article-body-image-wrapper"&gt;&lt;img src="https://media2.dev.to/dynamic/image/width=800%2Cheight=%2Cfit=scale-down%2Cgravity=auto%2Cformat=auto/https%3A%2F%2Fdev-to-uploads.s3.us-east-2.amazonaws.com%2Fuploads%2Farticles%2Fsxhjkf9i8r66nvceb095.jpg" alt="Three vertical clips from the NASA interview, each with &amp;quot;EXAMPLE ONLY: not a real ad&amp;quot; at the top, a hook in the middle, captions and the watermark &amp;quot;EXAMPLE ONLY&amp;quot;" width="800" height="474"&gt;&lt;/a&gt;&lt;/p&gt;

&lt;p&gt;Three accounts get three different moments: Groundhog Day on TikTok (29.6 s), "save the crew, not the mission" on Reels (20.8 s), and "future crews forget his name" on Shorts (26.7 s). That is the only honest way to post a campaign on several accounts. Campaigns reject "minimally changed" edits, and TikTok, Instagram and YouTube all limit reused or near-identical uploads. The skill never changes speed, mirrors or filters a clip to make one edit look like two.&lt;/p&gt;

&lt;h2&gt;
  
  
  Step 3: check every clip in code
&lt;/h2&gt;

&lt;p&gt;The agent writes a caption per account (&lt;code&gt;#ad&lt;/code&gt; first, what the clip says, then every required hashtag, @mention and link) into &lt;code&gt;clipjob.json&lt;/code&gt;, then:&lt;br&gt;
&lt;/p&gt;

&lt;div class="highlight js-code-highlight"&gt;
&lt;pre class="highlight shell"&gt;&lt;code&gt;python3 &lt;span class="o"&gt;&amp;lt;&lt;/span&gt;skill&lt;span class="o"&gt;&amp;gt;&lt;/span&gt;/scripts/check_clips.py clipjob.json
&lt;/code&gt;&lt;/pre&gt;

&lt;/div&gt;





&lt;div class="highlight js-code-highlight"&gt;
&lt;pre class="highlight plaintext"&gt;&lt;code&gt;groundhog-tiktok (TikTok @example.clipper, clips/clip-01.mp4, 29.6 s): ok
safe-path-reels (Reels @example.clipper.reels, clips/clip-02.mp4, 20.8 s): ok
curse-names-shorts (Shorts @exampleclipper, clips/clip-03.mp4, 26.7 s): ok
warning: clips: 3 clip(s) are 540x960, under 1080x1920: fine for a draft or an example; ...

0 error(s), 1 warning(s)
&lt;/code&gt;&lt;/pre&gt;

&lt;/div&gt;



&lt;p&gt;Per clip, it checks the file and length against its platform and the campaign, the caption (disclosure first, every required item, nothing banned, no engagement bait, Instagram's 5 hashtags), the disclosure and watermark on screen, that the footage is the campaign's and recent enough, that the account is verified, and that the budget isn't gone. Across clips, it refuses the same file on two accounts, two accounts on one platform sharing more than half their footage, and any edit already posted for this or another campaign. That last one works across folders: there is one ledger per user in &lt;code&gt;~/.ai-clipping/&lt;/code&gt;, holding a fingerprint of every file you posted.&lt;/p&gt;

&lt;p&gt;The warning is honest too: the example clips are 540x960 to keep the repo small. A real run cuts 1080x1920.&lt;/p&gt;

&lt;h2&gt;
  
  
  Step 4: one plan page, one yes
&lt;/h2&gt;



&lt;div class="highlight js-code-highlight"&gt;
&lt;pre class="highlight shell"&gt;&lt;code&gt;python3 &lt;span class="o"&gt;&amp;lt;&lt;/span&gt;skill&lt;span class="o"&gt;&amp;gt;&lt;/span&gt;/scripts/plan.py clipjob.json
&lt;/code&gt;&lt;/pre&gt;

&lt;/div&gt;



&lt;p&gt;&lt;a href="https://media2.dev.to/dynamic/image/width=800%2Cheight=%2Cfit=scale-down%2Cgravity=auto%2Cformat=auto/https%3A%2F%2Fdev-to-uploads.s3.us-east-2.amazonaws.com%2Fuploads%2Farticles%2F0dfvc945nbhrupydsi8t.png" class="article-body-image-wrapper"&gt;&lt;img src="https://media2.dev.to/dynamic/image/width=800%2Cheight=%2Cfit=scale-down%2Cgravity=auto%2Cformat=auto/https%3A%2F%2Fdev-to-uploads.s3.us-east-2.amazonaws.com%2Fuploads%2Farticles%2F0dfvc945nbhrupydsi8t.png" alt="The plan page for the made-up campaign: the campaign summary, an EXAMPLE ONLY banner, the flagged brief sentence, then a card each for TikTok, Instagram Reels and YouTube Shorts with the clip's first frame, hook, caption, settings, checks and expected earnings" width="800" height="1068"&gt;&lt;/a&gt;&lt;/p&gt;

&lt;p&gt;&lt;code&gt;plan.html&lt;/code&gt; opens with what to do before posting: set the bio link now, and for the audience and engagement rules, what actually helps (post from accounts whose viewers already live there, write hooks that pay off, answer comments in the first hour) next to what gets accounts banned (VPN tricks, engagement groups, bought views). Then a card per account: the clip's first frame, hook, caption, settings, checks, and what it earns. For the TikTok clip: $0.00 at 1,000 views, $9.00 at 10,000, $90.00 at 100,000, capped at $150. No promise, just the brief's math after the fee.&lt;/p&gt;

&lt;p&gt;The agent shows the page and waits for one clear yes that covers these clips, these accounts, now.&lt;/p&gt;

&lt;h2&gt;
  
  
  Step 5: post now, not later
&lt;/h2&gt;

&lt;p&gt;Campaigns accept a link only within 30 minutes of the post going live, so nothing is scheduled. After the yes, each clip is uploaded once, the request is validated, and the connected accounts post in one request with a child per account, each with its own clip and caption. Trimmed from &lt;code&gt;requests/example-orbit-crew-dragon.json&lt;/code&gt;:&lt;br&gt;
&lt;/p&gt;

&lt;div class="highlight js-code-highlight"&gt;
&lt;pre class="highlight json"&gt;&lt;code&gt;&lt;span class="p"&gt;{&lt;/span&gt;
  &lt;span class="nl"&gt;"schedule"&lt;/span&gt;&lt;span class="p"&gt;:&lt;/span&gt; &lt;span class="p"&gt;{&lt;/span&gt;&lt;span class="nl"&gt;"type"&lt;/span&gt;&lt;span class="p"&gt;:&lt;/span&gt; &lt;span class="s2"&gt;"PUBLISH_NOW"&lt;/span&gt;&lt;span class="p"&gt;}&lt;/span&gt;&lt;span class="p"&gt;,&lt;/span&gt;
  &lt;span class="nl"&gt;"socialAccounts"&lt;/span&gt;&lt;span class="p"&gt;:&lt;/span&gt; &lt;span class="p"&gt;[&lt;/span&gt;
    &lt;span class="p"&gt;{&lt;/span&gt;&lt;span class="nl"&gt;"platform"&lt;/span&gt;&lt;span class="p"&gt;:&lt;/span&gt; &lt;span class="s2"&gt;"TIKTOK"&lt;/span&gt;&lt;span class="p"&gt;,&lt;/span&gt; &lt;span class="nl"&gt;"postMode"&lt;/span&gt;&lt;span class="p"&gt;:&lt;/span&gt; &lt;span class="s2"&gt;"DIRECT_POST"&lt;/span&gt;&lt;span class="p"&gt;,&lt;/span&gt; &lt;span class="nl"&gt;"privacyLevel"&lt;/span&gt;&lt;span class="p"&gt;:&lt;/span&gt; &lt;span class="s2"&gt;"PUBLIC_TO_EVERYONE"&lt;/span&gt;&lt;span class="p"&gt;,&lt;/span&gt;
     &lt;span class="nl"&gt;"allowComment"&lt;/span&gt;&lt;span class="p"&gt;:&lt;/span&gt; &lt;span class="kc"&gt;true&lt;/span&gt;&lt;span class="p"&gt;,&lt;/span&gt; &lt;span class="nl"&gt;"allowDuet"&lt;/span&gt;&lt;span class="p"&gt;:&lt;/span&gt; &lt;span class="kc"&gt;false&lt;/span&gt;&lt;span class="p"&gt;,&lt;/span&gt; &lt;span class="nl"&gt;"allowStitch"&lt;/span&gt;&lt;span class="p"&gt;:&lt;/span&gt; &lt;span class="kc"&gt;false&lt;/span&gt;&lt;span class="p"&gt;,&lt;/span&gt;
     &lt;span class="nl"&gt;"isYourBrand"&lt;/span&gt;&lt;span class="p"&gt;:&lt;/span&gt; &lt;span class="kc"&gt;false&lt;/span&gt;&lt;span class="p"&gt;,&lt;/span&gt; &lt;span class="nl"&gt;"isBrandedContent"&lt;/span&gt;&lt;span class="p"&gt;:&lt;/span&gt; &lt;span class="kc"&gt;true&lt;/span&gt;&lt;span class="p"&gt;,&lt;/span&gt; &lt;span class="nl"&gt;"isAiGeneratedContent"&lt;/span&gt;&lt;span class="p"&gt;:&lt;/span&gt; &lt;span class="kc"&gt;false&lt;/span&gt;&lt;span class="p"&gt;,&lt;/span&gt;
     &lt;span class="nl"&gt;"contentOverride"&lt;/span&gt;&lt;span class="p"&gt;:&lt;/span&gt; &lt;span class="p"&gt;[&lt;/span&gt;&lt;span class="p"&gt;{&lt;/span&gt;&lt;span class="nl"&gt;"text"&lt;/span&gt;&lt;span class="p"&gt;:&lt;/span&gt; &lt;span class="s2"&gt;"#ad Astronaut Bob Behnken trains until launch day holds no surprises, ..."&lt;/span&gt;&lt;span class="p"&gt;,&lt;/span&gt;
                          &lt;span class="nl"&gt;"media"&lt;/span&gt;&lt;span class="p"&gt;:&lt;/span&gt; &lt;span class="p"&gt;[&lt;/span&gt;&lt;span class="p"&gt;{&lt;/span&gt;&lt;span class="nl"&gt;"type"&lt;/span&gt;&lt;span class="p"&gt;:&lt;/span&gt; &lt;span class="s2"&gt;"MEDIA_LIBRARY"&lt;/span&gt;&lt;span class="p"&gt;,&lt;/span&gt; &lt;span class="nl"&gt;"mediaId"&lt;/span&gt;&lt;span class="p"&gt;:&lt;/span&gt; &lt;span class="s2"&gt;"&amp;lt;clip-01&amp;gt;"&lt;/span&gt;&lt;span class="p"&gt;}&lt;/span&gt;&lt;span class="p"&gt;]&lt;/span&gt;&lt;span class="p"&gt;}&lt;/span&gt;&lt;span class="p"&gt;]&lt;/span&gt;&lt;span class="p"&gt;}&lt;/span&gt;&lt;span class="p"&gt;,&lt;/span&gt;
    &lt;span class="p"&gt;{&lt;/span&gt;&lt;span class="nl"&gt;"platform"&lt;/span&gt;&lt;span class="p"&gt;:&lt;/span&gt; &lt;span class="s2"&gt;"INSTAGRAM"&lt;/span&gt;&lt;span class="p"&gt;,&lt;/span&gt; &lt;span class="nl"&gt;"postType"&lt;/span&gt;&lt;span class="p"&gt;:&lt;/span&gt; &lt;span class="s2"&gt;"REEL"&lt;/span&gt;&lt;span class="p"&gt;,&lt;/span&gt; &lt;span class="nl"&gt;"isAiGeneratedContent"&lt;/span&gt;&lt;span class="p"&gt;:&lt;/span&gt; &lt;span class="kc"&gt;false&lt;/span&gt;&lt;span class="p"&gt;,&lt;/span&gt;
     &lt;span class="nl"&gt;"contentOverride"&lt;/span&gt;&lt;span class="p"&gt;:&lt;/span&gt; &lt;span class="p"&gt;[&lt;/span&gt;&lt;span class="p"&gt;{&lt;/span&gt;&lt;span class="nl"&gt;"text"&lt;/span&gt;&lt;span class="p"&gt;:&lt;/span&gt; &lt;span class="s2"&gt;"#ad \"Save the crew, not the mission.\" ..."&lt;/span&gt;&lt;span class="p"&gt;,&lt;/span&gt;
                          &lt;span class="nl"&gt;"media"&lt;/span&gt;&lt;span class="p"&gt;:&lt;/span&gt; &lt;span class="p"&gt;[&lt;/span&gt;&lt;span class="p"&gt;{&lt;/span&gt;&lt;span class="nl"&gt;"type"&lt;/span&gt;&lt;span class="p"&gt;:&lt;/span&gt; &lt;span class="s2"&gt;"MEDIA_LIBRARY"&lt;/span&gt;&lt;span class="p"&gt;,&lt;/span&gt; &lt;span class="nl"&gt;"mediaId"&lt;/span&gt;&lt;span class="p"&gt;:&lt;/span&gt; &lt;span class="s2"&gt;"&amp;lt;clip-02&amp;gt;"&lt;/span&gt;&lt;span class="p"&gt;}&lt;/span&gt;&lt;span class="p"&gt;]&lt;/span&gt;&lt;span class="p"&gt;}&lt;/span&gt;&lt;span class="p"&gt;]&lt;/span&gt;&lt;span class="p"&gt;}&lt;/span&gt;
  &lt;span class="p"&gt;]&lt;/span&gt;
&lt;span class="p"&gt;}&lt;/span&gt;
&lt;/code&gt;&lt;/pre&gt;

&lt;/div&gt;



&lt;p&gt;&lt;code&gt;isBrandedContent: true&lt;/code&gt; is TikTok's paid partnership label. Instagram's paid partnership label and YouTube's paid promotion box can't be set through the posting API, so they are steps in the app, listed on the plan and on the submission sheet.&lt;/p&gt;

&lt;p&gt;In the example the clipper is on the free plan with 2 accounts, so TikTok and Reels go through the request and Shorts is posted by hand from &lt;code&gt;by-hand/curse-names-shorts.md&lt;/code&gt;: the file, title, description, settings, the paid promotion box, and the 30-minute steps. Pasting the Short's link back records it like the others.&lt;/p&gt;

&lt;h2&gt;
  
  
  Step 6: the submission sheet
&lt;/h2&gt;

&lt;p&gt;After posting, &lt;code&gt;plan.py --status&lt;/code&gt; reads the live links back and writes &lt;code&gt;submit.md&lt;/code&gt;:&lt;br&gt;
&lt;/p&gt;

&lt;div class="highlight js-code-highlight"&gt;
&lt;pre class="highlight plaintext"&gt;&lt;code&gt;1. **TikTok @example.clipper** (TikTok), post `groundhog-tiktok`
   - Link: https://www.tiktok.com/@example.clipper/video/example-0001
   - Live since Mon 16 Nov 09:01 America/New_York (EST). Submit by **Mon 16 Nov 09:31 America/New_York (EST)** (26 min left).
2. **Reels @example.clipper.reels** (Instagram Reels), post `safe-path-reels`
   - Live since Mon 16 Nov 09:02 ... Submit by **Mon 16 Nov 09:32 ...**
   - Before then: Instagram: edit the Reel in the app and turn on the paid partnership label.
&lt;/code&gt;&lt;/pre&gt;

&lt;/div&gt;



&lt;p&gt;A deadline in your own time zone, per post, with the app steps that must happen inside it, and the times for the 48-hour, 72-hour and 7-day analytics screenshots. You paste each link into the campaign yourself. No marketplace has a submission API for clippers, and the skill never drives your logged-in browser.&lt;/p&gt;

&lt;h2&gt;
  
  
  Step 7: the ledger and the proof log
&lt;/h2&gt;



&lt;div class="highlight js-code-highlight"&gt;
&lt;pre class="highlight shell"&gt;&lt;code&gt;python3 &lt;span class="o"&gt;&amp;lt;&lt;/span&gt;skill&lt;span class="o"&gt;&amp;gt;&lt;/span&gt;/scripts/ledger.py --submitted &lt;span class="o"&gt;&amp;lt;&lt;/span&gt;link&lt;span class="o"&gt;&amp;gt;&lt;/span&gt; --campaign example-orbit-crew-dragon
python3 &lt;span class="o"&gt;&amp;lt;&lt;/span&gt;skill&lt;span class="o"&gt;&amp;gt;&lt;/span&gt;/scripts/ledger.py --proof analytics.json
python3 &lt;span class="o"&gt;&amp;lt;&lt;/span&gt;skill&lt;span class="o"&gt;&amp;gt;&lt;/span&gt;/scripts/ledger.py --earnings
&lt;/code&gt;&lt;/pre&gt;

&lt;/div&gt;



&lt;p&gt;&lt;code&gt;--submitted&lt;/code&gt; refuses a link already submitted, a share link, and a post made for another campaign. &lt;code&gt;--proof&lt;/code&gt; adds a dated row per post to one proof log, from the per-post numbers the posting service collects about once a day. &lt;code&gt;--earnings&lt;/code&gt; applies the minimum, cap and fee, with the 30-day earning window counted from the day a clip crosses the payout threshold. From the example's made-up day-3 numbers:&lt;br&gt;
&lt;/p&gt;

&lt;div class="highlight js-code-highlight"&gt;
&lt;pre class="highlight plaintext"&gt;&lt;code&gt;TikTok @example.clipper: 7,420 views, est. $6.68 after 10% fee, 30 day(s) left, submitted
Reels @example.clipper.reels (Instagram Reels): 2,380 views, est. $0.00 after 10% fee (under the $5.00 minimum payout) ...
  audience: 22% in the US (from your screenshot); the campaign needs 30%, so expect this clip to be rejected
Shorts @exampleclipper (YouTube Shorts): 1,120 views, est. $0.00 after 10% fee (under the $5.00 minimum payout) ...
submitted total: 10,920 views, est. $6.68
&lt;/code&gt;&lt;/pre&gt;

&lt;/div&gt;



&lt;p&gt;That is what most clippers see: a few thousand views, a few dollars at most. The value of the ledger is not that number. It is knowing which clip is under the minimum, which one breaks the audience rule, and having the dated record when a payout is checked.&lt;/p&gt;

&lt;h2&gt;
  
  
  What stays manual, on purpose
&lt;/h2&gt;

&lt;ul&gt;
&lt;li&gt;
&lt;strong&gt;Submitting.&lt;/strong&gt; You paste each link. The sheet tells you by when.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Instagram's paid partnership label and YouTube's paid promotion box.&lt;/strong&gt; Set in the app inside the 30 minutes.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;The bio link,&lt;/strong&gt; set before posting, and &lt;strong&gt;a required TikTok sound,&lt;/strong&gt; which no posting API can add: that clip is posted by hand with the sound.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Content rules a script can't judge:&lt;/strong&gt; topics, tone, what to say about the product. They are on the plan page for you to read before the yes.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;The owner's choices:&lt;/strong&gt; who can see a TikTok, made for kids, AI labels, marketplace verification. Asked, never filled in.&lt;/li&gt;
&lt;/ul&gt;

&lt;h2&gt;
  
  
  Posting, and what it costs
&lt;/h2&gt;

&lt;p&gt;Disclosure: the skill is made by &lt;a href="https://woopsocial.com/?utm_source=ai-clipping-skill&amp;amp;utm_medium=devto" rel="noopener noreferrer"&gt;the team that builds WoopSocial&lt;/a&gt;, and posting goes through WoopSocial. It posts through each platform's own API with its own approved apps, so you don't need a developer app or an app review of your own, and it returns each post's link and per-post numbers. It doesn't write captions, edit video, set Instagram's or YouTube's paid-partnership options, or submit to campaigns: the agent and the skill's scripts do the first two, and you do the rest. The free plan covers 2 social accounts, no card (&lt;a href="https://app.woopsocial.com/signup?utm_source=ai-clipping-skill&amp;amp;utm_medium=devto" rel="noopener noreferrer"&gt;make a free account&lt;/a&gt;). TikTok, Reels and Shorts are 3, so the agent asks up front: it posts 2 and gives you a ready page for the third, posts all of them on a paid plan, or you post all of them by hand with nothing to sign up for. The connect step for your agent is on &lt;a href="https://woopsocial.com/agents?utm_source=ai-clipping-skill&amp;amp;utm_medium=devto" rel="noopener noreferrer"&gt;WoopSocial's agent setup page&lt;/a&gt;.&lt;/p&gt;

&lt;h2&gt;
  
  
  Try it
&lt;/h2&gt;



&lt;div class="highlight js-code-highlight"&gt;
&lt;pre class="highlight shell"&gt;&lt;code&gt;npx skills add social-media-skills/ai-clipping-skill -g -a claude-code -y
&lt;/code&gt;&lt;/pre&gt;

&lt;/div&gt;



&lt;p&gt;Then tell your agent "I joined a clipping campaign" and paste the brief. It needs Python 3.9+, ffmpeg and Pillow; &lt;code&gt;scripts/check_setup.py&lt;/code&gt; prints the install command for anything missing. The whole example, every file the skill made, is in &lt;a href="https://github.com/social-media-skills/ai-clipping-skill/tree/main/skills/ai-clipping-skill/examples/orbit-campaign" rel="noopener noreferrer"&gt;&lt;code&gt;examples/orbit-campaign/&lt;/code&gt;&lt;/a&gt;. Whop and Content Rewards are named only as the marketplaces where these campaigns run; the skill isn't affiliated with either. Issues and pull requests are welcome, and there's &lt;a href="https://social-media-skills.com/?utm_source=ai-clipping-skill&amp;amp;utm_medium=devto" rel="noopener noreferrer"&gt;the rest of our free social media skills&lt;/a&gt; from the same team.&lt;/p&gt;

</description>
      <category>claudecode</category>
      <category>python</category>
      <category>video</category>
      <category>showdev</category>
    </item>
    <item>
      <title>Cross Post Video to TikTok, Reels and Shorts With Claude Code: Check the File, Write Three Texts, Send One Request</title>
      <dc:creator>Frank Heijdenrijk</dc:creator>
      <pubDate>Thu, 12 Nov 2026 19:59:50 +0000</pubDate>
      <link>https://dev.to/fheijdenrijk/cross-post-video-to-tiktok-reels-and-shorts-with-claude-code-check-the-file-write-three-texts-b30</link>
      <guid>https://dev.to/fheijdenrijk/cross-post-video-to-tiktok-reels-and-shorts-with-claude-code-check-the-file-write-three-texts-b30</guid>
      <description>&lt;p&gt;Cross post video to TikTok, Reels and Shorts from your agent: Cross Post Video Skill is a free, MIT-licensed skill for Claude Code that takes one finished video and gets it ready for all three. It checks the file against every platform before anything is uploaded, helps you pick a cover, has the agent write a separate text for each platform, asks the settings only the account owner may choose, and shows every post on one plan page. After one clear yes, it sends all accounts in one request with a child per account. It also works in Codex, Cursor and OpenClaw.&lt;/p&gt;

&lt;p&gt;Everything up to the upload runs on your own computer with Python and ffmpeg, and works without signing up for anything. This post walks through the NASA example in the repo, one command at a time.&lt;br&gt;
&lt;/p&gt;

&lt;div class="highlight js-code-highlight"&gt;
&lt;pre class="highlight shell"&gt;&lt;code&gt;npx skills add social-media-skills/cross-post-video-skill -g -a claude-code -y
&lt;/code&gt;&lt;/pre&gt;

&lt;/div&gt;



&lt;p&gt;Repo: &lt;a href="https://github.com/social-media-skills/cross-post-video-skill" rel="noopener noreferrer"&gt;github.com/social-media-skills/cross-post-video-skill&lt;/a&gt;&lt;/p&gt;

&lt;p&gt;&lt;a href="https://media2.dev.to/dynamic/image/width=800%2Cheight=%2Cfit=scale-down%2Cgravity=auto%2Cformat=auto/https%3A%2F%2Fdev-to-uploads.s3.us-east-2.amazonaws.com%2Fuploads%2Farticles%2F2muca9mk8pm1v2ww0vur.png" class="article-body-image-wrapper"&gt;&lt;img src="https://media2.dev.to/dynamic/image/width=800%2Cheight=%2Cfit=scale-down%2Cgravity=auto%2Cformat=auto/https%3A%2F%2Fdev-to-uploads.s3.us-east-2.amazonaws.com%2Fuploads%2Farticles%2F2muca9mk8pm1v2ww0vur.png" alt="The plan page for one NASA clip: the cover, then a card each for TikTok, Instagram Reels and YouTube Shorts with its own text, character count and settings" width="800" height="475"&gt;&lt;/a&gt;&lt;/p&gt;

&lt;p&gt;&lt;em&gt;The plan page from the example: a 14-second NASA clip about why astronauts exercise in space, planned for TikTok, Reels and Shorts (NASA footage, no endorsement implied).&lt;/em&gt;&lt;/p&gt;

&lt;h2&gt;
  
  
  Why one video breaks on TikTok, Reels or Shorts
&lt;/h2&gt;

&lt;p&gt;One vertical MP4 can be valid on all three platforms. What differs is everything around it, and each of these is easy to find only after the upload:&lt;/p&gt;

&lt;ul&gt;
&lt;li&gt;
&lt;strong&gt;Length.&lt;/strong&gt; TikTok's maximum belongs to the account, not the platform. A Short is square or vertical and up to 3 minutes; longer and YouTube makes it a normal video.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;The file itself.&lt;/strong&gt; Instagram is the strictest: 300 MB, up to 1920 px wide, 23 to 60 fps, AAC sound, the index at the start of the file, and no HDR.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;The text.&lt;/strong&gt; TikTok and Instagram take a caption. YouTube wants a title of up to 100 characters and counts the description in bytes. Instagram caps hashtags at 5.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;The settings.&lt;/strong&gt; TikTok makes the account owner choose who can see each post and whether comments, duets and stitches are on, with no defaults. YouTube asks whether a video is made for kids.&lt;/li&gt;
&lt;/ul&gt;

&lt;p&gt;The skill moves all of that in front of the upload. Scripts do the checking and counting; the agent writes the texts and asks the questions.&lt;/p&gt;

&lt;h2&gt;
  
  
  The prompt
&lt;/h2&gt;

&lt;blockquote&gt;
&lt;p&gt;Post clips/microgravity.mp4 to my TikTok, Reels and Shorts on 12 January at 6 pm New York time. It's my own video with the original sound, no AI, not promotional. TikTok: everyone can see it, comments on, duets and stitches off. Not made for kids.&lt;/p&gt;
&lt;/blockquote&gt;

&lt;p&gt;That prompt already answers what the agent would otherwise ask in its first message: is this the original export you have the right to post, what the sound is, whether it is AI-made or promotional, and the settings per account.&lt;/p&gt;

&lt;h2&gt;
  
  
  Step 1: check the file
&lt;/h2&gt;



&lt;div class="highlight js-code-highlight"&gt;
&lt;pre class="highlight shell"&gt;&lt;code&gt;python3 &lt;span class="o"&gt;&amp;lt;&lt;/span&gt;skill&lt;span class="o"&gt;&amp;gt;&lt;/span&gt;/scripts/check_video.py clips/microgravity.mp4
&lt;/code&gt;&lt;/pre&gt;

&lt;/div&gt;





&lt;div class="highlight js-code-highlight"&gt;
&lt;pre class="highlight plaintext"&gt;&lt;code&gt;clips/microgravity.mp4: 720x1280, 13.9 s, 29.97 fps, h264 yuv420p, aac 48000 Hz, 3.1 MB, 1.82 Mbps
  TikTok: ok
  Instagram Reels: ok
  YouTube Shorts: ok
&lt;/code&gt;&lt;/pre&gt;

&lt;/div&gt;



&lt;p&gt;This clip passes. When a file doesn't (4K, 120 fps, 5.1 sound, a MOV from a camera, HDR from a newer iPhone), &lt;code&gt;--fix&lt;/code&gt; writes a re-encoded copy next to it as &lt;code&gt;&amp;lt;name&amp;gt;.crosspost.mp4&lt;/code&gt; and leaves the original alone. It keeps the picture size and only scales down a video over 1920 px wide. HDR is converted to standard colour when your ffmpeg has the &lt;code&gt;zscale&lt;/code&gt; filter; when it doesn't, &lt;code&gt;--fix&lt;/code&gt; writes nothing and tells you how to export a standard-colour file instead. It never trims, crops, pads or speeds up a video: length and shape stay your call.&lt;/p&gt;

&lt;h2&gt;
  
  
  Step 2: pick a cover
&lt;/h2&gt;

&lt;p&gt;Without a cover, TikTok and Instagram show the first frame. &lt;code&gt;cover.py&lt;/code&gt; makes a sheet of 12 frames with their times, with a dashed box on each for the middle 3:4, the part Instagram's profile grid shows.&lt;br&gt;
&lt;/p&gt;

&lt;div class="highlight js-code-highlight"&gt;
&lt;pre class="highlight shell"&gt;&lt;code&gt;python3 &lt;span class="o"&gt;&amp;lt;&lt;/span&gt;skill&lt;span class="o"&gt;&amp;gt;&lt;/span&gt;/scripts/cover.py clips/microgravity.mp4
&lt;/code&gt;&lt;/pre&gt;

&lt;/div&gt;



&lt;p&gt;&lt;a href="https://media2.dev.to/dynamic/image/width=800%2Cheight=%2Cfit=scale-down%2Cgravity=auto%2Cformat=auto/https%3A%2F%2Fdev-to-uploads.s3.us-east-2.amazonaws.com%2Fuploads%2Farticles%2Foy6cvphfhcobzzgp7w7b.jpg" class="article-body-image-wrapper"&gt;&lt;img src="https://media2.dev.to/dynamic/image/width=800%2Cheight=%2Cfit=scale-down%2Cgravity=auto%2Cformat=auto/https%3A%2F%2Fdev-to-uploads.s3.us-east-2.amazonaws.com%2Fuploads%2Farticles%2Foy6cvphfhcobzzgp7w7b.jpg" alt="Twelve frames of the clip with their times; a dashed box on each marks the middle 3:4 that Instagram's profile grid shows" width="800" height="493"&gt;&lt;/a&gt;&lt;/p&gt;

&lt;p&gt;In the example, the frame at 8.7 seconds looked good full screen and cut the astronaut's head off in the grid crop. The frame at 12.2 seconds didn't:&lt;br&gt;
&lt;/p&gt;

&lt;div class="highlight js-code-highlight"&gt;
&lt;pre class="highlight shell"&gt;&lt;code&gt;python3 &lt;span class="o"&gt;&amp;lt;&lt;/span&gt;skill&lt;span class="o"&gt;&amp;gt;&lt;/span&gt;/scripts/cover.py clips/microgravity.mp4 --at 12.2
&lt;/code&gt;&lt;/pre&gt;

&lt;/div&gt;



&lt;p&gt;&lt;a href="https://media2.dev.to/dynamic/image/width=800%2Cheight=%2Cfit=scale-down%2Cgravity=auto%2Cformat=auto/https%3A%2F%2Fdev-to-uploads.s3.us-east-2.amazonaws.com%2Fuploads%2Farticles%2Fay7offij45a1qsw2f118.jpg" class="article-body-image-wrapper"&gt;&lt;img src="https://media2.dev.to/dynamic/image/width=800%2Cheight=%2Cfit=scale-down%2Cgravity=auto%2Cformat=auto/https%3A%2F%2Fdev-to-uploads.s3.us-east-2.amazonaws.com%2Fuploads%2Farticles%2Fay7offij45a1qsw2f118.jpg" alt="The chosen cover full screen, and next to it the 3:4 crop the profile grid shows" width="768" height="702"&gt;&lt;/a&gt;&lt;/p&gt;

&lt;p&gt;The cover goes to TikTok and Instagram with the post. YouTube's cover is set in the YouTube app, so the plan and the by-hand page both say "Cover frame: 12.2 s" and you pick the same frame there.&lt;/p&gt;

&lt;h2&gt;
  
  
  Step 3: a text per platform, in one job file
&lt;/h2&gt;

&lt;p&gt;The agent writes each text from what the video says, not from the other texts, and puts everything in &lt;code&gt;crosspost.json&lt;/code&gt;: the accounts with their settings, the four answers, and the texts. Trimmed:&lt;br&gt;
&lt;/p&gt;

&lt;div class="highlight js-code-highlight"&gt;
&lt;pre class="highlight json"&gt;&lt;code&gt;&lt;span class="p"&gt;{&lt;/span&gt;
  &lt;span class="nl"&gt;"timezone"&lt;/span&gt;&lt;span class="p"&gt;:&lt;/span&gt; &lt;span class="s2"&gt;"America/New_York"&lt;/span&gt;&lt;span class="p"&gt;,&lt;/span&gt;
  &lt;span class="nl"&gt;"accounts"&lt;/span&gt;&lt;span class="p"&gt;:&lt;/span&gt; &lt;span class="p"&gt;[&lt;/span&gt;
    &lt;span class="p"&gt;{&lt;/span&gt;&lt;span class="nl"&gt;"name"&lt;/span&gt;&lt;span class="p"&gt;:&lt;/span&gt; &lt;span class="s2"&gt;"TikTok"&lt;/span&gt;&lt;span class="p"&gt;,&lt;/span&gt; &lt;span class="nl"&gt;"platform"&lt;/span&gt;&lt;span class="p"&gt;:&lt;/span&gt; &lt;span class="s2"&gt;"TIKTOK"&lt;/span&gt;&lt;span class="p"&gt;,&lt;/span&gt; &lt;span class="nl"&gt;"privacyLevel"&lt;/span&gt;&lt;span class="p"&gt;:&lt;/span&gt; &lt;span class="s2"&gt;"PUBLIC_TO_EVERYONE"&lt;/span&gt;&lt;span class="p"&gt;,&lt;/span&gt;
     &lt;span class="nl"&gt;"allowComment"&lt;/span&gt;&lt;span class="p"&gt;:&lt;/span&gt; &lt;span class="kc"&gt;true&lt;/span&gt;&lt;span class="p"&gt;,&lt;/span&gt; &lt;span class="nl"&gt;"allowDuet"&lt;/span&gt;&lt;span class="p"&gt;:&lt;/span&gt; &lt;span class="kc"&gt;false&lt;/span&gt;&lt;span class="p"&gt;,&lt;/span&gt; &lt;span class="nl"&gt;"allowStitch"&lt;/span&gt;&lt;span class="p"&gt;:&lt;/span&gt; &lt;span class="kc"&gt;false&lt;/span&gt;&lt;span class="p"&gt;,&lt;/span&gt; &lt;span class="nl"&gt;"max_video_s"&lt;/span&gt;&lt;span class="p"&gt;:&lt;/span&gt; &lt;span class="mi"&gt;600&lt;/span&gt;&lt;span class="p"&gt;}&lt;/span&gt;&lt;span class="p"&gt;,&lt;/span&gt;
    &lt;span class="p"&gt;{&lt;/span&gt;&lt;span class="nl"&gt;"name"&lt;/span&gt;&lt;span class="p"&gt;:&lt;/span&gt; &lt;span class="s2"&gt;"Reels"&lt;/span&gt;&lt;span class="p"&gt;,&lt;/span&gt; &lt;span class="nl"&gt;"platform"&lt;/span&gt;&lt;span class="p"&gt;:&lt;/span&gt; &lt;span class="s2"&gt;"INSTAGRAM"&lt;/span&gt;&lt;span class="p"&gt;}&lt;/span&gt;&lt;span class="p"&gt;,&lt;/span&gt;
    &lt;span class="p"&gt;{&lt;/span&gt;&lt;span class="nl"&gt;"name"&lt;/span&gt;&lt;span class="p"&gt;:&lt;/span&gt; &lt;span class="s2"&gt;"Shorts"&lt;/span&gt;&lt;span class="p"&gt;,&lt;/span&gt; &lt;span class="nl"&gt;"platform"&lt;/span&gt;&lt;span class="p"&gt;:&lt;/span&gt; &lt;span class="s2"&gt;"YOUTUBE"&lt;/span&gt;&lt;span class="p"&gt;,&lt;/span&gt; &lt;span class="nl"&gt;"privacy"&lt;/span&gt;&lt;span class="p"&gt;:&lt;/span&gt; &lt;span class="s2"&gt;"public"&lt;/span&gt;&lt;span class="p"&gt;,&lt;/span&gt; &lt;span class="nl"&gt;"madeForKids"&lt;/span&gt;&lt;span class="p"&gt;:&lt;/span&gt; &lt;span class="kc"&gt;false&lt;/span&gt;&lt;span class="p"&gt;}&lt;/span&gt;
  &lt;span class="p"&gt;]&lt;/span&gt;&lt;span class="p"&gt;,&lt;/span&gt;
  &lt;span class="nl"&gt;"videos"&lt;/span&gt;&lt;span class="p"&gt;:&lt;/span&gt; &lt;span class="p"&gt;[&lt;/span&gt;&lt;span class="p"&gt;{&lt;/span&gt;
    &lt;span class="nl"&gt;"id"&lt;/span&gt;&lt;span class="p"&gt;:&lt;/span&gt; &lt;span class="s2"&gt;"microgravity"&lt;/span&gt;&lt;span class="p"&gt;,&lt;/span&gt; &lt;span class="nl"&gt;"file"&lt;/span&gt;&lt;span class="p"&gt;:&lt;/span&gt; &lt;span class="s2"&gt;"clips/microgravity.mp4"&lt;/span&gt;&lt;span class="p"&gt;,&lt;/span&gt; &lt;span class="nl"&gt;"when"&lt;/span&gt;&lt;span class="p"&gt;:&lt;/span&gt; &lt;span class="s2"&gt;"2027-01-12 18:00"&lt;/span&gt;&lt;span class="p"&gt;,&lt;/span&gt;
    &lt;span class="nl"&gt;"own_video"&lt;/span&gt;&lt;span class="p"&gt;:&lt;/span&gt; &lt;span class="kc"&gt;true&lt;/span&gt;&lt;span class="p"&gt;,&lt;/span&gt; &lt;span class="nl"&gt;"audio"&lt;/span&gt;&lt;span class="p"&gt;:&lt;/span&gt; &lt;span class="s2"&gt;"original"&lt;/span&gt;&lt;span class="p"&gt;,&lt;/span&gt; &lt;span class="nl"&gt;"ai_generated"&lt;/span&gt;&lt;span class="p"&gt;:&lt;/span&gt; &lt;span class="kc"&gt;false&lt;/span&gt;&lt;span class="p"&gt;,&lt;/span&gt; &lt;span class="nl"&gt;"promotional"&lt;/span&gt;&lt;span class="p"&gt;:&lt;/span&gt; &lt;span class="s2"&gt;"no"&lt;/span&gt;&lt;span class="p"&gt;,&lt;/span&gt;
    &lt;span class="nl"&gt;"text"&lt;/span&gt;&lt;span class="p"&gt;:&lt;/span&gt; &lt;span class="p"&gt;{&lt;/span&gt;
      &lt;span class="nl"&gt;"TIKTOK"&lt;/span&gt;&lt;span class="p"&gt;:&lt;/span&gt; &lt;span class="s2"&gt;"Why astronauts work out for hours every day\n..."&lt;/span&gt;&lt;span class="p"&gt;,&lt;/span&gt;
      &lt;span class="nl"&gt;"INSTAGRAM"&lt;/span&gt;&lt;span class="p"&gt;:&lt;/span&gt; &lt;span class="s2"&gt;"Astronauts on the space station exercise for hours every day.\n..."&lt;/span&gt;&lt;span class="p"&gt;,&lt;/span&gt;
      &lt;span class="nl"&gt;"YOUTUBE"&lt;/span&gt;&lt;span class="p"&gt;:&lt;/span&gt; &lt;span class="p"&gt;{&lt;/span&gt;&lt;span class="nl"&gt;"title"&lt;/span&gt;&lt;span class="p"&gt;:&lt;/span&gt; &lt;span class="s2"&gt;"Why astronauts exercise for hours every day in space"&lt;/span&gt;&lt;span class="p"&gt;,&lt;/span&gt;
                  &lt;span class="nl"&gt;"description"&lt;/span&gt;&lt;span class="p"&gt;:&lt;/span&gt; &lt;span class="s2"&gt;"..."&lt;/span&gt;&lt;span class="p"&gt;,&lt;/span&gt; &lt;span class="nl"&gt;"tags"&lt;/span&gt;&lt;span class="p"&gt;:&lt;/span&gt; &lt;span class="p"&gt;[&lt;/span&gt;&lt;span class="s2"&gt;"space station"&lt;/span&gt;&lt;span class="p"&gt;,&lt;/span&gt; &lt;span class="s2"&gt;"ISS"&lt;/span&gt;&lt;span class="p"&gt;,&lt;/span&gt; &lt;span class="s2"&gt;"microgravity"&lt;/span&gt;&lt;span class="p"&gt;]&lt;/span&gt;&lt;span class="p"&gt;}&lt;/span&gt;
    &lt;span class="p"&gt;}&lt;/span&gt;
  &lt;span class="p"&gt;}&lt;/span&gt;&lt;span class="p"&gt;]&lt;/span&gt;
&lt;span class="p"&gt;}&lt;/span&gt;
&lt;/code&gt;&lt;/pre&gt;

&lt;/div&gt;



&lt;p&gt;The settings and the four answers are the owner's. If one is missing, &lt;code&gt;plan.py&lt;/code&gt; lists it and the agent asks; it never fills one in. &lt;code&gt;max_video_s&lt;/code&gt; comes from the connected TikTok account itself, because TikTok sets it per account.&lt;/p&gt;

&lt;h2&gt;
  
  
  Step 4: the plan page
&lt;/h2&gt;



&lt;div class="highlight js-code-highlight"&gt;
&lt;pre class="highlight shell"&gt;&lt;code&gt;python3 &lt;span class="o"&gt;&amp;lt;&lt;/span&gt;skill&lt;span class="o"&gt;&amp;gt;&lt;/span&gt;/scripts/plan.py crosspost.json
&lt;/code&gt;&lt;/pre&gt;

&lt;/div&gt;





&lt;div class="highlight js-code-highlight"&gt;
&lt;pre class="highlight plaintext"&gt;&lt;code&gt;3 posts planned -&amp;gt; plan.md, plan.html (0 errors, 0 warnings, 0 notes)
by hand (1): by-hand/
requests: requests/microgravity.json
next: show plan.html (or paste plan.md) and wait for a clear yes. ...
&lt;/code&gt;&lt;/pre&gt;

&lt;/div&gt;



&lt;p&gt;&lt;code&gt;plan.py&lt;/code&gt; checks the file per account, every text and every setting, and counts each text the way its platform does. For the example: the TikTok caption is 198 of 2,200 characters with 3 hashtags, the Reels caption 329 of 2,200 with 4, the YouTube title 52 of 100 characters and the description 227 of 5,000 bytes. It also flags the things that aren't errors but are worth knowing: a link that won't be clickable, an @handle reused on another platform, a first line too long to stand alone, a hashtag that says nothing about the video.&lt;/p&gt;

&lt;p&gt;The result is &lt;code&gt;plan.html&lt;/code&gt; (the screenshot at the top) and &lt;code&gt;plan.md&lt;/code&gt;: every account, its text, settings, cover and time, plus a list of what to do in the apps after it's live. The agent shows it and waits for one yes that covers these posts, these accounts and these times.&lt;/p&gt;

&lt;p&gt;In the example only TikTok and Instagram are connected, so Shorts is planned by hand. &lt;code&gt;by-hand/microgravity/shorts.md&lt;/code&gt; has the title, description, tags, settings, the cover frame and numbered steps for YouTube Studio or the app.&lt;/p&gt;

&lt;h2&gt;
  
  
  Step 5: one request, a child per account
&lt;/h2&gt;

&lt;p&gt;After the yes, the agent uploads each file once, and &lt;code&gt;plan.py --media media.json&lt;/code&gt; fills in the media IDs. The request is one post with a child per connected account, each with its own text in &lt;code&gt;contentOverride&lt;/code&gt;. Trimmed from &lt;code&gt;requests/microgravity.json&lt;/code&gt;:&lt;br&gt;
&lt;/p&gt;

&lt;div class="highlight js-code-highlight"&gt;
&lt;pre class="highlight json"&gt;&lt;code&gt;&lt;span class="p"&gt;{&lt;/span&gt;
  &lt;span class="nl"&gt;"schedule"&lt;/span&gt;&lt;span class="p"&gt;:&lt;/span&gt; &lt;span class="p"&gt;{&lt;/span&gt;&lt;span class="nl"&gt;"type"&lt;/span&gt;&lt;span class="p"&gt;:&lt;/span&gt; &lt;span class="s2"&gt;"SCHEDULE_FOR_LATER"&lt;/span&gt;&lt;span class="p"&gt;,&lt;/span&gt; &lt;span class="nl"&gt;"scheduledFor"&lt;/span&gt;&lt;span class="p"&gt;:&lt;/span&gt; &lt;span class="s2"&gt;"2027-01-12T23:00:00Z"&lt;/span&gt;&lt;span class="p"&gt;}&lt;/span&gt;&lt;span class="p"&gt;,&lt;/span&gt;
  &lt;span class="nl"&gt;"socialAccounts"&lt;/span&gt;&lt;span class="p"&gt;:&lt;/span&gt; &lt;span class="p"&gt;[&lt;/span&gt;
    &lt;span class="p"&gt;{&lt;/span&gt;&lt;span class="nl"&gt;"platform"&lt;/span&gt;&lt;span class="p"&gt;:&lt;/span&gt; &lt;span class="s2"&gt;"TIKTOK"&lt;/span&gt;&lt;span class="p"&gt;,&lt;/span&gt; &lt;span class="nl"&gt;"postMode"&lt;/span&gt;&lt;span class="p"&gt;:&lt;/span&gt; &lt;span class="s2"&gt;"DIRECT_POST"&lt;/span&gt;&lt;span class="p"&gt;,&lt;/span&gt; &lt;span class="nl"&gt;"privacyLevel"&lt;/span&gt;&lt;span class="p"&gt;:&lt;/span&gt; &lt;span class="s2"&gt;"PUBLIC_TO_EVERYONE"&lt;/span&gt;&lt;span class="p"&gt;,&lt;/span&gt;
     &lt;span class="nl"&gt;"allowComment"&lt;/span&gt;&lt;span class="p"&gt;:&lt;/span&gt; &lt;span class="kc"&gt;true&lt;/span&gt;&lt;span class="p"&gt;,&lt;/span&gt; &lt;span class="nl"&gt;"allowDuet"&lt;/span&gt;&lt;span class="p"&gt;:&lt;/span&gt; &lt;span class="kc"&gt;false&lt;/span&gt;&lt;span class="p"&gt;,&lt;/span&gt; &lt;span class="nl"&gt;"allowStitch"&lt;/span&gt;&lt;span class="p"&gt;:&lt;/span&gt; &lt;span class="kc"&gt;false&lt;/span&gt;&lt;span class="p"&gt;,&lt;/span&gt;
     &lt;span class="nl"&gt;"isYourBrand"&lt;/span&gt;&lt;span class="p"&gt;:&lt;/span&gt; &lt;span class="kc"&gt;false&lt;/span&gt;&lt;span class="p"&gt;,&lt;/span&gt; &lt;span class="nl"&gt;"isBrandedContent"&lt;/span&gt;&lt;span class="p"&gt;:&lt;/span&gt; &lt;span class="kc"&gt;false&lt;/span&gt;&lt;span class="p"&gt;,&lt;/span&gt; &lt;span class="nl"&gt;"isAiGeneratedContent"&lt;/span&gt;&lt;span class="p"&gt;:&lt;/span&gt; &lt;span class="kc"&gt;false&lt;/span&gt;&lt;span class="p"&gt;,&lt;/span&gt;
     &lt;span class="nl"&gt;"cover"&lt;/span&gt;&lt;span class="p"&gt;:&lt;/span&gt; &lt;span class="p"&gt;{&lt;/span&gt;&lt;span class="nl"&gt;"type"&lt;/span&gt;&lt;span class="p"&gt;:&lt;/span&gt; &lt;span class="s2"&gt;"MEDIA_LIBRARY"&lt;/span&gt;&lt;span class="p"&gt;,&lt;/span&gt; &lt;span class="nl"&gt;"mediaId"&lt;/span&gt;&lt;span class="p"&gt;:&lt;/span&gt; &lt;span class="s2"&gt;"&amp;lt;cover&amp;gt;"&lt;/span&gt;&lt;span class="p"&gt;}&lt;/span&gt;&lt;span class="p"&gt;,&lt;/span&gt;
     &lt;span class="nl"&gt;"contentOverride"&lt;/span&gt;&lt;span class="p"&gt;:&lt;/span&gt; &lt;span class="p"&gt;[&lt;/span&gt;&lt;span class="p"&gt;{&lt;/span&gt;&lt;span class="nl"&gt;"text"&lt;/span&gt;&lt;span class="p"&gt;:&lt;/span&gt; &lt;span class="s2"&gt;"Why astronauts work out for hours every day\n..."&lt;/span&gt;&lt;span class="p"&gt;,&lt;/span&gt;
                          &lt;span class="nl"&gt;"media"&lt;/span&gt;&lt;span class="p"&gt;:&lt;/span&gt; &lt;span class="p"&gt;[&lt;/span&gt;&lt;span class="p"&gt;{&lt;/span&gt;&lt;span class="nl"&gt;"type"&lt;/span&gt;&lt;span class="p"&gt;:&lt;/span&gt; &lt;span class="s2"&gt;"MEDIA_LIBRARY"&lt;/span&gt;&lt;span class="p"&gt;,&lt;/span&gt; &lt;span class="nl"&gt;"mediaId"&lt;/span&gt;&lt;span class="p"&gt;:&lt;/span&gt; &lt;span class="s2"&gt;"&amp;lt;video&amp;gt;"&lt;/span&gt;&lt;span class="p"&gt;}&lt;/span&gt;&lt;span class="p"&gt;]&lt;/span&gt;&lt;span class="p"&gt;}&lt;/span&gt;&lt;span class="p"&gt;]&lt;/span&gt;&lt;span class="p"&gt;}&lt;/span&gt;&lt;span class="p"&gt;,&lt;/span&gt;
    &lt;span class="p"&gt;{&lt;/span&gt;&lt;span class="nl"&gt;"platform"&lt;/span&gt;&lt;span class="p"&gt;:&lt;/span&gt; &lt;span class="s2"&gt;"INSTAGRAM"&lt;/span&gt;&lt;span class="p"&gt;,&lt;/span&gt; &lt;span class="nl"&gt;"postType"&lt;/span&gt;&lt;span class="p"&gt;:&lt;/span&gt; &lt;span class="s2"&gt;"REEL"&lt;/span&gt;&lt;span class="p"&gt;,&lt;/span&gt; &lt;span class="nl"&gt;"isAiGeneratedContent"&lt;/span&gt;&lt;span class="p"&gt;:&lt;/span&gt; &lt;span class="kc"&gt;false&lt;/span&gt;&lt;span class="p"&gt;,&lt;/span&gt;
     &lt;span class="nl"&gt;"cover"&lt;/span&gt;&lt;span class="p"&gt;:&lt;/span&gt; &lt;span class="p"&gt;{&lt;/span&gt;&lt;span class="nl"&gt;"type"&lt;/span&gt;&lt;span class="p"&gt;:&lt;/span&gt; &lt;span class="s2"&gt;"MEDIA_LIBRARY"&lt;/span&gt;&lt;span class="p"&gt;,&lt;/span&gt; &lt;span class="nl"&gt;"mediaId"&lt;/span&gt;&lt;span class="p"&gt;:&lt;/span&gt; &lt;span class="s2"&gt;"&amp;lt;cover&amp;gt;"&lt;/span&gt;&lt;span class="p"&gt;}&lt;/span&gt;&lt;span class="p"&gt;,&lt;/span&gt;
     &lt;span class="nl"&gt;"contentOverride"&lt;/span&gt;&lt;span class="p"&gt;:&lt;/span&gt; &lt;span class="p"&gt;[&lt;/span&gt;&lt;span class="p"&gt;{&lt;/span&gt;&lt;span class="nl"&gt;"text"&lt;/span&gt;&lt;span class="p"&gt;:&lt;/span&gt; &lt;span class="s2"&gt;"Astronauts on the space station exercise for hours every day.\n..."&lt;/span&gt;&lt;span class="p"&gt;,&lt;/span&gt;
                          &lt;span class="nl"&gt;"media"&lt;/span&gt;&lt;span class="p"&gt;:&lt;/span&gt; &lt;span class="p"&gt;[&lt;/span&gt;&lt;span class="p"&gt;{&lt;/span&gt;&lt;span class="nl"&gt;"type"&lt;/span&gt;&lt;span class="p"&gt;:&lt;/span&gt; &lt;span class="s2"&gt;"MEDIA_LIBRARY"&lt;/span&gt;&lt;span class="p"&gt;,&lt;/span&gt; &lt;span class="nl"&gt;"mediaId"&lt;/span&gt;&lt;span class="p"&gt;:&lt;/span&gt; &lt;span class="s2"&gt;"&amp;lt;video&amp;gt;"&lt;/span&gt;&lt;span class="p"&gt;}&lt;/span&gt;&lt;span class="p"&gt;]&lt;/span&gt;&lt;span class="p"&gt;}&lt;/span&gt;&lt;span class="p"&gt;]&lt;/span&gt;&lt;span class="p"&gt;}&lt;/span&gt;
  &lt;span class="p"&gt;]&lt;/span&gt;
&lt;span class="p"&gt;}&lt;/span&gt;
&lt;/code&gt;&lt;/pre&gt;

&lt;/div&gt;



&lt;p&gt;The 6 pm New York time became &lt;code&gt;23:00Z&lt;/code&gt;: &lt;code&gt;plan.py&lt;/code&gt; stops when a clock time has no time zone, and never assumes UTC. The order is fixed: the yes, the upload, &lt;code&gt;posts_validate&lt;/code&gt; with the exact request, &lt;code&gt;posts_create&lt;/code&gt;, then &lt;code&gt;plan.py --posted&lt;/code&gt; with the post ID, then &lt;code&gt;posts_get&lt;/code&gt; for the status per account. The request is validated as a whole: if one child is invalid, nothing is created. If validation forces a change, the agent plans again and asks for a new yes.&lt;/p&gt;

&lt;h2&gt;
  
  
  Nothing goes out twice
&lt;/h2&gt;

&lt;p&gt;&lt;code&gt;--posted&lt;/code&gt; writes the video and every account in that request to &lt;code&gt;posted.json&lt;/code&gt;. From then on, &lt;code&gt;plan.py&lt;/code&gt; leaves those accounts out of every new request for that video and shows them as "already posted". That matters for the most common next step: you connect the third account next week, plan again, and the new request holds only Shorts. The same goes for a failure. If one account fails, &lt;code&gt;plan.py --failed&lt;/code&gt; releases just that account, and the accounts that went through are never sent again.&lt;/p&gt;

&lt;p&gt;A scheduled post can't be checked before its time. After it, "check my cross-post" makes the agent read &lt;code&gt;posted.json&lt;/code&gt;, call &lt;code&gt;posts_get&lt;/code&gt; for each recorded post and report one line per account: live with its link, still processing, or failed and why.&lt;/p&gt;

&lt;h2&gt;
  
  
  The free plan math, said up front
&lt;/h2&gt;

&lt;p&gt;Disclosure: the skill is made by &lt;a href="https://woopsocial.com/?utm_source=cross-post-video-skill&amp;amp;utm_medium=devto" rel="noopener noreferrer"&gt;the team behind WoopSocial&lt;/a&gt;, and posting goes through WoopSocial. It posts through each platform's own API with WoopSocial's approved apps, so you don't need a developer app or an app review of your own. The free plan covers 2 social accounts, no card. TikTok, Reels and Shorts are 3, so in its first message the agent asks which you want: it posts 2 and gives you the by-hand page for the third, posts all of them on a paid plan, or prepares everything and you post all of them by hand with nothing to sign up for. The connect step for your agent is on &lt;a href="https://woopsocial.com/agents?utm_source=cross-post-video-skill&amp;amp;utm_medium=devto" rel="noopener noreferrer"&gt;the WoopSocial setup page for AI agents&lt;/a&gt;.&lt;/p&gt;

&lt;p&gt;WoopSocial posts the video and the texts. It doesn't write captions or change the video: the agent writes the texts, and the skill's own script does any re-encoding. Some things can't be set through it, and the plan lists them under "After it's live": the YouTube cover, YouTube's altered content and paid promotion answers, and Instagram's paid partnership label. A sound picked from an app's library can't be added either; for TikTok the agent can send the video as a draft to your inbox, where you add the sound and post it.&lt;/p&gt;

&lt;h2&gt;
  
  
  Several accounts per platform
&lt;/h2&gt;

&lt;p&gt;The job file takes as many accounts as you have, including two on one platform (a Dutch and an English TikTok, say). Each account then gets its own text with &lt;code&gt;"text": {"&amp;lt;account name&amp;gt;": ...}&lt;/code&gt; and its audience written down, and the plan warns once per job with what that platform's own rules say about the same video on several accounts. The skill never offers to speed up, crop or filter copies to make them look different, and it posts the original export only: no copy saved from another app, and nothing added to the picture.&lt;/p&gt;

&lt;p&gt;It also makes no reach promises. No platform has said that posting order, a gap between platforms, a time of day or a hashtag count changes how a video is shown, so the plan doesn't suggest any, and it doesn't add #Shorts, which YouTube doesn't need.&lt;/p&gt;

&lt;h2&gt;
  
  
  Try it
&lt;/h2&gt;



&lt;div class="highlight js-code-highlight"&gt;
&lt;pre class="highlight shell"&gt;&lt;code&gt;npx skills add social-media-skills/cross-post-video-skill -g -a claude-code -y
&lt;/code&gt;&lt;/pre&gt;

&lt;/div&gt;



&lt;p&gt;Then tell your agent "post this video to TikTok, Reels and Shorts" or "cross-post clip.mp4 everywhere tomorrow at 6". It needs Python 3.9+, ffmpeg and Pillow; &lt;code&gt;scripts/check_setup.py&lt;/code&gt; prints the install command for anything missing. The full example, with every file the skill made, is in &lt;a href="https://github.com/social-media-skills/cross-post-video-skill/tree/main/skills/cross-post-video-skill/examples/microgravity" rel="noopener noreferrer"&gt;&lt;code&gt;examples/microgravity/&lt;/code&gt;&lt;/a&gt;. Issues and pull requests are welcome, and there are &lt;a href="https://social-media-skills.com/?utm_source=cross-post-video-skill&amp;amp;utm_medium=devto" rel="noopener noreferrer"&gt;more free skills for social media&lt;/a&gt; from the same team.&lt;/p&gt;

</description>
      <category>claudecode</category>
      <category>video</category>
      <category>python</category>
      <category>showdev</category>
    </item>
    <item>
      <title>Auto Captions Skill for Claude Code: Add Captions to Video, Make SRT Subtitles and Translate Them on Your Own Computer</title>
      <dc:creator>Frank Heijdenrijk</dc:creator>
      <pubDate>Wed, 11 Nov 2026 17:09:13 +0000</pubDate>
      <link>https://dev.to/fheijdenrijk/auto-captions-skill-for-claude-code-add-captions-to-video-make-srt-subtitles-and-translate-them-165d</link>
      <guid>https://dev.to/fheijdenrijk/auto-captions-skill-for-claude-code-add-captions-to-video-make-srt-subtitles-and-translate-them-165d</guid>
      <description>&lt;p&gt;Auto captions skill for Claude Code, free and MIT licensed: Auto Captions Skill lets your agent add captions to your videos on your own computer. It transcribes the speech with a time for every word, proofreads names and numbers before anything is burned in, renders creator-style captions in the shape each app wants, writes SRT and VTT subtitle files, and translates the captions with the same timing. It also works in Codex, Cursor and OpenClaw.&lt;/p&gt;

&lt;p&gt;There is no watermark, no per-minute limit and no paid API. Transcription runs locally with faster-whisper (or mlx-whisper on Apple Silicon), and the captions are drawn with Pillow and encoded with ffmpeg. This post covers what it does, the example in the repo, and the parts that are code.&lt;br&gt;
&lt;/p&gt;

&lt;div class="highlight js-code-highlight"&gt;
&lt;pre class="highlight shell"&gt;&lt;code&gt;npx skills add social-media-skills/auto-captions-skill -g -a claude-code -y
&lt;/code&gt;&lt;/pre&gt;

&lt;/div&gt;



&lt;p&gt;Repo: &lt;a href="https://github.com/social-media-skills/auto-captions-skill" rel="noopener noreferrer"&gt;github.com/social-media-skills/auto-captions-skill&lt;/a&gt;&lt;/p&gt;

&lt;p&gt;&lt;a href="https://media2.dev.to/dynamic/image/width=800%2Cheight=%2Cfit=scale-down%2Cgravity=auto%2Cformat=auto/https%3A%2F%2Fdev-to-uploads.s3.us-east-2.amazonaws.com%2Fuploads%2Farticles%2Fw55wxqn9fklme8hqqja7.png" class="article-body-image-wrapper"&gt;&lt;img src="https://media2.dev.to/dynamic/image/width=800%2Cheight=%2Cfit=scale-down%2Cgravity=auto%2Cformat=auto/https%3A%2F%2Fdev-to-uploads.s3.us-east-2.amazonaws.com%2Fuploads%2Farticles%2Fw55wxqn9fklme8hqqja7.png" alt="The same NASA clip five ways: the original frame, then captions in the pop, highlight and bold styles, and a Spanish version" width="799" height="297"&gt;&lt;/a&gt;&lt;/p&gt;

&lt;p&gt;&lt;em&gt;From the example in the repo: two clips of NASA astronaut Nichole Ayers on the International Space Station, captioned in English and Spanish (NASA footage, no endorsement implied).&lt;/em&gt;&lt;/p&gt;

&lt;h2&gt;
  
  
  Where auto captions go wrong
&lt;/h2&gt;

&lt;p&gt;A caption generator gets most words right and the important ones wrong. Names, brands and numbers are what speech engines mishear, and they drop commas, which decide where a caption breaks. A misspelled name burned into a video can't be fixed after posting. Then the captions sit under the app's buttons, split "9 or 10 minutes" across two screens, and nobody gets a subtitle file for accessibility.&lt;/p&gt;

&lt;p&gt;The agent you already use is a good proofreader. So the skill splits the work: scripts do the timing, chunking and rendering; the agent reads every line and fixes what the engine got wrong.&lt;/p&gt;

&lt;h2&gt;
  
  
  What the skill does
&lt;/h2&gt;

&lt;ol&gt;
&lt;li&gt;
&lt;strong&gt;Asks&lt;/strong&gt; where the videos will go, which names must be spelled right, and which languages.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Transcribes&lt;/strong&gt; with &lt;code&gt;transcribe.py&lt;/code&gt;: every word with its time.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Proofreads&lt;/strong&gt; with &lt;code&gt;fix.py&lt;/code&gt;: a report of unsure words and near-misses of your names; the agent writes the corrections, and every unchanged word keeps its time.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Writes subtitle files&lt;/strong&gt; with &lt;code&gt;captions.py&lt;/code&gt;: SRT, VTT and a plain transcript.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Shows the styles&lt;/strong&gt; on a frame of your own video with &lt;code&gt;styles.py&lt;/code&gt;.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Shows a review sheet&lt;/strong&gt; with &lt;code&gt;render.py --preview&lt;/code&gt;, with each app's buttons drawn on.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Renders&lt;/strong&gt; every video in every shape and language.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Plans the posts&lt;/strong&gt; if you want, and schedules them only after one clear yes.&lt;/li&gt;
&lt;/ol&gt;

&lt;h2&gt;
  
  
  Install
&lt;/h2&gt;

&lt;p&gt;Claude Code:&lt;br&gt;
&lt;/p&gt;

&lt;div class="highlight js-code-highlight"&gt;
&lt;pre class="highlight shell"&gt;&lt;code&gt;npx skills add social-media-skills/auto-captions-skill -g -a claude-code -y
&lt;/code&gt;&lt;/pre&gt;

&lt;/div&gt;



&lt;p&gt;Codex, Cursor and other agents: &lt;code&gt;npx skills add social-media-skills/auto-captions-skill&lt;/code&gt;, then pick your agent. OpenClaw: &lt;code&gt;openclaw skills install @social-media-skills/auto-captions-skill&lt;/code&gt;. GitHub CLI: &lt;code&gt;gh skill install social-media-skills/auto-captions-skill auto-captions-skill&lt;/code&gt;. Gemini CLI: &lt;code&gt;gemini extensions install https://github.com/social-media-skills/auto-captions-skill&lt;/code&gt;.&lt;/p&gt;

&lt;p&gt;It needs Python 3.9+ with Pillow, ffmpeg, and one way to transcribe: &lt;code&gt;pip install faster-whisper&lt;/code&gt; is free. &lt;code&gt;scripts/check_setup.py&lt;/code&gt; tells you what's missing and prints the install command for your system.&lt;/p&gt;

&lt;h2&gt;
  
  
  The example
&lt;/h2&gt;

&lt;p&gt;The prompt:&lt;/p&gt;

&lt;blockquote&gt;
&lt;p&gt;Add captions to clips/intro.mp4 and clips/speed.mp4 for TikTok, Reels and Shorts, a 16:9 version for YouTube, a 4:5 version for LinkedIn, and a Spanish version. Her name is Nichole Ayers.&lt;/p&gt;
&lt;/blockquote&gt;

&lt;p&gt;The agent writes a job file (videos, &lt;code&gt;vocab&lt;/code&gt;, &lt;code&gt;translations&lt;/code&gt;, &lt;code&gt;outputs&lt;/code&gt;) and runs the scripts. Transcribing both clips took 19 seconds on a laptop after the one-time model download. The raw transcript of the first clip, as faster-whisper heard it:&lt;br&gt;
&lt;/p&gt;

&lt;div class="highlight js-code-highlight"&gt;
&lt;pre class="highlight plaintext"&gt;&lt;code&gt;L1 [00:00.0-00:06.8] Hey everyone, I'm NASA astronaut Nichole Ayers and welcome aboard the International Space Station We're orbiting the Earth 16 times every single day
&lt;/code&gt;&lt;/pre&gt;

&lt;/div&gt;



&lt;p&gt;It dropped the word "where" and most commas. The agent's &lt;code&gt;fixes.json&lt;/code&gt; puts them back and picks the words to colour:&lt;br&gt;
&lt;/p&gt;

&lt;div class="highlight js-code-highlight"&gt;
&lt;pre class="highlight json"&gt;&lt;code&gt;&lt;span class="p"&gt;{&lt;/span&gt;
  &lt;span class="nl"&gt;"lines"&lt;/span&gt;&lt;span class="p"&gt;:&lt;/span&gt; &lt;span class="p"&gt;{&lt;/span&gt;&lt;span class="nl"&gt;"1"&lt;/span&gt;&lt;span class="p"&gt;:&lt;/span&gt; &lt;span class="s2"&gt;"Hey everyone, I'm NASA astronaut Nichole Ayers, and welcome aboard the International Space Station, where we're orbiting the Earth 16 times every single day."&lt;/span&gt;&lt;span class="p"&gt;}&lt;/span&gt;&lt;span class="p"&gt;,&lt;/span&gt;
  &lt;span class="nl"&gt;"emphasis"&lt;/span&gt;&lt;span class="p"&gt;:&lt;/span&gt; &lt;span class="p"&gt;[&lt;/span&gt;&lt;span class="s2"&gt;"16 times"&lt;/span&gt;&lt;span class="p"&gt;]&lt;/span&gt;
&lt;span class="p"&gt;}&lt;/span&gt;
&lt;/code&gt;&lt;/pre&gt;

&lt;/div&gt;



&lt;p&gt;One more detail from the example: NASA's own subtitle file for this video spells her name "Nicole". The captions use "Nichole", as on the name title in the video. That's the kind of thing only a proofreading step catches.&lt;/p&gt;

&lt;p&gt;The second clip's subtitle file, as &lt;code&gt;captions.py&lt;/code&gt; wrote it:&lt;br&gt;
&lt;/p&gt;

&lt;div class="highlight js-code-highlight"&gt;
&lt;pre class="highlight plaintext"&gt;&lt;code&gt;4
00:00:08,500 --&amp;gt; 00:00:09,597
So we'll actually only have

5
00:00:09,680 --&amp;gt; 00:00:11,557
about 9 or 10 minutes
of good radio contact
&lt;/code&gt;&lt;/pre&gt;

&lt;/div&gt;



&lt;p&gt;"9 or 10 minutes" stays on one line. Cues are at most 2 lines of 42 characters, last 5/6 of a second to 7 seconds, and break at punctuation.&lt;/p&gt;

&lt;h2&gt;
  
  
  The review sheet
&lt;/h2&gt;

&lt;p&gt;Before anything is rendered, &lt;code&gt;render.py --preview&lt;/code&gt; writes a sheet: three moments per version, and on the vertical ones the same frame with TikTok's, Reels' and Shorts' buttons drawn on. No app publishes a safe zone for normal posts, so the skill draws the buttons and you check that the captions miss them, the face and any name title. You move them with &lt;code&gt;position&lt;/code&gt; and the crop with &lt;code&gt;focus_x&lt;/code&gt;.&lt;/p&gt;

&lt;p&gt;&lt;a href="https://media2.dev.to/dynamic/image/width=800%2Cheight=%2Cfit=scale-down%2Cgravity=auto%2Cformat=auto/https%3A%2F%2Fdev-to-uploads.s3.us-east-2.amazonaws.com%2Fuploads%2Farticles%2F2mifoiybgefn0ocvdeao.jpg" class="article-body-image-wrapper"&gt;&lt;img src="https://media2.dev.to/dynamic/image/width=800%2Cheight=%2Cfit=scale-down%2Cgravity=auto%2Cformat=auto/https%3A%2F%2Fdev-to-uploads.s3.us-east-2.amazonaws.com%2Fuploads%2Farticles%2F2mifoiybgefn0ocvdeao.jpg" alt="Review sheet: the clip in the pop style at three moments, the same frame with TikTok, Reels and Shorts buttons drawn on, the Spanish version, the 16:9 version and the 4:5 LinkedIn version" width="800" height="661"&gt;&lt;/a&gt;&lt;/p&gt;

&lt;h2&gt;
  
  
  Styles are JSON
&lt;/h2&gt;

&lt;p&gt;The 8 presets (&lt;code&gt;pop&lt;/code&gt;, &lt;code&gt;bold&lt;/code&gt;, &lt;code&gt;highlight&lt;/code&gt;, &lt;code&gt;reveal&lt;/code&gt;, &lt;code&gt;lines&lt;/code&gt;, &lt;code&gt;clean&lt;/code&gt;, &lt;code&gt;minimal&lt;/code&gt;, &lt;code&gt;one-word&lt;/code&gt;) are plain JSON files. &lt;code&gt;pop&lt;/code&gt;, the default for vertical video:&lt;br&gt;
&lt;/p&gt;

&lt;div class="highlight js-code-highlight"&gt;
&lt;pre class="highlight json"&gt;&lt;code&gt;&lt;span class="p"&gt;{&lt;/span&gt;
  &lt;span class="nl"&gt;"font"&lt;/span&gt;&lt;span class="p"&gt;:&lt;/span&gt; &lt;span class="s2"&gt;"Poppins-ExtraBold.ttf"&lt;/span&gt;&lt;span class="p"&gt;,&lt;/span&gt; &lt;span class="nl"&gt;"size"&lt;/span&gt;&lt;span class="p"&gt;:&lt;/span&gt; &lt;span class="mi"&gt;84&lt;/span&gt;&lt;span class="p"&gt;,&lt;/span&gt;
  &lt;span class="nl"&gt;"color"&lt;/span&gt;&lt;span class="p"&gt;:&lt;/span&gt; &lt;span class="s2"&gt;"#FFFFFF"&lt;/span&gt;&lt;span class="p"&gt;,&lt;/span&gt; &lt;span class="nl"&gt;"active"&lt;/span&gt;&lt;span class="p"&gt;:&lt;/span&gt; &lt;span class="s2"&gt;"#FFD43B"&lt;/span&gt;&lt;span class="p"&gt;,&lt;/span&gt; &lt;span class="nl"&gt;"keyword"&lt;/span&gt;&lt;span class="p"&gt;:&lt;/span&gt; &lt;span class="s2"&gt;"#4ADE80"&lt;/span&gt;&lt;span class="p"&gt;,&lt;/span&gt;
  &lt;span class="nl"&gt;"stroke"&lt;/span&gt;&lt;span class="p"&gt;:&lt;/span&gt; &lt;span class="mi"&gt;9&lt;/span&gt;&lt;span class="p"&gt;,&lt;/span&gt; &lt;span class="nl"&gt;"words"&lt;/span&gt;&lt;span class="p"&gt;:&lt;/span&gt; &lt;span class="mi"&gt;3&lt;/span&gt;&lt;span class="p"&gt;,&lt;/span&gt; &lt;span class="nl"&gt;"chars"&lt;/span&gt;&lt;span class="p"&gt;:&lt;/span&gt; &lt;span class="mi"&gt;18&lt;/span&gt;&lt;span class="p"&gt;,&lt;/span&gt; &lt;span class="nl"&gt;"lines"&lt;/span&gt;&lt;span class="p"&gt;:&lt;/span&gt; &lt;span class="mi"&gt;1&lt;/span&gt;&lt;span class="p"&gt;,&lt;/span&gt;
  &lt;span class="nl"&gt;"pause"&lt;/span&gt;&lt;span class="p"&gt;:&lt;/span&gt; &lt;span class="mf"&gt;0.3&lt;/span&gt;&lt;span class="p"&gt;,&lt;/span&gt; &lt;span class="nl"&gt;"position"&lt;/span&gt;&lt;span class="p"&gt;:&lt;/span&gt; &lt;span class="mf"&gt;0.62&lt;/span&gt;
&lt;span class="p"&gt;}&lt;/span&gt;
&lt;/code&gt;&lt;/pre&gt;

&lt;/div&gt;



&lt;p&gt;&lt;code&gt;active&lt;/code&gt; is the spoken word, &lt;code&gt;keyword&lt;/code&gt; the emphasis words from &lt;code&gt;fixes.json&lt;/code&gt;, &lt;code&gt;pause&lt;/code&gt; the silence that starts a new caption. Change any of it per output with &lt;code&gt;style_overrides&lt;/code&gt;, or copy a preset into your project as your brand style. Captions are drawn with Pillow, so any ffmpeg build works: it needs no subtitle library.&lt;/p&gt;

&lt;h2&gt;
  
  
  Translations keep the timing
&lt;/h2&gt;

&lt;p&gt;For each language, &lt;code&gt;captions.py&lt;/code&gt; writes a template with one entry per cue of the original: &lt;code&gt;i&lt;/code&gt;, &lt;code&gt;start&lt;/code&gt;, &lt;code&gt;end&lt;/code&gt;, &lt;code&gt;source&lt;/code&gt; and an empty &lt;code&gt;text&lt;/code&gt;. The agent fills in &lt;code&gt;text&lt;/code&gt;; the times never change, so the translation stays in sync. From the Spanish file:&lt;br&gt;
&lt;/p&gt;

&lt;div class="highlight js-code-highlight"&gt;
&lt;pre class="highlight json"&gt;&lt;code&gt;&lt;span class="p"&gt;{&lt;/span&gt;&lt;span class="nl"&gt;"i"&lt;/span&gt;&lt;span class="p"&gt;:&lt;/span&gt; &lt;span class="mi"&gt;5&lt;/span&gt;&lt;span class="p"&gt;,&lt;/span&gt; &lt;span class="nl"&gt;"start"&lt;/span&gt;&lt;span class="p"&gt;:&lt;/span&gt; &lt;span class="mf"&gt;9.68&lt;/span&gt;&lt;span class="p"&gt;,&lt;/span&gt; &lt;span class="nl"&gt;"end"&lt;/span&gt;&lt;span class="p"&gt;:&lt;/span&gt; &lt;span class="mf"&gt;11.557&lt;/span&gt;&lt;span class="p"&gt;,&lt;/span&gt;
 &lt;span class="nl"&gt;"source"&lt;/span&gt;&lt;span class="p"&gt;:&lt;/span&gt; &lt;span class="s2"&gt;"about 9 or 10 minutes of good radio contact"&lt;/span&gt;&lt;span class="p"&gt;,&lt;/span&gt;
 &lt;span class="nl"&gt;"text"&lt;/span&gt;&lt;span class="p"&gt;:&lt;/span&gt; &lt;span class="s2"&gt;"unos 9 o 10 minutos de contacto por radio"&lt;/span&gt;&lt;span class="p"&gt;}&lt;/span&gt;
&lt;/code&gt;&lt;/pre&gt;

&lt;/div&gt;



&lt;p&gt;The script warns when a translation reads faster than about 20 characters per second. The first English cue, "We won't have a lot of time to talk,", is on screen for 1.3 seconds; the agent shortened the Spanish to "No tendremos mucho tiempo,". Units are converted for the audience (17,500 miles an hour became 28 000 km/h). Translated versions show whole captions without the moving word highlight, because a translation has no spoken word to follow.&lt;/p&gt;

&lt;h2&gt;
  
  
  Subtitle files, and what posting can carry
&lt;/h2&gt;

&lt;p&gt;Burned-in captions can't be resized or read aloud, so every video also gets an SRT, a VTT and a transcript. TikTok and Instagram take no subtitle file through their APIs, which is why the skill burns captions in. YouTube and Facebook let you add one; &lt;code&gt;plan.md&lt;/code&gt; says where.&lt;/p&gt;

&lt;p&gt;Posting is optional. &lt;code&gt;post_plan.py&lt;/code&gt; checks every file against each platform, plans one post a day per account at most, and writes &lt;code&gt;plan.md&lt;/code&gt;, which works by hand. To have the agent schedule it, the skill uses WoopSocial's &lt;a href="https://woopsocial.com/social-media-mcp?utm_source=auto-captions-skill&amp;amp;utm_medium=devto" rel="noopener noreferrer"&gt;social media MCP&lt;/a&gt;, made by the team that maintains the skill. It posts the video and the post text; it doesn't attach subtitle files or change the video. The free plan covers 2 social accounts, no card.&lt;/p&gt;

&lt;h2&gt;
  
  
  Try it
&lt;/h2&gt;



&lt;div class="highlight js-code-highlight"&gt;
&lt;pre class="highlight shell"&gt;&lt;code&gt;npx skills add social-media-skills/auto-captions-skill -g -a claude-code -y
&lt;/code&gt;&lt;/pre&gt;

&lt;/div&gt;



&lt;p&gt;Then tell your agent "add captions to this video for TikTok and Reels" and give it the names that must be spelled right. Rendering takes one to three times the video's length per output, depending on the computer. The repo is &lt;a href="https://github.com/social-media-skills/auto-captions-skill" rel="noopener noreferrer"&gt;social-media-skills/auto-captions-skill&lt;/a&gt;. Issues and pull requests are welcome.&lt;/p&gt;

</description>
      <category>claudecode</category>
      <category>video</category>
      <category>python</category>
      <category>showdev</category>
    </item>
    <item>
      <title>LinkedIn Skill for Claude Code: Write, Check, Preview and Schedule LinkedIn Posts in Your Own Voice</title>
      <dc:creator>Frank Heijdenrijk</dc:creator>
      <pubDate>Tue, 10 Nov 2026 09:28:36 +0000</pubDate>
      <link>https://dev.to/fheijdenrijk/linkedin-skill-for-claude-code-write-check-preview-and-schedule-linkedin-posts-in-your-own-voice-26kb</link>
      <guid>https://dev.to/fheijdenrijk/linkedin-skill-for-claude-code-write-check-preview-and-schedule-linkedin-posts-in-your-own-voice-26kb</guid>
      <description>&lt;p&gt;LinkedIn skill for Claude Code, free and MIT licensed: LinkedIn Skill lets your agent write posts for your personal profile and your company Pages in your own voice, make the images or captioned video from your material, check every post against LinkedIn's limits and the patterns that cost reach, show each post the way the feed will on a phone, and schedule it after you say yes. It also works in Codex, Cursor and OpenClaw.&lt;/p&gt;

&lt;p&gt;Everything up to scheduling runs on your own computer with Python and Pillow (ffmpeg only for video). There is no paid API, no LinkedIn developer app and no browser automation. This post covers what it does, how to install it, the example in the repo, and the parts that are code.&lt;br&gt;
&lt;/p&gt;

&lt;div class="highlight js-code-highlight"&gt;
&lt;pre class="highlight shell"&gt;&lt;code&gt;npx skills add social-media-skills/linkedin-skill -g -a claude-code -y
&lt;/code&gt;&lt;/pre&gt;

&lt;/div&gt;



&lt;p&gt;Repo: &lt;a href="https://github.com/social-media-skills/linkedin-skill" rel="noopener noreferrer"&gt;github.com/social-media-skills/linkedin-skill&lt;/a&gt;&lt;/p&gt;

&lt;p&gt;&lt;a href="https://media2.dev.to/dynamic/image/width=800%2Cheight=%2Cfit=scale-down%2Cgravity=auto%2Cformat=auto/https%3A%2F%2Fdev-to-uploads.s3.us-east-2.amazonaws.com%2Fuploads%2Farticles%2Fttqpjajifjqvkwrsdnrs.png" class="article-body-image-wrapper"&gt;&lt;img src="https://media2.dev.to/dynamic/image/width=800%2Cheight=%2Cfit=scale-down%2Cgravity=auto%2Cformat=auto/https%3A%2F%2Fdev-to-uploads.s3.us-east-2.amazonaws.com%2Fuploads%2Farticles%2Fttqpjajifjqvkwrsdnrs.png" alt="Five LinkedIn posts the skill made about NASA's Artemis II mission, shown as the feed shows them on a phone" width="800" height="290"&gt;&lt;/a&gt;&lt;/p&gt;

&lt;p&gt;&lt;em&gt;The example in the repo: a week of posts about Artemis II for a placeholder profile and an astronomy club's Page, made from NASA photos and video (no endorsement implied).&lt;/em&gt;&lt;/p&gt;

&lt;h2&gt;
  
  
  The problem with AI LinkedIn posts
&lt;/h2&gt;

&lt;p&gt;Most AI help for LinkedIn produces the same post: a dramatic one-liner, one sentence per line, three emoji, five hashtags, "Agree?". Readers spot it in a second, and LinkedIn has said it now down-ranks low-quality AI text and cuts reach for engagement bait. The fix isn't a better prompt. It's writing from the person's own material, in their own voice, and checking the result before it goes out.&lt;/p&gt;

&lt;h2&gt;
  
  
  What the skill does
&lt;/h2&gt;

&lt;ol&gt;
&lt;li&gt;
&lt;strong&gt;Learns your voice.&lt;/strong&gt; &lt;code&gt;voice.py&lt;/code&gt; reads 5 to 20 of your own past posts and writes down how you write: typical length, how you open, sentence length, emoji and hashtag habits, the words you use. The check compares every draft with it.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Writes from your material.&lt;/strong&gt; Notes, a changelog, a call summary, a photo from the week. One idea per post, the specific thing in the first line.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Makes the media.&lt;/strong&gt; &lt;code&gt;render_images.py&lt;/code&gt; crops your photos to 4:5 or 1:1 and makes quote, number and list cards; &lt;code&gt;video.py&lt;/code&gt; fits a clip to 4:5 and burns in captions from your transcript.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Checks everything&lt;/strong&gt; with &lt;code&gt;li_check.py&lt;/code&gt;.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Picks the dates&lt;/strong&gt; with &lt;code&gt;schedule.py&lt;/code&gt;: your days and time per account, one post a day, a few a week.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Previews&lt;/strong&gt; each post on a phone, cut at "...see more", with LinkedIn's image grid, the video poster or the link card.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Schedules&lt;/strong&gt; after one yes, and logs every post.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Learns&lt;/strong&gt; from day 2: impressions against your own median, then comments, reposts, saves and sends per 1,000 impressions.&lt;/li&gt;
&lt;/ol&gt;

&lt;h2&gt;
  
  
  Install
&lt;/h2&gt;

&lt;p&gt;Claude Code:&lt;br&gt;
&lt;/p&gt;

&lt;div class="highlight js-code-highlight"&gt;
&lt;pre class="highlight shell"&gt;&lt;code&gt;npx skills add social-media-skills/linkedin-skill -g -a claude-code -y
&lt;/code&gt;&lt;/pre&gt;

&lt;/div&gt;



&lt;p&gt;Codex, Cursor and other agents: &lt;code&gt;npx skills add social-media-skills/linkedin-skill&lt;/code&gt;, then pick your agent. OpenClaw: &lt;code&gt;openclaw skills install @social-media-skills/linkedin-skill&lt;/code&gt;. Gemini CLI: &lt;code&gt;gemini extensions install https://github.com/social-media-skills/linkedin-skill&lt;/code&gt;.&lt;/p&gt;

&lt;p&gt;It needs Python 3.9+ with Pillow, and ffmpeg for video. &lt;code&gt;scripts/check_setup.py&lt;/code&gt; tells you what's missing.&lt;/p&gt;

&lt;h2&gt;
  
  
  The example
&lt;/h2&gt;

&lt;p&gt;&lt;code&gt;examples/artemis-ii/&lt;/code&gt; is a week of posts about NASA's crewed lunar flyby, for a placeholder personal profile and a placeholder company Page:&lt;/p&gt;

&lt;ul&gt;
&lt;li&gt;
&lt;strong&gt;Profile:&lt;/strong&gt; Earthset as a single photo, the flyby in four photos as an image set, and NASA's mission page as a link card.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Page:&lt;/strong&gt; the liftoff seen from a camera on Orion (30 seconds, 4:5, captions burned in), and the crew on the raft after splashdown, each with starter lines for the team to reshare.&lt;/li&gt;
&lt;/ul&gt;

&lt;p&gt;The "past posts" for the voice step are written for the example: short, plain, no emoji, no hashtags. The drafts follow that, and the check would flag a dash or an emoji the voice never uses.&lt;/p&gt;

&lt;h2&gt;
  
  
  The checker
&lt;/h2&gt;

&lt;p&gt;&lt;code&gt;li_check.py&lt;/code&gt; covers LinkedIn's published limits and then the things that cost reach or trust:&lt;/p&gt;

&lt;ul&gt;
&lt;li&gt;
&lt;strong&gt;The fold.&lt;/strong&gt; About 140 characters show on a phone before "...see more" (about 3 lines with media). A first line longer than that gets cut, so the check says so.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Bait and stock phrasing.&lt;/strong&gt; "Comment YES if you agree", "like if", "tag someone", and a list of phrases readers take as generic AI text. Long dashes are flagged too, unless the user's own posts use them.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Formatting LinkedIn can't use.&lt;/strong&gt; Unicode bold letters (screen readers spell them out, search can't read them), more than 3 hashtags, @names (the posting API can't tag people), URLs pasted into the text.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Media.&lt;/strong&gt; Image types, size, minimum size and shape (taller than 4:5 is cropped), 2 to 20 images of one shape, video length and shape, and never media and a link card together: WoopSocial rejects that combination too, because LinkedIn shows one or the other.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Accounts.&lt;/strong&gt; The same post on profile and Page on the same day, more than one post a day, the week's count, and a picture or opening line already used (from the log).&lt;/li&gt;
&lt;/ul&gt;

&lt;h2&gt;
  
  
  Links: a trade-off, not a rule
&lt;/h2&gt;

&lt;p&gt;Many practitioners measure less reach for posts with a link card; LinkedIn has never confirmed it, the measurements disagree, and putting the link in the first comment no longer seems to avoid it. The skill doesn't pretend otherwise: it notes the trade-off on every link post and lets you choose.&lt;/p&gt;

&lt;h2&gt;
  
  
  Profiles and company Pages without your own LinkedIn app
&lt;/h2&gt;

&lt;p&gt;Posting to a profile through LinkedIn's API needs a developer app. Posting to a company Page needs Community Management API access, which LinkedIn grants to registered organisations after a review. The skill schedules through &lt;a href="https://woopsocial.com/social-media-mcp/linkedin-mcp?utm_source=linkedin-skill&amp;amp;utm_medium=devto" rel="noopener noreferrer"&gt;WoopSocial&lt;/a&gt;, a social media API and MCP server made by the team that maintains the skill, which has that access. You connect your profile and the Pages you administer, and the agent posts text, images, 2 to 20 images, video and link cards.&lt;/p&gt;

&lt;p&gt;What it doesn't send: documents (PDF carousels), polls, @mentions and comments. For carousels the skill renders the slides and either posts them as an image set or gives you a PDF to upload in the app.&lt;/p&gt;

&lt;h2&gt;
  
  
  Results against yourself
&lt;/h2&gt;

&lt;p&gt;Page numbers come back through WoopSocial. Personal profile numbers don't, because LinkedIn restricts them for posting apps, so you paste them into a CSV. &lt;code&gt;log.py&lt;/code&gt; compares each post with the same account's own posts, never a benchmark, and names a best post only after six.&lt;/p&gt;

&lt;h2&gt;
  
  
  Try it
&lt;/h2&gt;



&lt;div class="highlight js-code-highlight"&gt;
&lt;pre class="highlight shell"&gt;&lt;code&gt;npx skills add social-media-skills/linkedin-skill -g -a claude-code -y
&lt;/code&gt;&lt;/pre&gt;

&lt;/div&gt;



&lt;p&gt;Then tell your agent "write this week's LinkedIn posts from these notes" and give it a few of your past posts. The repo is &lt;a href="https://github.com/social-media-skills/linkedin-skill" rel="noopener noreferrer"&gt;social-media-skills/linkedin-skill&lt;/a&gt;. Issues and pull requests are welcome.&lt;/p&gt;

</description>
      <category>claudecode</category>
      <category>linkedin</category>
      <category>python</category>
      <category>showdev</category>
    </item>
    <item>
      <title>Pinterest Skill for Claude Code: Make, Check, Preview and Schedule Pinterest Pins From Your Agent</title>
      <dc:creator>Frank Heijdenrijk</dc:creator>
      <pubDate>Mon, 09 Nov 2026 10:06:25 +0000</pubDate>
      <link>https://dev.to/fheijdenrijk/pinterest-skill-for-claude-code-make-check-preview-and-schedule-pinterest-pins-from-your-agent-3ggc</link>
      <guid>https://dev.to/fheijdenrijk/pinterest-skill-for-claude-code-make-check-preview-and-schedule-pinterest-pins-from-your-agent-3ggc</guid>
      <description>&lt;p&gt;Pinterest skill for Claude Code, free and MIT licensed: Pinterest Skill lets your agent turn a page (a blog post, a product, a recipe, a guide) and your own photos or videos into several genuinely different pins, check every pin against Pinterest's limits and the patterns that get accounts flagged, show you the batch in the home feed, and schedule it over the coming weeks after you say yes. It also works in Codex, Cursor and OpenClaw.&lt;/p&gt;

&lt;p&gt;Everything up to scheduling runs on your own computer with Python and Pillow (ffmpeg only for video pins). There is no paid API. This post covers what the skill does, how to install it, the example in the repo, and then the parts that are code: the renderer, the checker and its repeat detection, the scheduler and the results.&lt;br&gt;
&lt;/p&gt;

&lt;div class="highlight js-code-highlight"&gt;
&lt;pre class="highlight shell"&gt;&lt;code&gt;npx skills add social-media-skills/pinterest-skill &lt;span class="nt"&gt;-g&lt;/span&gt; &lt;span class="nt"&gt;-a&lt;/span&gt; claude-code &lt;span class="nt"&gt;-y&lt;/span&gt;
&lt;/code&gt;&lt;/pre&gt;

&lt;/div&gt;



&lt;p&gt;Repo: &lt;a href="https://github.com/social-media-skills/pinterest-skill" rel="noopener noreferrer"&gt;github.com/social-media-skills/pinterest-skill&lt;/a&gt;&lt;/p&gt;

&lt;p&gt;&lt;a href="https://media2.dev.to/dynamic/image/width=800%2Cheight=%2Cfit=scale-down%2Cgravity=auto%2Cformat=auto/https%3A%2F%2Fdev-to-uploads.s3.us-east-2.amazonaws.com%2Fuploads%2Farticles%2Fceliqumwyl9pdk9hh2jl.png" class="article-body-image-wrapper"&gt;&lt;img src="https://media2.dev.to/dynamic/image/width=800%2Cheight=%2Cfit=scale-down%2Cgravity=auto%2Cformat=auto/https%3A%2F%2Fdev-to-uploads.s3.us-east-2.amazonaws.com%2Fuploads%2Farticles%2Fceliqumwyl9pdk9hh2jl.png" alt="Five pins the Pinterest skill made for one page about Moon phases, shown in the home feed: a grid of all 8 phases, a 12-second NASA video, a full Moon photo pin, a big 29.5 days fact pin and a viewing guide" width="799" height="284"&gt;&lt;/a&gt;&lt;/p&gt;

&lt;p&gt;&lt;em&gt;The example in the repo: five pins for NASA's Moon phases page, made only from NASA images, NASA's Moon phase animation and illustrations based on them (no endorsement implied).&lt;/em&gt;&lt;/p&gt;

&lt;h2&gt;
  
  
  What is a Pinterest skill?
&lt;/h2&gt;

&lt;p&gt;An agent skill is a folder with a &lt;code&gt;SKILL.md&lt;/code&gt; that tells an agent like Claude Code how to do a job, plus the scripts it runs. Most Claude Pinterest help stops at text: a few titles, a description, a list of keywords. The work that decides whether pins get seen is different and repetitive: several fresh designs for every page, the keyword where the feed shows it, the right board, a steady pace, and never the same picture pinned to the same page again.&lt;/p&gt;

&lt;p&gt;This skill ships the scripts that do that work, so the agent hands you real pins:&lt;/p&gt;

&lt;ul&gt;
&lt;li&gt;1000x1500 pins in six layouts from your photos, with the subject kept in the crop;&lt;/li&gt;
&lt;li&gt;video pins from your own clip or photos, at 2:3 or 9:16, opening on the strongest frame;&lt;/li&gt;
&lt;li&gt;titles, descriptions and alt text that use the page's own words;&lt;/li&gt;
&lt;li&gt;a check that refuses what Pinterest treats as spam or rejects;&lt;/li&gt;
&lt;li&gt;a preview of the batch in the home feed at real size, then each close-up;&lt;/li&gt;
&lt;li&gt;a schedule that spreads pins out, and one validated request per pin.&lt;/li&gt;
&lt;/ul&gt;

&lt;h2&gt;
  
  
  Install
&lt;/h2&gt;

&lt;p&gt;Claude Code:&lt;br&gt;
&lt;/p&gt;

&lt;div class="highlight js-code-highlight"&gt;
&lt;pre class="highlight shell"&gt;&lt;code&gt;npx skills add social-media-skills/pinterest-skill &lt;span class="nt"&gt;-g&lt;/span&gt; &lt;span class="nt"&gt;-a&lt;/span&gt; claude-code &lt;span class="nt"&gt;-y&lt;/span&gt;
&lt;/code&gt;&lt;/pre&gt;

&lt;/div&gt;



&lt;p&gt;Codex, Cursor and other agents: &lt;code&gt;npx skills add social-media-skills/pinterest-skill&lt;/code&gt;, then pick your agent. OpenClaw: &lt;code&gt;openclaw skills install @social-media-skills/pinterest-skill&lt;/code&gt;. Gemini CLI: &lt;code&gt;gemini extensions install https://github.com/social-media-skills/pinterest-skill&lt;/code&gt;.&lt;/p&gt;

&lt;p&gt;It needs Python 3.9+ with Pillow, and ffmpeg for video pins. &lt;code&gt;scripts/check_setup.py&lt;/code&gt; tells you what's missing and prints the install command for your OS.&lt;/p&gt;

&lt;h2&gt;
  
  
  The example
&lt;/h2&gt;

&lt;p&gt;&lt;code&gt;examples/moon-phases/&lt;/code&gt; holds five pins for one page, NASA's guide to the Moon phases, each with a different angle and its own picture:&lt;/p&gt;

&lt;ol&gt;
&lt;li&gt;
&lt;strong&gt;All 8 phases in a grid&lt;/strong&gt;, as seen from the Northern Hemisphere (illustrations made by shading NASA's Lunar Reconnaissance Orbiter view of the Moon).&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;A 12-second video&lt;/strong&gt; of one cycle, cut from NASA's Moon phase animation for 2026 with &lt;code&gt;video.py fit --start 1.5 --end 25.3 --speed 2&lt;/code&gt;: full Moon to full Moon, and it opens on the full Moon, because the first frame is the cover.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;A full Moon over Kennedy Space Center&lt;/strong&gt; for "why we always see the same side of the Moon".&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;A big 29.5&lt;/strong&gt; for the length of the cycle, over an astronaut photo of a crescent Moon above Earth.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;When to see the main phases&lt;/strong&gt;, from NASA's rise and set times, over a NASA photo of the full Moon above the U.S. Capitol.&lt;/li&gt;
&lt;/ol&gt;

&lt;p&gt;Every pin, check report, preview, schedule and WoopSocial request in that folder came from the scripts.&lt;/p&gt;

&lt;h2&gt;
  
  
  The renderer: designs that read at feed size
&lt;/h2&gt;

&lt;p&gt;&lt;code&gt;render_pins.py&lt;/code&gt; takes a &lt;code&gt;pins.json&lt;/code&gt; with your brand colours and one entry per pin: a layout, the photo, where the subject sits (&lt;code&gt;focus&lt;/code&gt;), and the headline. It renders at twice the size and scales down for crisp text, fits the headline to its box, balances the line lengths so no word sits alone on the last line, and keeps shrinking a single long word rather than cutting it.&lt;/p&gt;

&lt;p&gt;It writes a manifest per batch: each pin's source photos and its headline height at the feed's column width (236 pixels). The check reads both. A headline under about 14 pixels in the feed can't be read on a phone, so the check says so. If the text contains a character the font can't draw (an emoji, CJK in a Latin font), the renderer reports it instead of printing empty boxes, and &lt;code&gt;brand.font_file&lt;/code&gt; takes a font that has them.&lt;/p&gt;

&lt;h2&gt;
  
  
  The checker: repeats, pace and the page
&lt;/h2&gt;

&lt;p&gt;&lt;code&gt;pin_check.py&lt;/code&gt; covers the published limits (formats, sizes, the 100, 800 and 500 character fields, video length) and then the parts that decide whether an account gets flagged:&lt;/p&gt;

&lt;ul&gt;
&lt;li&gt;
&lt;strong&gt;Repeats.&lt;/strong&gt; Each picture gets a fingerprint: a 256-bit difference hash plus a tiny colour thumbnail. A difference hash alone calls two dark photos "the same", and the colour check stops that. Two pins for one page that look alike are refused, and so are two pins built from the same source photo, even in different layouts.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;The log.&lt;/strong&gt; &lt;code&gt;log.py record&lt;/code&gt; writes every posted pin, with its fingerprint, to &lt;code&gt;pins-log.jsonl&lt;/code&gt;. A picture already pinned to the same page is refused next time. Pages are compared by a canonical URL (no tracking parameters, no fragment, no &lt;code&gt;www&lt;/code&gt;), so adding &lt;code&gt;?utm_source=pinterest&lt;/code&gt; doesn't make an old picture new.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Boards and pace.&lt;/strong&gt; One page sent to several boards on one day is refused, and so is a day with more pins than the account's cap, counting pins already posted.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;The page.&lt;/strong&gt; Links must be full https addresses: a typo'd key or a &lt;code&gt;javascript:&lt;/code&gt; link fails the check instead of reaching Pinterest. Shorteners are refused. If &lt;code&gt;page.py&lt;/code&gt; found a redirect, the check asks for the final URL, and a pin that shares no words with its page's title and headings gets a warning.&lt;/li&gt;
&lt;/ul&gt;

&lt;h2&gt;
  
  
  The scheduler: a steady trickle
&lt;/h2&gt;

&lt;p&gt;&lt;code&gt;schedule.py&lt;/code&gt; gives each unscheduled pin the first free day that keeps one pin a day per account and new pins for the same page 7 days apart, counting the log. Pins for a seasonal page (a holiday, a launch) go out 90 to 45 days before the moment, configurable per page. A start date in the past moves to today, a moment that has passed turns the pin into an evergreen one, and requests for pins you removed from the plan are deleted so they can't be sent by mistake.&lt;/p&gt;

&lt;h2&gt;
  
  
  Scheduling without a Pinterest developer app
&lt;/h2&gt;

&lt;p&gt;A new Pinterest developer app starts on trial access, where pins are visible only to you; public pins need an app review. The skill schedules through WoopSocial instead, a social media API and MCP server for agents, made by the team that maintains the skill. Here is &lt;a href="https://woopsocial.com/social-media-mcp/pinterest-mcp?utm_source=pinterest-skill&amp;amp;utm_medium=devto" rel="noopener noreferrer"&gt;how scheduling pins through WoopSocial works&lt;/a&gt;. The agent uploads each pin, validates every request, and only then shows you the preview. It schedules after a clear yes.&lt;/p&gt;

&lt;p&gt;WoopSocial's Pinterest fields are the board, title, description, link and one image or video. Alt text isn't one of them yet, so &lt;code&gt;schedule.md&lt;/code&gt; lists each pin's alt text to paste in Pinterest after it goes live.&lt;/p&gt;

&lt;h2&gt;
  
  
  Results, from day 30
&lt;/h2&gt;

&lt;p&gt;&lt;code&gt;log.py results&lt;/code&gt; reads the numbers WoopSocial collects for each pin, plus a CSV from Pinterest Analytics if you have one. It ranks designs for the same page by outbound clicks per 1,000 impressions, then saves per 1,000, and only ranks pins from day 30 with at least 1,000 impressions. Pinterest says there's no set engagement window and older pins can take off later, so the advice it gives is about the next pins, never about deleting the slow ones.&lt;/p&gt;

&lt;h2&gt;
  
  
  Try it
&lt;/h2&gt;



&lt;div class="highlight js-code-highlight"&gt;
&lt;pre class="highlight shell"&gt;&lt;code&gt;npx skills add social-media-skills/pinterest-skill &lt;span class="nt"&gt;-g&lt;/span&gt; &lt;span class="nt"&gt;-a&lt;/span&gt; claude-code &lt;span class="nt"&gt;-y&lt;/span&gt;
&lt;/code&gt;&lt;/pre&gt;

&lt;/div&gt;



&lt;p&gt;Then tell your agent "make Pinterest pins for this post" with a link and a folder of your photos. The repo is &lt;a href="https://github.com/social-media-skills/pinterest-skill" rel="noopener noreferrer"&gt;social-media-skills/pinterest-skill&lt;/a&gt;. Issues and pull requests are welcome.&lt;/p&gt;

</description>
      <category>claudecode</category>
      <category>pinterest</category>
      <category>python</category>
      <category>showdev</category>
    </item>
    <item>
      <title>TikTok Skill for Claude Code: Make, Check, Preview and Post TikTok Videos From Your Agent</title>
      <dc:creator>Frank Heijdenrijk</dc:creator>
      <pubDate>Thu, 05 Nov 2026 16:44:08 +0000</pubDate>
      <link>https://dev.to/fheijdenrijk/tiktok-skill-for-claude-code-make-check-preview-and-post-tiktok-videos-from-your-agent-42e6</link>
      <guid>https://dev.to/fheijdenrijk/tiktok-skill-for-claude-code-make-check-preview-and-post-tiktok-videos-from-your-agent-42e6</guid>
      <description>&lt;p&gt;TikTok skill for Claude Code, free and MIT licensed: TikTok Skill lets your agent turn an idea, notes, your clips or your photos into finished TikTok videos and photo posts, check every file and setting against TikTok's posting rules, show you each post as the For You feed shows it, and schedule them after you say yes. It also works in Codex, Cursor and OpenClaw.&lt;/p&gt;

&lt;p&gt;Everything up to posting runs on your own computer with Python, Pillow and ffmpeg. There is no paid API. This post covers what the skill does, how to install it, the example in the repo, and then the parts that are code: the checker, the video tools, inbox drafts, per-account versions and the learning loop.&lt;br&gt;
&lt;/p&gt;

&lt;div class="highlight js-code-highlight"&gt;
&lt;pre class="highlight shell"&gt;&lt;code&gt;npx skills add social-media-skills/tiktok-skill &lt;span class="nt"&gt;-g&lt;/span&gt; &lt;span class="nt"&gt;-a&lt;/span&gt; claude-code &lt;span class="nt"&gt;-y&lt;/span&gt;
&lt;/code&gt;&lt;/pre&gt;

&lt;/div&gt;



&lt;p&gt;Repo: &lt;a href="https://github.com/social-media-skills/tiktok-skill" rel="noopener noreferrer"&gt;github.com/social-media-skills/tiktok-skill&lt;/a&gt;&lt;/p&gt;

&lt;p&gt;&lt;a href="https://media2.dev.to/dynamic/image/width=800%2Cheight=%2Cfit=scale-down%2Cgravity=auto%2Cformat=auto/https%3A%2F%2Fdev-to-uploads.s3.us-east-2.amazonaws.com%2Fuploads%2Farticles%2F81rnylu51gv80c6k8te6.png" class="article-body-image-wrapper"&gt;&lt;img src="https://media2.dev.to/dynamic/image/width=800%2Cheight=%2Cfit=scale-down%2Cgravity=auto%2Cformat=auto/https%3A%2F%2Fdev-to-uploads.s3.us-east-2.amazonaws.com%2Fuploads%2Farticles%2F81rnylu51gv80c6k8te6.png" alt="The example made with the TikTok skill: a video cut from five NASA/JPL shots with the hook &amp;quot;Only 2.5% of Earth's water is fresh&amp;quot;, and a photo post asking how big all of Earth's water would be as one drop, in English and Spanish, each shown as the For You feed shows it" width="800" height="325"&gt;&lt;/a&gt;&lt;/p&gt;

&lt;p&gt;&lt;em&gt;The example in the repo, for "Earth in Numbers", a made-up science account. Footage and photos: NASA/JPL-Caltech and NASA; figures: USGS and NASA (no endorsement implied).&lt;/em&gt;&lt;/p&gt;

&lt;h2&gt;
  
  
  What is a TikTok skill?
&lt;/h2&gt;

&lt;p&gt;An agent skill is a folder with a &lt;code&gt;SKILL.md&lt;/code&gt; that tells an agent like Claude Code how to do a job, plus the scripts it runs. Most Claude TikTok help stops at text: a hook, a script, some hashtags. Then the upload fails on something nobody checked: a 50 fps export the API refuses, a PNG in a photo post, a video longer than the account may post, or a missing privacy choice, which TikTok requires the user to make every time.&lt;/p&gt;

&lt;p&gt;This skill ships the scripts that make and check the posts, so the agent hands you real files:&lt;/p&gt;

&lt;ul&gt;
&lt;li&gt;videos at 1080x1920, H.264 and AAC, constant frame rate, with the hook in TikTok's own text style;&lt;/li&gt;
&lt;li&gt;several shots cut into one 9:16 video, each cropped where its subject is, with timed points on screen;&lt;/li&gt;
&lt;li&gt;photo posts rendered as 1080x1920 JPEGs, with the payoff on its own slide;&lt;/li&gt;
&lt;li&gt;a cover taken from the first second, because WoopSocial adds the cover as the video's first frame;&lt;/li&gt;
&lt;li&gt;a preview of every post as the For You feed shows it, with TikTok's posting panel;&lt;/li&gt;
&lt;li&gt;a schedule per account in its own time zone, and one validated request per post and account.&lt;/li&gt;
&lt;/ul&gt;

&lt;h2&gt;
  
  
  Install
&lt;/h2&gt;

&lt;p&gt;Claude Code:&lt;br&gt;
&lt;/p&gt;

&lt;div class="highlight js-code-highlight"&gt;
&lt;pre class="highlight shell"&gt;&lt;code&gt;npx skills add social-media-skills/tiktok-skill &lt;span class="nt"&gt;-g&lt;/span&gt; &lt;span class="nt"&gt;-a&lt;/span&gt; claude-code &lt;span class="nt"&gt;-y&lt;/span&gt;
&lt;/code&gt;&lt;/pre&gt;

&lt;/div&gt;



&lt;p&gt;Codex, Cursor and other agents: &lt;code&gt;npx skills add social-media-skills/tiktok-skill&lt;/code&gt;, then pick your agent. OpenClaw: &lt;code&gt;openclaw skills install @social-media-skills/tiktok-skill&lt;/code&gt;. Gemini CLI: &lt;code&gt;gemini extensions install https://github.com/social-media-skills/tiktok-skill&lt;/code&gt;.&lt;/p&gt;

&lt;p&gt;It needs Python 3.9+ with Pillow, and ffmpeg for video. &lt;code&gt;scripts/check_setup.py&lt;/code&gt; tells you what's missing and prints the install command for your OS.&lt;/p&gt;

&lt;h2&gt;
  
  
  The example
&lt;/h2&gt;

&lt;p&gt;&lt;code&gt;examples/fresh-water/&lt;/code&gt; is a full batch for a made-up science account with an English profile (New York) and a Spanish one (Madrid):&lt;/p&gt;

&lt;ul&gt;
&lt;li&gt;
&lt;strong&gt;A video&lt;/strong&gt;, 15 seconds, cut from five shots of NASA/JPL's Earth Water Budget animation. No NASA narration, no NASA labels: its own points on screen from USGS figures. "Only 2.5% of Earth's water is fresh", "Two thirds of it is ice", "Almost a third is underground", "Lakes, rivers and the rest: about 1%", "That's all the fresh water we can reach". It goes to the TikTok inbox as a draft, so a sound can be added in the app.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;A photo post&lt;/strong&gt;: "If all of Earth's water were one drop, how big would it be?" Five swipes later the answer, 860 miles wide, gets a slide of its own, then "less than half as wide as the Moon". The caption repeats the question, not the answer. The Spanish account gets its own slides in km (1.385 km de ancho), not a translation pasted over the English ones.&lt;/li&gt;
&lt;/ul&gt;

&lt;p&gt;Every slide, check, preview, schedule and request in that folder was made by the skill's scripts.&lt;/p&gt;

&lt;h2&gt;
  
  
  The checker: TikTok's posting rules in one file
&lt;/h2&gt;

&lt;p&gt;&lt;code&gt;tt_check.py&lt;/code&gt; reads &lt;code&gt;plan.json&lt;/code&gt; (the batch: accounts, posts, settings) and checks each file and target. A few rules that cost people uploads:&lt;/p&gt;

&lt;div class="table-wrapper-paragraph"&gt;&lt;table&gt;
&lt;thead&gt;
&lt;tr&gt;
&lt;th&gt;Rule&lt;/th&gt;
&lt;th&gt;Why it matters&lt;/th&gt;
&lt;/tr&gt;
&lt;/thead&gt;
&lt;tbody&gt;
&lt;tr&gt;
&lt;td&gt;23 to 60 fps, constant&lt;/td&gt;
&lt;td&gt;Other frame rates fail with &lt;code&gt;frame_rate_check_failed&lt;/code&gt;
&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;H.264 (or H.265, VP8, VP9), 8-bit&lt;/td&gt;
&lt;td&gt;10-bit and odd codecs get rejected or re-encoded badly&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;Length up to the account's own maximum&lt;/td&gt;
&lt;td&gt;Many accounts may post 3 minutes through the API; it reads the account's limit from WoopSocial&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;JPEG or WebP photos, up to 1080p and 20 MB&lt;/td&gt;
&lt;td&gt;PNG is rejected for photo posts&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;Privacy chosen by the user, no default&lt;/td&gt;
&lt;td&gt;TikTok's rule for every posting app; the plan records &lt;code&gt;"privacy_from": "user"&lt;/code&gt;
&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;Branded content never private&lt;/td&gt;
&lt;td&gt;TikTok refuses "Paid partnership" with "only me"&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;Interactions off unless turned on&lt;/td&gt;
&lt;td&gt;Another of TikTok's posting-app rules&lt;/td&gt;
&lt;/tr&gt;
&lt;/tbody&gt;
&lt;/table&gt;&lt;/div&gt;

&lt;p&gt;&lt;code&gt;--fix&lt;/code&gt; writes fixed copies to &lt;code&gt;fixed/&lt;/code&gt; (frame rate, codecs, shape, PNG to JPEG) and keeps the originals. A landscape video that would lose a lot of its width is an error, not a silent crop: the agent shows you the frames and asks.&lt;/p&gt;

&lt;p&gt;The checker also knows things a file probe can't: the same media on two accounts (by content hash, so renaming doesn't hide it), two posts on one account within a week that teach the same fact, and a single clip you didn't film that keeps its own audio, which TikTok treats as reused material.&lt;/p&gt;

&lt;h2&gt;
  
  
  Footage you didn't film
&lt;/h2&gt;

&lt;p&gt;Public domain means you may use a clip. It doesn't mean TikTok counts it as yours. A hook plate on someone else's single clip is the easiest thing for TikTok to match, especially when the original audio is still there.&lt;/p&gt;

&lt;p&gt;So the skill treats such footage as material for an edit of your own:&lt;br&gt;
&lt;/p&gt;

&lt;div class="highlight js-code-highlight"&gt;
&lt;pre class="highlight shell"&gt;&lt;code&gt;python3 scripts/video.py montage clips.json &lt;span class="nt"&gt;--out&lt;/span&gt; video/montage.mp4
python3 scripts/video.py text video/montage.mp4 &lt;span class="nt"&gt;--beats&lt;/span&gt; beats.json &lt;span class="nt"&gt;--out&lt;/span&gt; video/fresh-water.mp4
&lt;/code&gt;&lt;/pre&gt;

&lt;/div&gt;



&lt;p&gt;&lt;code&gt;clips.json&lt;/code&gt; lists the shots with a start, an end and &lt;code&gt;x&lt;/code&gt;, where the subject sits (0 left, 0.5 middle, 1 right), so each shot is cropped around its subject instead of the middle. &lt;code&gt;beats.json&lt;/code&gt; holds the timed points. Each point picks TikTok's classic outlined text or the highlight style, depending on how bright the background behind it is, and stays clear of the buttons and caption.&lt;/p&gt;

&lt;h2&gt;
  
  
  Sounds and inbox drafts
&lt;/h2&gt;

&lt;p&gt;No posting API can attach a TikTok sound. For photo posts, &lt;code&gt;autoAddMusic&lt;/code&gt; lets TikTok add recommended music on a direct post. For anything that needs a chosen sound, the skill sends the post to the account's TikTok inbox as a draft (&lt;code&gt;postMode: MEDIA_UPLOAD&lt;/code&gt;).&lt;/p&gt;

&lt;p&gt;Two details make drafts usable. A draft lands at its scheduled time, not at the time it should go live, so &lt;code&gt;schedule.py&lt;/code&gt; sends it 2 hours early. And video drafts arrive without a caption, so &lt;code&gt;schedule.md&lt;/code&gt; lists each draft with the caption to paste. Privacy and &lt;code&gt;autoAddMusic&lt;/code&gt; are left out of the draft request: you pick them in the app.&lt;/p&gt;

&lt;h2&gt;
  
  
  Posting through the TikTok MCP
&lt;/h2&gt;

&lt;p&gt;Posting goes through &lt;a href="https://woopsocial.com/?utm_source=tiktok-skill&amp;amp;utm_medium=devto" rel="noopener noreferrer"&gt;WoopSocial&lt;/a&gt;, a social media API and MCP server for AI agents. Its &lt;a href="https://woopsocial.com/social-media-mcp/tiktok-mcp?utm_source=tiktok-skill&amp;amp;utm_medium=devto" rel="noopener noreferrer"&gt;TikTok MCP server&lt;/a&gt; posts videos and photo posts, directly or as inbox drafts, now or scheduled, with a cover, privacy, interaction settings, Promotional content and Paid partnership labels, and the AI label on videos.&lt;/p&gt;

&lt;p&gt;This matters for anyone who searched for a TikTok developer API key: TikTok's Content Posting API only lets unaudited apps post "only me", for at most 5 users a day. WoopSocial's app is audited, so posts can be public from the start, and you don't need a developer account. Disclosure: the skill is maintained by Frank Heijdenrijk and the team behind WoopSocial. The free plan covers 2 social accounts.&lt;/p&gt;

&lt;p&gt;The agent uploads each file, writes one request per post and account, validates them, and posts only after you say yes on the preview. WoopSocial doesn't write captions or edit media; the skill does that on your computer.&lt;/p&gt;

&lt;h2&gt;
  
  
  Several accounts
&lt;/h2&gt;

&lt;p&gt;Each account gets its own version of a post: another hook, cut, order or language written for that country (in Spain "grifo", in Mexico "llave"; metric units; decimal commas). Two accounts in the same language get their own idea, not the same slides with new words. Each account posts at its own times in its own time zone.&lt;/p&gt;

&lt;h2&gt;
  
  
  Learning from your own results
&lt;/h2&gt;

&lt;p&gt;After a few days, &lt;code&gt;log.py results&lt;/code&gt; reads the post numbers WoopSocial collects and measures each post at the value nearest to day 7, so posts from different weeks are compared at the same age. It ranks against the same account's posts of the same type:&lt;/p&gt;

&lt;ol&gt;
&lt;li&gt;views against that account's median (on TikTok, views show how far TikTok pushed the post);&lt;/li&gt;
&lt;li&gt;retention for videos (average watch time divided by length);&lt;/li&gt;
&lt;li&gt;shares, saves and follows per 1,000 views.&lt;/li&gt;
&lt;/ol&gt;

&lt;p&gt;It names a best post only after 6 of a type. Saves, follows and watch time that the API doesn't report come from TikTok Studio, via &lt;code&gt;insights.csv&lt;/code&gt;. Missing numbers are "not reported", never zero.&lt;/p&gt;

&lt;h2&gt;
  
  
  What it won't do
&lt;/h2&gt;

&lt;ul&gt;
&lt;li&gt;Invent numbers, results, views or quotes.&lt;/li&gt;
&lt;li&gt;Post anything without your yes, or pick your privacy for you.&lt;/li&gt;
&lt;li&gt;Repost someone else's video as is, or crop your landscape video without asking.&lt;/li&gt;
&lt;li&gt;Add a sound through the API, or give a generic "best time to post".&lt;/li&gt;
&lt;/ul&gt;

&lt;p&gt;The repo is &lt;a href="https://github.com/social-media-skills/tiktok-skill" rel="noopener noreferrer"&gt;social-media-skills/tiktok-skill&lt;/a&gt;. Issues and pull requests are welcome, especially when a TikTok limit in &lt;code&gt;references/tiktok-specs.md&lt;/code&gt; changes. More free skills from the same team are at &lt;a href="https://social-media-skills.com/?utm_source=tiktok-skill&amp;amp;utm_medium=devto" rel="noopener noreferrer"&gt;social-media-skills.com&lt;/a&gt;.&lt;/p&gt;

</description>
      <category>claudecode</category>
      <category>tiktok</category>
      <category>python</category>
      <category>showdev</category>
    </item>
    <item>
      <title>Instagram Skill for Claude Code: Make, Check, Preview and Schedule Instagram Posts From Your Agent</title>
      <dc:creator>Frank Heijdenrijk</dc:creator>
      <pubDate>Wed, 04 Nov 2026 12:47:24 +0000</pubDate>
      <link>https://dev.to/fheijdenrijk/instagram-skill-for-claude-code-make-check-preview-and-schedule-instagram-posts-from-your-agent-2m9j</link>
      <guid>https://dev.to/fheijdenrijk/instagram-skill-for-claude-code-make-check-preview-and-schedule-instagram-posts-from-your-agent-2m9j</guid>
      <description>&lt;p&gt;Instagram skill for Claude Code, free and MIT licensed: Instagram Skill lets your agent turn an idea, a link, notes, your photos or a video into finished Instagram posts (carousels, single images, Reels and Stories), check every file against Instagram's current limits, show you the feed and the grid, and schedule them after you say yes. It also works in Codex, Cursor and OpenClaw.&lt;/p&gt;

&lt;p&gt;Everything up to posting runs on your own computer with Python, Pillow and ffmpeg. There is no paid API. This post covers what the skill does, how to install it, the example in the repo, and then the parts that are code: the file checker, the preview, per-account versions and time zones, and the rules it uses to learn from your results.&lt;br&gt;
&lt;/p&gt;

&lt;div class="highlight js-code-highlight"&gt;
&lt;pre class="highlight shell"&gt;&lt;code&gt;npx skills add social-media-skills/instagram-skill &lt;span class="nt"&gt;-g&lt;/span&gt; &lt;span class="nt"&gt;-a&lt;/span&gt; claude-code &lt;span class="nt"&gt;-y&lt;/span&gt;
&lt;/code&gt;&lt;/pre&gt;

&lt;/div&gt;



&lt;p&gt;Repo: &lt;a href="https://github.com/social-media-skills/instagram-skill" rel="noopener noreferrer"&gt;github.com/social-media-skills/instagram-skill&lt;/a&gt;&lt;/p&gt;

&lt;p&gt;&lt;a href="https://media2.dev.to/dynamic/image/width=800%2Cheight=%2Cfit=scale-down%2Cgravity=auto%2Cformat=auto/https%3A%2F%2Fdev-to-uploads.s3.us-east-2.amazonaws.com%2Fuploads%2Farticles%2Fl4dclh1li6kggqk5bk25.png" class="article-body-image-wrapper"&gt;&lt;img src="https://media2.dev.to/dynamic/image/width=800%2Cheight=%2Cfit=scale-down%2Cgravity=auto%2Cformat=auto/https%3A%2F%2Fdev-to-uploads.s3.us-east-2.amazonaws.com%2Fuploads%2Farticles%2Fl4dclh1li6kggqk5bk25.png" alt="An example week made with the Instagram skill: a carousel on all of Earth's water in a 1-liter bottle in English and in Spanish, a Story teaser an hour before it, and a Reel with its chosen cover (6x more water floats in the air than flows in every river), shown as feed posts next to the profile grid"&gt;&lt;/a&gt;&lt;/p&gt;

&lt;p&gt;&lt;em&gt;The example week in the repo, for "Earth in Numbers", a made-up science account. Figures from the USGS Water Science School; Earth photo NASA Blue Marble 2012 (no endorsement by NASA or USGS is implied). The Reel is the account's own 16-second animated short, made for the example from the same USGS figures.&lt;/em&gt;&lt;/p&gt;

&lt;h2&gt;
  
  
  What is an Instagram skill?
&lt;/h2&gt;

&lt;p&gt;An agent skill is a folder with a &lt;code&gt;SKILL.md&lt;/code&gt; that tells an agent like Claude Code how to do a job. Most Claude Instagram help stops at text: a hook, a caption, some hashtags, and "now design the post yourself". Then the post fails on details nobody checked: a PNG the API won't take, a 3:4 photo it rejects, a Reel title under the caption, a cover the grid cuts in half, six hashtags where Instagram now allows five.&lt;/p&gt;

&lt;p&gt;This skill ships the scripts that make and check the posts, so the agent produces real files, not advice:&lt;/p&gt;

&lt;ul&gt;
&lt;li&gt;carousels and single images rendered as 1080x1350 JPEGs, with a cover that survives the 3:4 profile grid;&lt;/li&gt;
&lt;li&gt;your own photos and screenshots converted and fitted (PNG, HEIC, 3:4, panoramas);&lt;/li&gt;
&lt;li&gt;your own videos finished as Reels: 9:16, H.264 and AAC, a chosen cover, and a check of what Instagram's caption and buttons cover;&lt;/li&gt;
&lt;li&gt;Story frames that tease a post before it goes out;&lt;/li&gt;
&lt;li&gt;one schedule for several accounts, each with its own version of the post, its own caption and its own time zone;&lt;/li&gt;
&lt;li&gt;results read back and ranked against your own posts of the same format.&lt;/li&gt;
&lt;/ul&gt;

&lt;h2&gt;
  
  
  Instagram MCP or Instagram API: how posting works
&lt;/h2&gt;

&lt;p&gt;Posting through the Instagram API yourself means setting up a Meta app and its permissions. The skill posts through &lt;a href="https://woopsocial.com/?utm_source=instagram-skill&amp;amp;utm_medium=devto" rel="noopener noreferrer"&gt;WoopSocial&lt;/a&gt;'s &lt;a href="https://woopsocial.com/social-media-mcp/instagram-mcp?utm_source=instagram-skill&amp;amp;utm_medium=devto" rel="noopener noreferrer"&gt;Instagram MCP&lt;/a&gt; instead, so your agent gets Instagram tools without a Meta app of your own. Disclosure: the skill is maintained by Frank Heijdenrijk and the team behind WoopSocial. The free plan covers 2 social accounts. Setup for every agent is at &lt;a href="https://woopsocial.com/agents?utm_source=instagram-skill&amp;amp;utm_medium=devto" rel="noopener noreferrer"&gt;woopsocial.com/agents&lt;/a&gt;.&lt;/p&gt;

&lt;p&gt;An MCP alone gives the agent a "post" call. The skill adds what comes before it: making the media, checking it, the preview, per-account versions and the learning loop. Through WoopSocial it can post feed images, carousels, Reels (with a cover, and as Trial Reels when your account qualifies) and Stories, now or scheduled, and validate each post first.&lt;/p&gt;

&lt;p&gt;Some things can't be set this way: collaborators, location, music, alt text and first comments. The skill writes them into a checklist per post so you add them in the app. WoopSocial doesn't write captions or edit media; the skill does that on your computer and you approve it. If you'd rather not connect anything, you get the files, captions and schedule to post yourself.&lt;/p&gt;

&lt;p&gt;Instagram posting through any API needs a Business or Creator account. Switching is free in the app.&lt;/p&gt;

&lt;h2&gt;
  
  
  Install
&lt;/h2&gt;

&lt;p&gt;&lt;strong&gt;Claude Code&lt;/strong&gt;&lt;br&gt;
&lt;/p&gt;

&lt;div class="highlight js-code-highlight"&gt;
&lt;pre class="highlight shell"&gt;&lt;code&gt;npx skills add social-media-skills/instagram-skill &lt;span class="nt"&gt;-g&lt;/span&gt; &lt;span class="nt"&gt;-a&lt;/span&gt; claude-code &lt;span class="nt"&gt;-y&lt;/span&gt;
&lt;/code&gt;&lt;/pre&gt;

&lt;/div&gt;



&lt;p&gt;&lt;strong&gt;Codex, Cursor and other agents:&lt;/strong&gt; run &lt;code&gt;npx skills add social-media-skills/instagram-skill&lt;/code&gt; and pick your agent, or copy &lt;code&gt;skills/instagram-skill/&lt;/code&gt; into your agent's skills folder.&lt;/p&gt;

&lt;p&gt;&lt;strong&gt;OpenClaw:&lt;/strong&gt; &lt;code&gt;openclaw skills install @social-media-skills/instagram-skill&lt;/code&gt;&lt;/p&gt;

&lt;p&gt;Requirements: Python 3.9 or newer with Pillow, and ffmpeg for anything with video. &lt;code&gt;pillow-heif&lt;/code&gt; is optional, for iPhone .heic photos. &lt;code&gt;python3 scripts/check_setup.py&lt;/code&gt; prints what's missing and the install command for your system. CI runs the full pipeline on macOS, Windows and Ubuntu.&lt;/p&gt;

&lt;h2&gt;
  
  
  The example: a week of Instagram posts in two languages
&lt;/h2&gt;

&lt;p&gt;The repo has one full run in &lt;code&gt;examples/water-week/&lt;/code&gt;. "Earth in Numbers" is a made-up science account with an English profile (@earthinnumbers, New York) and a Spanish one (@latierraennumeros, Madrid). The brief, shortened:&lt;br&gt;
&lt;/p&gt;

&lt;div class="highlight js-code-highlight"&gt;
&lt;pre class="highlight plaintext"&gt;&lt;code&gt;You:   A week of posts for @earthinnumbers (New York, followers online
       around 12:30) and the Spanish account @latierraennumeros (Madrid,
       13:30) about Earth's water. Real numbers. Our animated short about
       the air is attached.

Agent: Mon: a carousel, "All of Earth's water in a 1-liter bottle": pour
       out the oceans and 35 ml is left; every river is less than a
       tenth of a drop. Spanish slides for the Spanish account. A Story
       an hour before asks the question.
       Tue: a single image, 55% of the water under our feet is salty.
       Wed: your short as a Reel (English only), cover picked from its
       sharpest frames; nothing sits under Instagram's caption or buttons.

       Checks: everything passes. Preview: preview/index.html.
&lt;/code&gt;&lt;/pre&gt;

&lt;/div&gt;



&lt;p&gt;Each post carries its own fact, so nothing is posted twice. The carousel walks through the bottle: the cover, "35 ml is all that's left once you pour out the oceans", where those 35 ml are (ice 17.4, groundwater 16.9, permafrost 0.22, lakes 0.13, the rest 0.03 ml), "25 ml is fresh water: 17.4 ml ice, 7.6 ml underground", "0.002 ml for every river: less than a tenth of one drop", and a save ask. The single image says 55% of groundwater is saline (USGS: 12.87 of 23.4 million km³). The Reel, the account's own 16-second animated short, says 6 times more water floats in the air than flows in every river.&lt;/p&gt;

&lt;p&gt;The folder holds the plan, the slides in both languages, the Reel cover sheets, the preview page, &lt;code&gt;schedule.md&lt;/code&gt; and one WoopSocial request per post and account. Seven posts in all, and every figure comes from the USGS Water Science School with the source in the caption.&lt;/p&gt;

&lt;h2&gt;
  
  
  The checker: Instagram's limits as code
&lt;/h2&gt;

&lt;p&gt;&lt;code&gt;ig_check.py&lt;/code&gt; reads the plan and checks every file and caption. The limits sit in one dict at the top, with the date they were checked, so updating them is a one-line change:&lt;br&gt;
&lt;/p&gt;

&lt;div class="highlight js-code-highlight"&gt;
&lt;pre class="highlight python"&gt;&lt;code&gt;&lt;span class="c1"&gt;# Checked 2026-11-04 against Meta's content publishing docs and Instagram's own announcements.
&lt;/span&gt;&lt;span class="n"&gt;LIMITS&lt;/span&gt; &lt;span class="o"&gt;=&lt;/span&gt; &lt;span class="p"&gt;{&lt;/span&gt;
    &lt;span class="sh"&gt;"&lt;/span&gt;&lt;span class="s"&gt;image_min_w&lt;/span&gt;&lt;span class="sh"&gt;"&lt;/span&gt;&lt;span class="p"&gt;:&lt;/span&gt; &lt;span class="mi"&gt;320&lt;/span&gt;&lt;span class="p"&gt;,&lt;/span&gt; &lt;span class="sh"&gt;"&lt;/span&gt;&lt;span class="s"&gt;image_max_w&lt;/span&gt;&lt;span class="sh"&gt;"&lt;/span&gt;&lt;span class="p"&gt;:&lt;/span&gt; &lt;span class="mi"&gt;1440&lt;/span&gt;&lt;span class="p"&gt;,&lt;/span&gt; &lt;span class="sh"&gt;"&lt;/span&gt;&lt;span class="s"&gt;image_bytes&lt;/span&gt;&lt;span class="sh"&gt;"&lt;/span&gt;&lt;span class="p"&gt;:&lt;/span&gt; &lt;span class="mi"&gt;8&lt;/span&gt; &lt;span class="o"&gt;*&lt;/span&gt; &lt;span class="mi"&gt;1024&lt;/span&gt; &lt;span class="o"&gt;*&lt;/span&gt; &lt;span class="mi"&gt;1024&lt;/span&gt;&lt;span class="p"&gt;,&lt;/span&gt;
    &lt;span class="sh"&gt;"&lt;/span&gt;&lt;span class="s"&gt;feed_min_ratio&lt;/span&gt;&lt;span class="sh"&gt;"&lt;/span&gt;&lt;span class="p"&gt;:&lt;/span&gt; &lt;span class="mf"&gt;0.8&lt;/span&gt;&lt;span class="p"&gt;,&lt;/span&gt; &lt;span class="sh"&gt;"&lt;/span&gt;&lt;span class="s"&gt;feed_max_ratio&lt;/span&gt;&lt;span class="sh"&gt;"&lt;/span&gt;&lt;span class="p"&gt;:&lt;/span&gt; &lt;span class="mf"&gt;1.91&lt;/span&gt;&lt;span class="p"&gt;,&lt;/span&gt;           &lt;span class="c1"&gt;# width / height
&lt;/span&gt;    &lt;span class="sh"&gt;"&lt;/span&gt;&lt;span class="s"&gt;carousel_min&lt;/span&gt;&lt;span class="sh"&gt;"&lt;/span&gt;&lt;span class="p"&gt;:&lt;/span&gt; &lt;span class="mi"&gt;2&lt;/span&gt;&lt;span class="p"&gt;,&lt;/span&gt; &lt;span class="sh"&gt;"&lt;/span&gt;&lt;span class="s"&gt;carousel_max&lt;/span&gt;&lt;span class="sh"&gt;"&lt;/span&gt;&lt;span class="p"&gt;:&lt;/span&gt; &lt;span class="mi"&gt;20&lt;/span&gt;&lt;span class="p"&gt;,&lt;/span&gt;
    &lt;span class="sh"&gt;"&lt;/span&gt;&lt;span class="s"&gt;reel_min_s&lt;/span&gt;&lt;span class="sh"&gt;"&lt;/span&gt;&lt;span class="p"&gt;:&lt;/span&gt; &lt;span class="mi"&gt;3&lt;/span&gt;&lt;span class="p"&gt;,&lt;/span&gt; &lt;span class="sh"&gt;"&lt;/span&gt;&lt;span class="s"&gt;reel_max_s&lt;/span&gt;&lt;span class="sh"&gt;"&lt;/span&gt;&lt;span class="p"&gt;:&lt;/span&gt; &lt;span class="mi"&gt;15&lt;/span&gt; &lt;span class="o"&gt;*&lt;/span&gt; &lt;span class="mi"&gt;60&lt;/span&gt;&lt;span class="p"&gt;,&lt;/span&gt; &lt;span class="sh"&gt;"&lt;/span&gt;&lt;span class="s"&gt;reel_bytes&lt;/span&gt;&lt;span class="sh"&gt;"&lt;/span&gt;&lt;span class="p"&gt;:&lt;/span&gt; &lt;span class="mi"&gt;300&lt;/span&gt; &lt;span class="o"&gt;*&lt;/span&gt; &lt;span class="mi"&gt;1024&lt;/span&gt; &lt;span class="o"&gt;*&lt;/span&gt; &lt;span class="mi"&gt;1024&lt;/span&gt;&lt;span class="p"&gt;,&lt;/span&gt;
    &lt;span class="sh"&gt;"&lt;/span&gt;&lt;span class="s"&gt;story_min_s&lt;/span&gt;&lt;span class="sh"&gt;"&lt;/span&gt;&lt;span class="p"&gt;:&lt;/span&gt; &lt;span class="mi"&gt;3&lt;/span&gt;&lt;span class="p"&gt;,&lt;/span&gt; &lt;span class="sh"&gt;"&lt;/span&gt;&lt;span class="s"&gt;story_max_s&lt;/span&gt;&lt;span class="sh"&gt;"&lt;/span&gt;&lt;span class="p"&gt;:&lt;/span&gt; &lt;span class="mi"&gt;60&lt;/span&gt;&lt;span class="p"&gt;,&lt;/span&gt; &lt;span class="sh"&gt;"&lt;/span&gt;&lt;span class="s"&gt;story_bytes&lt;/span&gt;&lt;span class="sh"&gt;"&lt;/span&gt;&lt;span class="p"&gt;:&lt;/span&gt; &lt;span class="mi"&gt;100&lt;/span&gt; &lt;span class="o"&gt;*&lt;/span&gt; &lt;span class="mi"&gt;1024&lt;/span&gt; &lt;span class="o"&gt;*&lt;/span&gt; &lt;span class="mi"&gt;1024&lt;/span&gt;&lt;span class="p"&gt;,&lt;/span&gt;
    &lt;span class="sh"&gt;"&lt;/span&gt;&lt;span class="s"&gt;carousel_video_max_s&lt;/span&gt;&lt;span class="sh"&gt;"&lt;/span&gt;&lt;span class="p"&gt;:&lt;/span&gt; &lt;span class="mi"&gt;60&lt;/span&gt;&lt;span class="p"&gt;,&lt;/span&gt;
    &lt;span class="sh"&gt;"&lt;/span&gt;&lt;span class="s"&gt;fps_min&lt;/span&gt;&lt;span class="sh"&gt;"&lt;/span&gt;&lt;span class="p"&gt;:&lt;/span&gt; &lt;span class="mi"&gt;23&lt;/span&gt;&lt;span class="p"&gt;,&lt;/span&gt; &lt;span class="sh"&gt;"&lt;/span&gt;&lt;span class="s"&gt;fps_max&lt;/span&gt;&lt;span class="sh"&gt;"&lt;/span&gt;&lt;span class="p"&gt;:&lt;/span&gt; &lt;span class="mi"&gt;60&lt;/span&gt;&lt;span class="p"&gt;,&lt;/span&gt;
    &lt;span class="sh"&gt;"&lt;/span&gt;&lt;span class="s"&gt;caption_chars&lt;/span&gt;&lt;span class="sh"&gt;"&lt;/span&gt;&lt;span class="p"&gt;:&lt;/span&gt; &lt;span class="mi"&gt;2200&lt;/span&gt;&lt;span class="p"&gt;,&lt;/span&gt; &lt;span class="sh"&gt;"&lt;/span&gt;&lt;span class="s"&gt;hashtags&lt;/span&gt;&lt;span class="sh"&gt;"&lt;/span&gt;&lt;span class="p"&gt;:&lt;/span&gt; &lt;span class="mi"&gt;5&lt;/span&gt;&lt;span class="p"&gt;,&lt;/span&gt; &lt;span class="sh"&gt;"&lt;/span&gt;&lt;span class="s"&gt;mentions&lt;/span&gt;&lt;span class="sh"&gt;"&lt;/span&gt;&lt;span class="p"&gt;:&lt;/span&gt; &lt;span class="mi"&gt;20&lt;/span&gt;&lt;span class="p"&gt;,&lt;/span&gt; &lt;span class="sh"&gt;"&lt;/span&gt;&lt;span class="s"&gt;fold_chars&lt;/span&gt;&lt;span class="sh"&gt;"&lt;/span&gt;&lt;span class="p"&gt;:&lt;/span&gt; &lt;span class="mi"&gt;125&lt;/span&gt;&lt;span class="p"&gt;,&lt;/span&gt;
&lt;span class="p"&gt;}&lt;/span&gt;
&lt;/code&gt;&lt;/pre&gt;

&lt;/div&gt;



&lt;p&gt;A few of these surprise people. The API takes JPEG only, so a PNG screenshot or an iPhone HEIC fails. Feed images must be between 4:5 and 1.91:1, so a 3:4 photo (which the app accepts) is rejected through the API. Every carousel item is shown at the first item's shape. Five hashtags is Instagram's cap since December 2025.&lt;/p&gt;

&lt;p&gt;Videos go through &lt;code&gt;ffprobe&lt;/code&gt;: container, codec, 8-bit pixel format, HDR, frame rate, length, audio and size. 10-bit and HDR phone video is a common cause of grey or failed Reels, so the checker flags it and re-encodes it. Screen recordings with a variable frame rate get a warning, because re-encoding them to a constant 30 fps avoids audio drift.&lt;/p&gt;

&lt;p&gt;Captions are checked for length, hashtags, mentions, links (not clickable in captions), a first line longer than the roughly 125 characters the feed shows before "... more", and two accounts opening with the same line.&lt;/p&gt;

&lt;h3&gt;
  
  
  Fixes are new files
&lt;/h3&gt;

&lt;p&gt;&lt;code&gt;--fix&lt;/code&gt; never touches your originals. It writes fixed copies to &lt;code&gt;fixed/&lt;/code&gt;, points the plan at them and keeps a &lt;code&gt;plan.json.bak&lt;/code&gt;:&lt;br&gt;
&lt;/p&gt;

&lt;div class="highlight js-code-highlight"&gt;
&lt;pre class="highlight shell"&gt;&lt;code&gt;python3 scripts/ig_check.py plan.json          &lt;span class="c"&gt;# report&lt;/span&gt;
python3 scripts/ig_check.py plan.json &lt;span class="nt"&gt;--fix&lt;/span&gt;    &lt;span class="c"&gt;# fixed copies in fixed/, plan updated, originals kept&lt;/span&gt;
&lt;/code&gt;&lt;/pre&gt;

&lt;/div&gt;



&lt;p&gt;The exit code is 1 while errors remain, so the agent runs it until it gets 0. Anything it can't fix safely (a video that needs trimming, a photo too small to scale up) comes back as an error with the reason in plain words.&lt;/p&gt;

&lt;h3&gt;
  
  
  Your own landscape footage is framed, not cropped blind
&lt;/h3&gt;

&lt;p&gt;The interesting case is landscape footage for a Reel. Filling 9:16 from 16:9 throws away most of the width, and padding with a blurred copy looks like a repost. So the checker refuses to guess once the loss is 12% or more:&lt;br&gt;
&lt;/p&gt;

&lt;div class="highlight js-code-highlight"&gt;
&lt;pre class="highlight python"&gt;&lt;code&gt;&lt;span class="k"&gt;if&lt;/span&gt; &lt;span class="n"&gt;kind&lt;/span&gt; &lt;span class="o"&gt;==&lt;/span&gt; &lt;span class="sh"&gt;"&lt;/span&gt;&lt;span class="s"&gt;video&lt;/span&gt;&lt;span class="sh"&gt;"&lt;/span&gt; &lt;span class="ow"&gt;and&lt;/span&gt; &lt;span class="n"&gt;fmt&lt;/span&gt; &lt;span class="ow"&gt;in&lt;/span&gt; &lt;span class="p"&gt;(&lt;/span&gt;&lt;span class="sh"&gt;"&lt;/span&gt;&lt;span class="s"&gt;reel&lt;/span&gt;&lt;span class="sh"&gt;"&lt;/span&gt;&lt;span class="p"&gt;,&lt;/span&gt; &lt;span class="sh"&gt;"&lt;/span&gt;&lt;span class="s"&gt;story&lt;/span&gt;&lt;span class="sh"&gt;"&lt;/span&gt;&lt;span class="p"&gt;)&lt;/span&gt; &lt;span class="ow"&gt;and&lt;/span&gt; &lt;span class="n"&gt;loss&lt;/span&gt; &lt;span class="o"&gt;&amp;gt;=&lt;/span&gt; &lt;span class="mf"&gt;0.12&lt;/span&gt;&lt;span class="p"&gt;:&lt;/span&gt;
    &lt;span class="nf"&gt;report&lt;/span&gt;&lt;span class="p"&gt;(&lt;/span&gt;&lt;span class="sh"&gt;"&lt;/span&gt;&lt;span class="s"&gt;error&lt;/span&gt;&lt;span class="sh"&gt;"&lt;/span&gt;&lt;span class="p"&gt;,&lt;/span&gt; &lt;span class="n"&gt;pid&lt;/span&gt;&lt;span class="p"&gt;,&lt;/span&gt; &lt;span class="n"&gt;label&lt;/span&gt;&lt;span class="p"&gt;,&lt;/span&gt; &lt;span class="sa"&gt;f&lt;/span&gt;&lt;span class="sh"&gt;"&lt;/span&gt;&lt;span class="s"&gt;this video would lose &lt;/span&gt;&lt;span class="si"&gt;{&lt;/span&gt;&lt;span class="n"&gt;loss&lt;/span&gt;&lt;span class="si"&gt;:&lt;/span&gt;&lt;span class="p"&gt;.&lt;/span&gt;&lt;span class="mi"&gt;0&lt;/span&gt;&lt;span class="o"&gt;%&lt;/span&gt;&lt;span class="si"&gt;}&lt;/span&gt;&lt;span class="s"&gt; of its width to fill 9:16. Check reel.py frames, then: &lt;/span&gt;&lt;span class="sh"&gt;"&lt;/span&gt;
           &lt;span class="sh"&gt;"&lt;/span&gt;&lt;span class="s"&gt;reframe it on the subject in an editor; or --fit crop when the subject stays in the middle; or give the &lt;/span&gt;&lt;span class="sh"&gt;"&lt;/span&gt;
           &lt;span class="sh"&gt;"&lt;/span&gt;&lt;span class="s"&gt;post a &lt;/span&gt;&lt;span class="se"&gt;\"&lt;/span&gt;&lt;span class="s"&gt;frame&lt;/span&gt;&lt;span class="se"&gt;\"&lt;/span&gt;&lt;span class="s"&gt;: {&lt;/span&gt;&lt;span class="se"&gt;\"&lt;/span&gt;&lt;span class="s"&gt;title&lt;/span&gt;&lt;span class="se"&gt;\"&lt;/span&gt;&lt;span class="s"&gt;: &lt;/span&gt;&lt;span class="se"&gt;\"&lt;/span&gt;&lt;span class="s"&gt;...&lt;/span&gt;&lt;span class="se"&gt;\"&lt;/span&gt;&lt;span class="s"&gt;} for a framed Reel (video full width under a headline)&lt;/span&gt;&lt;span class="sh"&gt;"&lt;/span&gt;&lt;span class="p"&gt;)&lt;/span&gt;
&lt;/code&gt;&lt;/pre&gt;

&lt;/div&gt;



&lt;p&gt;A framed Reel puts the video full width on a plain 1080x1920 background in a colour sampled from the clip's edges, with the hook as a headline above it. The headline sits below the top bar (the top 14%) and the video above the caption area (the bottom 35%), the safe zones Meta publishes for Reels. Nothing is cut off.&lt;/p&gt;

&lt;p&gt;Framing is only for your own footage, so the post also needs &lt;code&gt;"own_footage": true&lt;/code&gt;. Without it the checker warns: a headline or border around someone else's clip still reads as a repost to Instagram.&lt;/p&gt;

&lt;p&gt;&lt;code&gt;reel.py&lt;/code&gt; handles the rest of a Reel: &lt;code&gt;frames&lt;/code&gt; picks sharp, well-lit cover candidates spread over the video into one sheet, &lt;code&gt;cover&lt;/code&gt; writes a 1080x1920 JPEG cover with an optional title, and &lt;code&gt;zones&lt;/code&gt; shades what Instagram's caption, buttons and top bar will cover. The example's Reel cover was picked from that frames sheet.&lt;/p&gt;

&lt;h2&gt;
  
  
  The preview: the feed cut and the 3:4 grid, per account
&lt;/h2&gt;

&lt;p&gt;&lt;code&gt;preview.py&lt;/code&gt; writes &lt;code&gt;preview/index.html&lt;/code&gt; and the same thing as one &lt;code&gt;preview.png&lt;/code&gt; the agent looks at before you do:&lt;br&gt;
&lt;/p&gt;

&lt;div class="highlight js-code-highlight"&gt;
&lt;pre class="highlight shell"&gt;&lt;code&gt;python3 scripts/preview.py plan.json &lt;span class="nt"&gt;--out&lt;/span&gt; preview/ &lt;span class="nt"&gt;--grid&lt;/span&gt; current-grid.png
&lt;/code&gt;&lt;/pre&gt;

&lt;/div&gt;



&lt;p&gt;&lt;a href="https://media2.dev.to/dynamic/image/width=800%2Cheight=%2Cfit=scale-down%2Cgravity=auto%2Cformat=auto/https%3A%2F%2Fdev-to-uploads.s3.us-east-2.amazonaws.com%2Fuploads%2Farticles%2F3vikfn18ytxxd6mj0rlw.png" class="article-body-image-wrapper"&gt;&lt;img src="https://media2.dev.to/dynamic/image/width=800%2Cheight=%2Cfit=scale-down%2Cgravity=auto%2Cformat=auto/https%3A%2F%2Fdev-to-uploads.s3.us-east-2.amazonaws.com%2Fuploads%2Farticles%2F3vikfn18ytxxd6mj0rlw.png" alt="The preview page from the example: the profile grid after the batch, then each post as a feed card on each account with its own caption cut at &amp;quot;... more&amp;quot;, the Stories full screen and the Reel at its 4:5 feed crop"&gt;&lt;/a&gt;&lt;/p&gt;

&lt;p&gt;Three things in it are easy to get wrong by eye:&lt;/p&gt;

&lt;ul&gt;
&lt;li&gt;
&lt;strong&gt;The grid crop.&lt;/strong&gt; The profile grid shows posts at 3:4 since January 2025, but the API only accepts 4:5 or wider. Each grid tile is the middle 3:4 of the post, so you see what the grid hides. The checker also warns when the grid would hide more than 12% of a cover's width.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;The "... more" cut.&lt;/strong&gt; Each feed card cuts the caption at about 125 characters, where the feed does, so you can see whether the hook made it.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Per account.&lt;/strong&gt; Every post is shown on every account with that account's own media and caption, and each account gets its own grid, newest first. &lt;code&gt;--grid&lt;/code&gt; adds a screenshot of your current grid below the new posts.&lt;/li&gt;
&lt;/ul&gt;

&lt;p&gt;Reels are shown at the middle 4:5 the main feed commonly shows (creators observe this crop; Meta doesn't document it). Stories are full screen.&lt;/p&gt;

&lt;h2&gt;
  
  
  Per-account versions and time zones
&lt;/h2&gt;

&lt;p&gt;Instagram's 2026 originality rules cover photos and carousels too, not only video, so the same media on a second account is likely seen as a repost of the first upload. In &lt;code&gt;plan.json&lt;/code&gt;, a post has default media and a caption, and each target can replace them:&lt;br&gt;
&lt;/p&gt;

&lt;div class="highlight js-code-highlight"&gt;
&lt;pre class="highlight json"&gt;&lt;code&gt;&lt;span class="nl"&gt;"targets"&lt;/span&gt;&lt;span class="p"&gt;:&lt;/span&gt;&lt;span class="w"&gt; &lt;/span&gt;&lt;span class="p"&gt;[&lt;/span&gt;&lt;span class="w"&gt;
  &lt;/span&gt;&lt;span class="p"&gt;{&lt;/span&gt;&lt;span class="nl"&gt;"account"&lt;/span&gt;&lt;span class="p"&gt;:&lt;/span&gt;&lt;span class="w"&gt; &lt;/span&gt;&lt;span class="s2"&gt;"en"&lt;/span&gt;&lt;span class="p"&gt;,&lt;/span&gt;&lt;span class="w"&gt; &lt;/span&gt;&lt;span class="nl"&gt;"at"&lt;/span&gt;&lt;span class="p"&gt;:&lt;/span&gt;&lt;span class="w"&gt; &lt;/span&gt;&lt;span class="s2"&gt;"2026-11-16T12:30"&lt;/span&gt;&lt;span class="p"&gt;},&lt;/span&gt;&lt;span class="w"&gt;
  &lt;/span&gt;&lt;span class="p"&gt;{&lt;/span&gt;&lt;span class="nl"&gt;"account"&lt;/span&gt;&lt;span class="p"&gt;:&lt;/span&gt;&lt;span class="w"&gt; &lt;/span&gt;&lt;span class="s2"&gt;"es"&lt;/span&gt;&lt;span class="p"&gt;,&lt;/span&gt;&lt;span class="w"&gt; &lt;/span&gt;&lt;span class="nl"&gt;"media"&lt;/span&gt;&lt;span class="p"&gt;:&lt;/span&gt;&lt;span class="w"&gt; &lt;/span&gt;&lt;span class="p"&gt;[&lt;/span&gt;&lt;span class="s2"&gt;"slides/bottle-es/slide-01.jpg"&lt;/span&gt;&lt;span class="p"&gt;,&lt;/span&gt;&lt;span class="w"&gt; &lt;/span&gt;&lt;span class="s2"&gt;"..."&lt;/span&gt;&lt;span class="p"&gt;],&lt;/span&gt;&lt;span class="w"&gt;
   &lt;/span&gt;&lt;span class="nl"&gt;"caption"&lt;/span&gt;&lt;span class="p"&gt;:&lt;/span&gt;&lt;span class="w"&gt; &lt;/span&gt;&lt;span class="s2"&gt;"¿Cuánta agua de la Tierra hay en los ríos? ..."&lt;/span&gt;&lt;span class="p"&gt;,&lt;/span&gt;&lt;span class="w"&gt;
   &lt;/span&gt;&lt;span class="nl"&gt;"at"&lt;/span&gt;&lt;span class="p"&gt;:&lt;/span&gt;&lt;span class="w"&gt; &lt;/span&gt;&lt;span class="s2"&gt;"2026-11-16T13:30"&lt;/span&gt;&lt;span class="p"&gt;}&lt;/span&gt;&lt;span class="w"&gt;
&lt;/span&gt;&lt;span class="p"&gt;]&lt;/span&gt;&lt;span class="w"&gt;
&lt;/span&gt;&lt;/code&gt;&lt;/pre&gt;

&lt;/div&gt;



&lt;p&gt;The checker warns when a second account has no media of its own, and when two accounts open with the same first line.&lt;/p&gt;

&lt;p&gt;Each account has its own &lt;code&gt;timezone&lt;/code&gt; and &lt;code&gt;times&lt;/code&gt;, taken from its Insights. &lt;code&gt;schedule.py&lt;/code&gt; gives every feed post a slot per account (one a day by default), converts the local time to UTC in the request and writes &lt;code&gt;schedule.md&lt;/code&gt;. The example's Reel at 12:30 in New York becomes &lt;code&gt;"scheduledFor": "2026-11-18T17:30:00Z"&lt;/code&gt;. Without times it uses 12:00 and says so at the top of the schedule. It never claims a "best time".&lt;/p&gt;

&lt;p&gt;Stories don't take a feed slot. A Story tied to a post with &lt;code&gt;"with": {"post": "bottle", "minutes": -60}&lt;/code&gt; goes out an hour before that post on each account, so the example's teaser runs at 11:30 New York time and 12:30 Madrid time.&lt;br&gt;
&lt;/p&gt;

&lt;div class="highlight js-code-highlight"&gt;
&lt;pre class="highlight shell"&gt;&lt;code&gt;python3 scripts/schedule.py plan.json &lt;span class="nt"&gt;--start&lt;/span&gt; 2026-11-16
python3 scripts/schedule.py plan.json &lt;span class="nt"&gt;--bodies-only&lt;/span&gt;   &lt;span class="c"&gt;# after upload, with media IDs&lt;/span&gt;
&lt;/code&gt;&lt;/pre&gt;

&lt;/div&gt;



&lt;p&gt;The agent uploads every file, writes the IDs into the plan, rebuilds the request bodies, validates each one, and posts only after a clear yes on the preview.&lt;/p&gt;

&lt;h2&gt;
  
  
  Learning from your own numbers
&lt;/h2&gt;

&lt;p&gt;&lt;code&gt;log.py record&lt;/code&gt; logs every post after it's created. A few days later, &lt;code&gt;log.py results&lt;/code&gt; reads the post metrics back (views, likes, comments, saves, shares and reach, plus an optional &lt;code&gt;insights.csv&lt;/code&gt; for what only the app shows, like follows and Reel watch time) and writes &lt;code&gt;results.md&lt;/code&gt;. The rules are in the script's docstring, so it doesn't mislead:&lt;/p&gt;

&lt;ul&gt;
&lt;li&gt;Rates are per reach. A post without reach is listed but not ranked.&lt;/li&gt;
&lt;li&gt;Posts are compared only within the same format and the same account. Stories are left out.&lt;/li&gt;
&lt;li&gt;Each post is measured once, at the value collected nearest to day 7 (within days 3 to 14), and that measurement is saved in &lt;code&gt;posts.jsonl&lt;/code&gt;. Posts from different weeks are compared at the same age, and the comparison grows with every run.&lt;/li&gt;
&lt;li&gt;A best post is named only once a format has 6 or more ranked posts.&lt;/li&gt;
&lt;li&gt;Posting-time advice needs 3 or more ranked posts in at least 2 time slots (weekday or weekend, morning, afternoon or evening).&lt;/li&gt;
&lt;li&gt;A missing number is "not reported", never zero.&lt;/li&gt;
&lt;/ul&gt;

&lt;p&gt;Posts are ranked by sends, saves and follows per reach against the account's own median. Sends per reach is one of the main signals Adam Mosseri (head of Instagram) has named, with watch time and likes per reach. From that ranking the agent picks the next step from a small table in &lt;code&gt;SKILL.md&lt;/code&gt;: same format with a new topic, add a checklist slide, end with a send-style ask, or change the cover first.&lt;/p&gt;

&lt;h2&gt;
  
  
  What the skill won't do
&lt;/h2&gt;

&lt;ul&gt;
&lt;li&gt;Invent numbers, results, reviews or quotes.&lt;/li&gt;
&lt;li&gt;Post anything without your yes on the preview.&lt;/li&gt;
&lt;li&gt;Give a generic "best time to post". It uses your Insights, then your results.&lt;/li&gt;
&lt;li&gt;Post the same media and caption to several accounts.&lt;/li&gt;
&lt;li&gt;Edit your footage (cuts, captions, music), or crop or pad landscape video without asking.&lt;/li&gt;
&lt;/ul&gt;

&lt;h2&gt;
  
  
  FAQ
&lt;/h2&gt;

&lt;h3&gt;
  
  
  Can Claude post to Instagram?
&lt;/h3&gt;

&lt;p&gt;Yes. With this skill and WoopSocial's MCP connected, Claude Code, Codex, Cursor and OpenClaw can post and schedule Instagram posts after you approve each batch. Without a connection, the skill still makes, checks and previews everything.&lt;/p&gt;

&lt;h3&gt;
  
  
  Is this an Instagram MCP?
&lt;/h3&gt;

&lt;p&gt;No, it's a skill that uses &lt;a href="https://woopsocial.com/social-media-mcp/instagram-mcp?utm_source=instagram-skill&amp;amp;utm_medium=devto" rel="noopener noreferrer"&gt;WoopSocial's Instagram MCP&lt;/a&gt; to post. The skill adds the media, the checks, the preview, per-account versions and the results.&lt;/p&gt;

&lt;h3&gt;
  
  
  Can I schedule Instagram posts to several accounts?
&lt;/h3&gt;

&lt;p&gt;Yes. One plan, one request per account, each with its own version of the post, its own caption and its own time zone.&lt;/p&gt;

&lt;h3&gt;
  
  
  Does it work with a personal Instagram account?
&lt;/h3&gt;

&lt;p&gt;No app can post to a personal account through Instagram's API. Switch to a free Creator or Business account in the app.&lt;/p&gt;

&lt;h3&gt;
  
  
  Why 4:5 and not 3:4?
&lt;/h3&gt;

&lt;p&gt;The grid shows 3:4, but the API only accepts 4:5 to 1.91:1. The skill renders 4:5 and keeps the cover text inside what the grid shows.&lt;/p&gt;

&lt;h2&gt;
  
  
  Try the Instagram Skill
&lt;/h2&gt;



&lt;div class="highlight js-code-highlight"&gt;
&lt;pre class="highlight shell"&gt;&lt;code&gt;npx skills add social-media-skills/instagram-skill &lt;span class="nt"&gt;-g&lt;/span&gt; &lt;span class="nt"&gt;-a&lt;/span&gt; claude-code &lt;span class="nt"&gt;-y&lt;/span&gt;
&lt;/code&gt;&lt;/pre&gt;

&lt;/div&gt;



&lt;p&gt;Then ask your agent for a week of posts about something you know well. The repo, the full example and every Instagram limit with its source: &lt;a href="https://github.com/social-media-skills/instagram-skill" rel="noopener noreferrer"&gt;github.com/social-media-skills/instagram-skill&lt;/a&gt;. Issues and pull requests are welcome.&lt;/p&gt;

&lt;p&gt;More free skills from the same team at &lt;a href="https://social-media-skills.com/?utm_source=instagram-skill&amp;amp;utm_medium=devto" rel="noopener noreferrer"&gt;social-media-skills.com&lt;/a&gt;.&lt;/p&gt;

</description>
      <category>claudecode</category>
      <category>instagram</category>
      <category>python</category>
      <category>showdev</category>
    </item>
    <item>
      <title>Remotion Skill for Claude Code: Finished Shorts From One JSON File, Checked Before Render</title>
      <dc:creator>Frank Heijdenrijk</dc:creator>
      <pubDate>Tue, 03 Nov 2026 09:22:13 +0000</pubDate>
      <link>https://dev.to/fheijdenrijk/remotion-skill-for-claude-code-finished-shorts-from-one-json-file-checked-before-render-m23</link>
      <guid>https://dev.to/fheijdenrijk/remotion-skill-for-claude-code-finished-shorts-from-one-json-file-checked-before-render-m23</guid>
      <description>&lt;p&gt;Remotion skill for Claude Code, open source and MIT licensed: Remotion Shorts Skill lets your agent turn a topic, an article, a dataset or a changelog into a finished vertical short for TikTok, Instagram Reels and YouTube Shorts. It also works in Codex, Cursor and OpenClaw.&lt;/p&gt;

&lt;p&gt;The agent writes one JSON file of scenes. A bundled Remotion project lays out every word, times every scene, renders review sheets and then the MP4, and the agent can post it after you say yes. This post covers how to use it, then the parts that are code: how text is fitted, how timing is decided, and how the skill checks its own output before you see it.&lt;br&gt;
&lt;/p&gt;

&lt;div class="highlight js-code-highlight"&gt;
&lt;pre class="highlight shell"&gt;&lt;code&gt;npx skills add social-media-skills/remotion-shorts-skill &lt;span class="nt"&gt;-g&lt;/span&gt; &lt;span class="nt"&gt;-a&lt;/span&gt; claude-code &lt;span class="nt"&gt;-y&lt;/span&gt;
&lt;/code&gt;&lt;/pre&gt;

&lt;/div&gt;



&lt;p&gt;Repo: &lt;a href="https://github.com/social-media-skills/remotion-shorts-skill" rel="noopener noreferrer"&gt;github.com/social-media-skills/remotion-shorts-skill&lt;/a&gt;&lt;/p&gt;

&lt;p&gt;&lt;a href="https://media2.dev.to/dynamic/image/width=800%2Cheight=%2Cfit=scale-down%2Cgravity=auto%2Cformat=auto/https%3A%2F%2Fdev-to-uploads.s3.us-east-2.amazonaws.com%2Fuploads%2Farticles%2Fli9wlx6di4vpq4secrzn.png" class="article-body-image-wrapper"&gt;&lt;img src="https://media2.dev.to/dynamic/image/width=800%2Cheight=%2Cfit=scale-down%2Cgravity=auto%2Cformat=auto/https%3A%2F%2Fdev-to-uploads.s3.us-east-2.amazonaws.com%2Fuploads%2Farticles%2Fli9wlx6di4vpq4secrzn.png" alt="Four frames from shorts made with the skill: 2.5% of Earth's water is fresh over a NASA image, a 96.5% stat, a numbered launch list item and a highlighted text scene" width="800" height="420"&gt;&lt;/a&gt;&lt;/p&gt;

&lt;p&gt;&lt;em&gt;Frames from the two examples in the repo. Figures in the water short come from the USGS Water Science School; the Earth image is NASA's Blue Marble 2012. Both scripts were written by an AI agent with the skill; no endorsement by USGS or NASA is implied.&lt;/em&gt;&lt;/p&gt;

&lt;h2&gt;
  
  
  What is a Remotion skill?
&lt;/h2&gt;

&lt;p&gt;A Remotion skill is an agent skill (a folder with a &lt;code&gt;SKILL.md&lt;/code&gt; and supporting files) that tells a coding agent like Claude Code how to make video with &lt;a href="https://www.remotion.dev" rel="noopener noreferrer"&gt;Remotion&lt;/a&gt;, the React framework for programmatic video. Most Remotion skills teach the agent to write Remotion components.&lt;/p&gt;

&lt;p&gt;This one works the other way round. The Remotion project is already written and tested, and the agent only writes the script. You get one job done well: kinetic-text shorts (big animated words, stat count-ups, top-N lists, quotes, launch and changelog videos) at 1080x1920.&lt;/p&gt;

&lt;h2&gt;
  
  
  How to use Remotion with Claude Code, step by step
&lt;/h2&gt;

&lt;ol&gt;
&lt;li&gt;
&lt;strong&gt;Install the skill&lt;/strong&gt; with the command above. Other agents: run &lt;code&gt;npx skills add social-media-skills/remotion-shorts-skill&lt;/code&gt; and pick yours, or &lt;code&gt;openclaw skills install @social-media-skills/remotion-shorts-skill&lt;/code&gt;.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Ask for a short.&lt;/strong&gt; "Make a short about how much of Earth's water is fresh. Use real numbers." or "Turn this changelog into a TikTok."&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;First run only:&lt;/strong&gt; the agent runs &lt;code&gt;setup.mjs&lt;/code&gt;, which copies the Remotion project, installs pinned packages and downloads the headless browser Remotion renders with (1 to 3 minutes).&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;The agent writes &lt;code&gt;short.json&lt;/code&gt;&lt;/strong&gt; and runs &lt;code&gt;node render.mjs short.json --stills&lt;/code&gt; to get review sheets in a few seconds.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;You look at the sheets&lt;/strong&gt;, ask for changes, then say "render it". You get an MP4, a cover frame and captions per app.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Posting is optional.&lt;/strong&gt; The agent shows you the accounts, captions and privacy settings and posts only after a clear yes.&lt;/li&gt;
&lt;/ol&gt;

&lt;p&gt;Here is the exchange from the water example in the README, shortened:&lt;br&gt;
&lt;/p&gt;

&lt;div class="highlight js-code-highlight"&gt;
&lt;pre class="highlight plaintext"&gt;&lt;code&gt;You:   Make a short about how much of Earth's water is fresh. Use real numbers. Ocean colours.
Agent: Using the USGS Water Science School figures. Review sheets are ready: 8 scenes, 22.3 seconds, no warnings.
You:   Looks good. Render it.
&lt;/code&gt;&lt;/pre&gt;

&lt;/div&gt;



&lt;h2&gt;
  
  
  Why a template, not generated Remotion code
&lt;/h2&gt;

&lt;p&gt;Coding agents can write Remotion components from scratch, and for a one-off motion piece that is the right approach. For shorts posted every week it causes problems. Each video is new code, so each video can have new bugs. Nothing checks that text stays out from under the app's buttons. And nobody decides how long a scene should stay up, so the timing is whatever the agent guessed.&lt;/p&gt;

&lt;p&gt;This skill splits the job. The Remotion project is fixed: six scene types (&lt;code&gt;hook&lt;/code&gt;, &lt;code&gt;text&lt;/code&gt;, &lt;code&gt;stat&lt;/code&gt;, &lt;code&gt;item&lt;/code&gt;, &lt;code&gt;quote&lt;/code&gt;, &lt;code&gt;end&lt;/code&gt;), five themes and a layout engine. The agent writes only the script:&lt;br&gt;
&lt;/p&gt;

&lt;div class="highlight js-code-highlight"&gt;
&lt;pre class="highlight json"&gt;&lt;code&gt;&lt;span class="p"&gt;{&lt;/span&gt;&lt;span class="w"&gt;
  &lt;/span&gt;&lt;span class="nl"&gt;"theme"&lt;/span&gt;&lt;span class="p"&gt;:&lt;/span&gt;&lt;span class="w"&gt; &lt;/span&gt;&lt;span class="s2"&gt;"ocean"&lt;/span&gt;&lt;span class="p"&gt;,&lt;/span&gt;&lt;span class="w"&gt;
  &lt;/span&gt;&lt;span class="nl"&gt;"footer"&lt;/span&gt;&lt;span class="p"&gt;:&lt;/span&gt;&lt;span class="w"&gt; &lt;/span&gt;&lt;span class="s2"&gt;"Data: USGS. Earth image: NASA"&lt;/span&gt;&lt;span class="p"&gt;,&lt;/span&gt;&lt;span class="w"&gt;
  &lt;/span&gt;&lt;span class="nl"&gt;"scenes"&lt;/span&gt;&lt;span class="p"&gt;:&lt;/span&gt;&lt;span class="w"&gt; &lt;/span&gt;&lt;span class="p"&gt;[&lt;/span&gt;&lt;span class="w"&gt;
    &lt;/span&gt;&lt;span class="p"&gt;{&lt;/span&gt;&lt;span class="w"&gt; &lt;/span&gt;&lt;span class="nl"&gt;"type"&lt;/span&gt;&lt;span class="p"&gt;:&lt;/span&gt;&lt;span class="w"&gt; &lt;/span&gt;&lt;span class="s2"&gt;"hook"&lt;/span&gt;&lt;span class="p"&gt;,&lt;/span&gt;&lt;span class="w"&gt; &lt;/span&gt;&lt;span class="nl"&gt;"big"&lt;/span&gt;&lt;span class="p"&gt;:&lt;/span&gt;&lt;span class="w"&gt; &lt;/span&gt;&lt;span class="s2"&gt;"2.5%"&lt;/span&gt;&lt;span class="p"&gt;,&lt;/span&gt;&lt;span class="w"&gt; &lt;/span&gt;&lt;span class="nl"&gt;"text"&lt;/span&gt;&lt;span class="p"&gt;:&lt;/span&gt;&lt;span class="w"&gt; &lt;/span&gt;&lt;span class="s2"&gt;"of Earth's water is fresh"&lt;/span&gt;&lt;span class="p"&gt;,&lt;/span&gt;&lt;span class="w"&gt; &lt;/span&gt;&lt;span class="nl"&gt;"image"&lt;/span&gt;&lt;span class="p"&gt;:&lt;/span&gt;&lt;span class="w"&gt; &lt;/span&gt;&lt;span class="s2"&gt;"earth.jpg"&lt;/span&gt;&lt;span class="w"&gt; &lt;/span&gt;&lt;span class="p"&gt;},&lt;/span&gt;&lt;span class="w"&gt;
    &lt;/span&gt;&lt;span class="p"&gt;{&lt;/span&gt;&lt;span class="w"&gt; &lt;/span&gt;&lt;span class="nl"&gt;"type"&lt;/span&gt;&lt;span class="p"&gt;:&lt;/span&gt;&lt;span class="w"&gt; &lt;/span&gt;&lt;span class="s2"&gt;"stat"&lt;/span&gt;&lt;span class="p"&gt;,&lt;/span&gt;&lt;span class="w"&gt; &lt;/span&gt;&lt;span class="nl"&gt;"value"&lt;/span&gt;&lt;span class="p"&gt;:&lt;/span&gt;&lt;span class="w"&gt; &lt;/span&gt;&lt;span class="s2"&gt;"96.5"&lt;/span&gt;&lt;span class="p"&gt;,&lt;/span&gt;&lt;span class="w"&gt; &lt;/span&gt;&lt;span class="nl"&gt;"suffix"&lt;/span&gt;&lt;span class="p"&gt;:&lt;/span&gt;&lt;span class="w"&gt; &lt;/span&gt;&lt;span class="s2"&gt;"%"&lt;/span&gt;&lt;span class="p"&gt;,&lt;/span&gt;&lt;span class="w"&gt; &lt;/span&gt;&lt;span class="nl"&gt;"label"&lt;/span&gt;&lt;span class="p"&gt;:&lt;/span&gt;&lt;span class="w"&gt; &lt;/span&gt;&lt;span class="s2"&gt;"is salt water in the oceans"&lt;/span&gt;&lt;span class="w"&gt; &lt;/span&gt;&lt;span class="p"&gt;},&lt;/span&gt;&lt;span class="w"&gt;
    &lt;/span&gt;&lt;span class="p"&gt;{&lt;/span&gt;&lt;span class="w"&gt; &lt;/span&gt;&lt;span class="nl"&gt;"type"&lt;/span&gt;&lt;span class="p"&gt;:&lt;/span&gt;&lt;span class="w"&gt; &lt;/span&gt;&lt;span class="s2"&gt;"stat"&lt;/span&gt;&lt;span class="p"&gt;,&lt;/span&gt;&lt;span class="w"&gt; &lt;/span&gt;&lt;span class="nl"&gt;"value"&lt;/span&gt;&lt;span class="p"&gt;:&lt;/span&gt;&lt;span class="w"&gt; &lt;/span&gt;&lt;span class="s2"&gt;"68.7"&lt;/span&gt;&lt;span class="p"&gt;,&lt;/span&gt;&lt;span class="w"&gt; &lt;/span&gt;&lt;span class="nl"&gt;"suffix"&lt;/span&gt;&lt;span class="p"&gt;:&lt;/span&gt;&lt;span class="w"&gt; &lt;/span&gt;&lt;span class="s2"&gt;"%"&lt;/span&gt;&lt;span class="p"&gt;,&lt;/span&gt;&lt;span class="w"&gt; &lt;/span&gt;&lt;span class="nl"&gt;"label"&lt;/span&gt;&lt;span class="p"&gt;:&lt;/span&gt;&lt;span class="w"&gt; &lt;/span&gt;&lt;span class="s2"&gt;"of the fresh water is locked in ice and glaciers"&lt;/span&gt;&lt;span class="p"&gt;,&lt;/span&gt;&lt;span class="w"&gt; &lt;/span&gt;&lt;span class="nl"&gt;"highlight"&lt;/span&gt;&lt;span class="p"&gt;:&lt;/span&gt;&lt;span class="w"&gt; &lt;/span&gt;&lt;span class="s2"&gt;"ice and glaciers"&lt;/span&gt;&lt;span class="w"&gt; &lt;/span&gt;&lt;span class="p"&gt;},&lt;/span&gt;&lt;span class="w"&gt;
    &lt;/span&gt;&lt;span class="p"&gt;{&lt;/span&gt;&lt;span class="w"&gt; &lt;/span&gt;&lt;span class="nl"&gt;"type"&lt;/span&gt;&lt;span class="p"&gt;:&lt;/span&gt;&lt;span class="w"&gt; &lt;/span&gt;&lt;span class="s2"&gt;"end"&lt;/span&gt;&lt;span class="p"&gt;,&lt;/span&gt;&lt;span class="w"&gt; &lt;/span&gt;&lt;span class="nl"&gt;"text"&lt;/span&gt;&lt;span class="p"&gt;:&lt;/span&gt;&lt;span class="w"&gt; &lt;/span&gt;&lt;span class="s2"&gt;"Save this for trivia night"&lt;/span&gt;&lt;span class="p"&gt;,&lt;/span&gt;&lt;span class="w"&gt; &lt;/span&gt;&lt;span class="nl"&gt;"highlight"&lt;/span&gt;&lt;span class="p"&gt;:&lt;/span&gt;&lt;span class="w"&gt; &lt;/span&gt;&lt;span class="s2"&gt;"Save this"&lt;/span&gt;&lt;span class="w"&gt; &lt;/span&gt;&lt;span class="p"&gt;}&lt;/span&gt;&lt;span class="w"&gt;
  &lt;/span&gt;&lt;span class="p"&gt;]&lt;/span&gt;&lt;span class="w"&gt;
&lt;/span&gt;&lt;span class="p"&gt;}&lt;/span&gt;&lt;span class="w"&gt;
&lt;/span&gt;&lt;/code&gt;&lt;/pre&gt;

&lt;/div&gt;



&lt;p&gt;The agent's effort goes into what each screen says, which is the part that decides whether people keep watching.&lt;/p&gt;

&lt;h2&gt;
  
  
  One safe zone for TikTok, Reels and Shorts
&lt;/h2&gt;

&lt;p&gt;TikTok, Reels and Shorts all draw on top of the video: search and tabs at the top, the caption, username and music line at the bottom, like and share buttons on the right. Text under any of them is lost.&lt;/p&gt;

&lt;p&gt;Only Meta publishes numbers (keep the top 14%, the bottom 35% and 6% on each side clear). TikTok ships overlay files and YouTube shows its safe area as an image. The template takes the strictest edge of the three on a 1080x1920 frame:&lt;br&gt;
&lt;/p&gt;

&lt;div class="highlight js-code-highlight"&gt;
&lt;pre class="highlight typescript"&gt;&lt;code&gt;&lt;span class="c1"&gt;// src/theme.ts&lt;/span&gt;
&lt;span class="k"&gt;export&lt;/span&gt; &lt;span class="kd"&gt;const&lt;/span&gt; &lt;span class="nx"&gt;SAFE&lt;/span&gt; &lt;span class="o"&gt;=&lt;/span&gt; &lt;span class="p"&gt;{&lt;/span&gt; &lt;span class="na"&gt;top&lt;/span&gt;&lt;span class="p"&gt;:&lt;/span&gt; &lt;span class="mi"&gt;288&lt;/span&gt;&lt;span class="p"&gt;,&lt;/span&gt; &lt;span class="na"&gt;bottom&lt;/span&gt;&lt;span class="p"&gt;:&lt;/span&gt; &lt;span class="mi"&gt;672&lt;/span&gt;&lt;span class="p"&gt;,&lt;/span&gt; &lt;span class="na"&gt;left&lt;/span&gt;&lt;span class="p"&gt;:&lt;/span&gt; &lt;span class="mi"&gt;96&lt;/span&gt;&lt;span class="p"&gt;,&lt;/span&gt; &lt;span class="na"&gt;right&lt;/span&gt;&lt;span class="p"&gt;:&lt;/span&gt; &lt;span class="mi"&gt;192&lt;/span&gt; &lt;span class="p"&gt;};&lt;/span&gt;
&lt;/code&gt;&lt;/pre&gt;

&lt;/div&gt;



&lt;p&gt;That leaves 792x960 pixels for text. Centred text uses 192 px on both sides so it sits on the frame's centre line. Photos and backgrounds still fill the frame; only text stays inside.&lt;/p&gt;

&lt;h2&gt;
  
  
  Fitting text: the biggest size with balanced lines
&lt;/h2&gt;

&lt;p&gt;&lt;code&gt;calculateMetadata&lt;/code&gt; runs in the headless browser before any frame renders, so it can measure text with the real fonts. &lt;code&gt;measureText&lt;/code&gt; from &lt;code&gt;@remotion/layout-utils&lt;/code&gt; measures every word once at 100 px; widths then scale linearly, and the layout searches sizes from the scene's maximum down to a readable floor.&lt;/p&gt;

&lt;p&gt;The biggest size that fits is not always the best. A seven-word hook at maximum size can come out as four lines with one word stranded on the last. So each candidate (a size plus a line count) gets a cost:&lt;br&gt;
&lt;/p&gt;

&lt;div class="highlight js-code-highlight"&gt;
&lt;pre class="highlight typescript"&gt;&lt;code&gt;&lt;span class="c1"&gt;// src/layout.ts&lt;/span&gt;
&lt;span class="kd"&gt;const&lt;/span&gt; &lt;span class="nx"&gt;lonely&lt;/span&gt; &lt;span class="o"&gt;=&lt;/span&gt; &lt;span class="nx"&gt;words&lt;/span&gt;&lt;span class="p"&gt;.&lt;/span&gt;&lt;span class="nx"&gt;length&lt;/span&gt; &lt;span class="o"&gt;&amp;gt;&lt;/span&gt; &lt;span class="mi"&gt;2&lt;/span&gt; &lt;span class="p"&gt;?&lt;/span&gt; &lt;span class="nx"&gt;lines&lt;/span&gt;&lt;span class="p"&gt;.&lt;/span&gt;&lt;span class="nf"&gt;filter&lt;/span&gt;&lt;span class="p"&gt;((&lt;/span&gt;&lt;span class="nx"&gt;l&lt;/span&gt;&lt;span class="p"&gt;)&lt;/span&gt; &lt;span class="o"&gt;=&amp;gt;&lt;/span&gt; &lt;span class="nx"&gt;l&lt;/span&gt;&lt;span class="p"&gt;.&lt;/span&gt;&lt;span class="nx"&gt;length&lt;/span&gt; &lt;span class="o"&gt;===&lt;/span&gt; &lt;span class="mi"&gt;1&lt;/span&gt;&lt;span class="p"&gt;).&lt;/span&gt;&lt;span class="nx"&gt;length&lt;/span&gt; &lt;span class="o"&gt;*&lt;/span&gt; &lt;span class="mf"&gt;0.2&lt;/span&gt; &lt;span class="p"&gt;:&lt;/span&gt; &lt;span class="mi"&gt;0&lt;/span&gt;&lt;span class="p"&gt;;&lt;/span&gt;
&lt;span class="kd"&gt;const&lt;/span&gt; &lt;span class="nx"&gt;widow&lt;/span&gt; &lt;span class="o"&gt;=&lt;/span&gt; &lt;span class="nx"&gt;lines&lt;/span&gt;&lt;span class="p"&gt;.&lt;/span&gt;&lt;span class="nx"&gt;length&lt;/span&gt; &lt;span class="o"&gt;&amp;gt;&lt;/span&gt; &lt;span class="mi"&gt;1&lt;/span&gt; &lt;span class="o"&gt;&amp;amp;&amp;amp;&lt;/span&gt; &lt;span class="nx"&gt;lines&lt;/span&gt;&lt;span class="p"&gt;[&lt;/span&gt;&lt;span class="nx"&gt;lines&lt;/span&gt;&lt;span class="p"&gt;.&lt;/span&gt;&lt;span class="nx"&gt;length&lt;/span&gt; &lt;span class="o"&gt;-&lt;/span&gt; &lt;span class="mi"&gt;1&lt;/span&gt;&lt;span class="p"&gt;].&lt;/span&gt;&lt;span class="nx"&gt;length&lt;/span&gt; &lt;span class="o"&gt;===&lt;/span&gt; &lt;span class="mi"&gt;1&lt;/span&gt; &lt;span class="o"&gt;&amp;amp;&amp;amp;&lt;/span&gt; &lt;span class="nx"&gt;words&lt;/span&gt;&lt;span class="p"&gt;.&lt;/span&gt;&lt;span class="nx"&gt;length&lt;/span&gt; &lt;span class="o"&gt;&amp;gt;&lt;/span&gt; &lt;span class="mi"&gt;2&lt;/span&gt; &lt;span class="p"&gt;?&lt;/span&gt; &lt;span class="mf"&gt;0.1&lt;/span&gt; &lt;span class="p"&gt;:&lt;/span&gt; &lt;span class="mi"&gt;0&lt;/span&gt;&lt;span class="p"&gt;;&lt;/span&gt;
&lt;span class="kd"&gt;const&lt;/span&gt; &lt;span class="nx"&gt;cost&lt;/span&gt; &lt;span class="o"&gt;=&lt;/span&gt; &lt;span class="p"&gt;(&lt;/span&gt;&lt;span class="nx"&gt;max&lt;/span&gt; &lt;span class="o"&gt;-&lt;/span&gt; &lt;span class="nx"&gt;size&lt;/span&gt;&lt;span class="p"&gt;)&lt;/span&gt; &lt;span class="o"&gt;/&lt;/span&gt; &lt;span class="nx"&gt;max&lt;/span&gt; &lt;span class="o"&gt;+&lt;/span&gt; &lt;span class="mf"&gt;0.3&lt;/span&gt; &lt;span class="o"&gt;*&lt;/span&gt; &lt;span class="nb"&gt;Math&lt;/span&gt;&lt;span class="p"&gt;.&lt;/span&gt;&lt;span class="nf"&gt;max&lt;/span&gt;&lt;span class="p"&gt;(&lt;/span&gt;&lt;span class="mi"&gt;0&lt;/span&gt;&lt;span class="p"&gt;,&lt;/span&gt; &lt;span class="nx"&gt;lines&lt;/span&gt;&lt;span class="p"&gt;.&lt;/span&gt;&lt;span class="nx"&gt;length&lt;/span&gt; &lt;span class="o"&gt;-&lt;/span&gt; &lt;span class="mi"&gt;3&lt;/span&gt;&lt;span class="p"&gt;)&lt;/span&gt; &lt;span class="o"&gt;+&lt;/span&gt; &lt;span class="nx"&gt;lonely&lt;/span&gt; &lt;span class="o"&gt;+&lt;/span&gt; &lt;span class="nx"&gt;widow&lt;/span&gt; &lt;span class="o"&gt;+&lt;/span&gt; &lt;span class="mf"&gt;0.05&lt;/span&gt; &lt;span class="o"&gt;*&lt;/span&gt; &lt;span class="nx"&gt;ragged&lt;/span&gt;&lt;span class="p"&gt;;&lt;/span&gt;
&lt;/code&gt;&lt;/pre&gt;

&lt;/div&gt;



&lt;p&gt;Bigger is better, more than 3 lines costs extra, a one-word line costs extra (more so on the last line), and &lt;code&gt;ragged&lt;/code&gt; penalises lines of very different length. For each line count it binary-searches the narrowest measure that still gives that many lines, which balances them. Three details keep text inside the box:&lt;/p&gt;

&lt;ul&gt;
&lt;li&gt;
&lt;strong&gt;Highlight boxes have width.&lt;/strong&gt; A highlighted run gets a coloured block that overhangs the letters, so its edge words are measured 0.12 em wider on each side.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;The punch has width.&lt;/strong&gt; When the key word lands, the block springs to 1.04x around the centre of the safe box. Layout fits to the safe width divided by 1.04, so the punch never pushes a letter out.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Numbers are measured as drawn.&lt;/strong&gt; A stat like &lt;code&gt;96.5%&lt;/code&gt; sets the suffix at 0.62 em and the digits with &lt;code&gt;tabular-nums&lt;/code&gt;, so the measurement uses the same parts.&lt;/li&gt;
&lt;/ul&gt;

&lt;p&gt;A single word too wide for the screen at the smallest size is an error, not a warning. &lt;code&gt;--force&lt;/code&gt; can override warnings but never puts text off screen.&lt;/p&gt;

&lt;h2&gt;
  
  
  Timing: how long each scene stays on screen
&lt;/h2&gt;

&lt;p&gt;Each scene stays up long enough to read, and no longer:&lt;br&gt;
&lt;/p&gt;

&lt;div class="highlight js-code-highlight"&gt;
&lt;pre class="highlight typescript"&gt;&lt;code&gt;&lt;span class="kd"&gt;const&lt;/span&gt; &lt;span class="nx"&gt;CPS&lt;/span&gt; &lt;span class="o"&gt;=&lt;/span&gt; &lt;span class="mi"&gt;17&lt;/span&gt;&lt;span class="p"&gt;;&lt;/span&gt;        &lt;span class="c1"&gt;// characters per second&lt;/span&gt;
&lt;span class="kd"&gt;const&lt;/span&gt; &lt;span class="nx"&gt;PER_WORD&lt;/span&gt; &lt;span class="o"&gt;=&lt;/span&gt; &lt;span class="mf"&gt;0.3&lt;/span&gt;&lt;span class="p"&gt;;&lt;/span&gt;  &lt;span class="c1"&gt;// seconds per word&lt;/span&gt;
&lt;span class="kd"&gt;const&lt;/span&gt; &lt;span class="nx"&gt;read&lt;/span&gt; &lt;span class="o"&gt;=&lt;/span&gt; &lt;span class="p"&gt;(&lt;/span&gt;&lt;span class="nx"&gt;s&lt;/span&gt;&lt;span class="p"&gt;)&lt;/span&gt; &lt;span class="o"&gt;=&amp;gt;&lt;/span&gt; &lt;span class="nb"&gt;Math&lt;/span&gt;&lt;span class="p"&gt;.&lt;/span&gt;&lt;span class="nf"&gt;max&lt;/span&gt;&lt;span class="p"&gt;(&lt;/span&gt;&lt;span class="nx"&gt;s&lt;/span&gt;&lt;span class="p"&gt;.&lt;/span&gt;&lt;span class="nx"&gt;length&lt;/span&gt; &lt;span class="o"&gt;/&lt;/span&gt; &lt;span class="nx"&gt;CPS&lt;/span&gt;&lt;span class="p"&gt;,&lt;/span&gt; &lt;span class="nf"&gt;countWords&lt;/span&gt;&lt;span class="p"&gt;(&lt;/span&gt;&lt;span class="nx"&gt;s&lt;/span&gt;&lt;span class="p"&gt;)&lt;/span&gt; &lt;span class="o"&gt;*&lt;/span&gt; &lt;span class="nx"&gt;PER_WORD&lt;/span&gt;&lt;span class="p"&gt;);&lt;/span&gt;
&lt;span class="kd"&gt;const&lt;/span&gt; &lt;span class="nx"&gt;needed&lt;/span&gt; &lt;span class="o"&gt;=&lt;/span&gt; &lt;span class="nx"&gt;ENTRY&lt;/span&gt; &lt;span class="o"&gt;+&lt;/span&gt; &lt;span class="mf"&gt;0.4&lt;/span&gt; &lt;span class="o"&gt;*&lt;/span&gt; &lt;span class="nx"&gt;reveal&lt;/span&gt; &lt;span class="o"&gt;+&lt;/span&gt; &lt;span class="nx"&gt;hold&lt;/span&gt;&lt;span class="p"&gt;;&lt;/span&gt;
&lt;/code&gt;&lt;/pre&gt;

&lt;/div&gt;



&lt;p&gt;The numbers come from subtitle practice: BBC guidance of about 0.33 seconds per word and Netflix's 17 to 20 characters per second for adults, slightly faster here because the text is the whole picture. People read while the words rise into place, so only 40% of the reveal time is added on top. Stats get time for the count; the small line under a list item counts at 60%, because it is skimmed.&lt;/p&gt;

&lt;p&gt;In practice a hook lands in 2 to 3 seconds, a stat in 2.5 to 4 and a list item in 2 to 3. The water example is 8 scenes in 22 seconds. With a &lt;code&gt;bpm&lt;/code&gt;, every cut moves to the next beat, never an earlier one, so snapping cannot cut a scene short of its reading time.&lt;/p&gt;

&lt;h2&gt;
  
  
  Frame 0 is the thumbnail
&lt;/h2&gt;

&lt;p&gt;TikTok and Reels show frame 0 in the profile grid, and it is the first thing a scroller sees. So the first scene is fully set on frame 0, and only the highlight wipe and the punch move after it. Every later scene starts already moving, so no cut lands on an empty frame.&lt;/p&gt;

&lt;p&gt;Words rise from behind a mask instead of fading, so no frame shows grey, half-transparent text. The highlighted run is drawn twice: the normal colour underneath and the accent-text colour on top, clipped by the same edge as the wiping block, so the text is readable at every frame of the wipe.&lt;/p&gt;

&lt;h2&gt;
  
  
  Review sheets before the render
&lt;/h2&gt;

&lt;p&gt;&lt;code&gt;node render.mjs short.json --stills&lt;/code&gt; runs in a few seconds and writes:&lt;/p&gt;

&lt;ul&gt;
&lt;li&gt;
&lt;code&gt;sheet.jpg&lt;/code&gt;: the last frame of every scene, with the safe area marked by a thin dashed line;&lt;/li&gt;
&lt;li&gt;
&lt;code&gt;cuts.jpg&lt;/code&gt;: frame 0 and the frames around every cut;&lt;/li&gt;
&lt;li&gt;
&lt;code&gt;cover.jpg&lt;/code&gt;: the hook once it has landed;&lt;/li&gt;
&lt;li&gt;
&lt;code&gt;report.json&lt;/code&gt;: the plan, line breaks and warnings.&lt;/li&gt;
&lt;/ul&gt;

&lt;p&gt;The agent opens both sheets before rendering the video. Warnings stop the render with exit code 2: text too long for one screen, a hook over about 10 words, a highlight that is not in the text, a scene too fast to read, a short over 59 seconds (YouTube blocks Shorts over a minute worldwide if any copyright claim lands on them).&lt;/p&gt;

&lt;p&gt;The sheets are Remotion compositions too: a grid of &lt;code&gt;&amp;lt;Short&amp;gt;&lt;/code&gt; instances, each inside &lt;code&gt;&amp;lt;Freeze frame={n}&amp;gt;&lt;/code&gt;. One gotcha if you build something similar: the sheet composition needs the same &lt;code&gt;durationInFrames&lt;/code&gt; as the short, or frames past its length are clamped and every tile shows frame 0.&lt;/p&gt;

&lt;h2&gt;
  
  
  Sound without licensing problems
&lt;/h2&gt;

&lt;p&gt;Silent text videos tend to look like low-effort templates, but music from a free library is a licensing risk on at least one of the three apps. The template ships its own sound effects, generated with ffmpeg's signal generators (no samples, no recordings): a low hit on the hook, a whoosh on each cut, a pop when a highlight lands and ticks under a count-up. They are MIT like the rest of the repo, on by default, and placed from the same timing the motion uses.&lt;/p&gt;

&lt;h2&gt;
  
  
  Rendering on ordinary laptops
&lt;/h2&gt;

&lt;p&gt;The output is 1080x1920, 30 fps, H.264, yuv420p, BT.709, AAC with the moov atom at the front: a format all three apps accept. Machines with 8 GB of memory render 2 frames at a time; &lt;code&gt;--concurrency 1&lt;/code&gt; is there if a render is still killed. A 20-second short renders in under a minute on a recent laptop.&lt;/p&gt;

&lt;p&gt;Setup is one command: &lt;code&gt;node scripts/setup.mjs&lt;/code&gt; copies the project, installs pinned packages (Remotion 4.0.536) and downloads Chrome Headless Shell. Run it again after the skill updates and it refreshes the template, keeping a &lt;code&gt;.bak&lt;/code&gt; of anything you changed. CI sets up a fresh project and renders both examples on macOS, Windows and Ubuntu.&lt;/p&gt;

&lt;h2&gt;
  
  
  Posting to TikTok, Reels and Shorts
&lt;/h2&gt;

&lt;p&gt;Posting is optional and goes through &lt;a href="https://woopsocial.com/?utm_source=remotion-shorts-skill&amp;amp;utm_medium=devto" rel="noopener noreferrer"&gt;WoopSocial&lt;/a&gt;, a social media API and MCP server for agents. Disclosure: the skill is maintained by Frank Heijdenrijk and the team behind WoopSocial. Setup for every agent is at &lt;a href="https://woopsocial.com/agents?utm_source=remotion-shorts-skill&amp;amp;utm_medium=devto" rel="noopener noreferrer"&gt;woopsocial.com/agents&lt;/a&gt;.&lt;/p&gt;

&lt;p&gt;&lt;code&gt;upload.mjs&lt;/code&gt; sends the MP4 in parts with retries; the agent reads the TikTok account's allowed privacy settings, writes a caption per app, validates the post and posts only after a clear yes. Posting to several accounts, it makes a different version for each one, because Instagram stops recommending near-duplicate posts and YouTube treats mass-produced templated videos as inauthentic content. If you would rather not connect anything, you get the MP4 and the captions to post yourself.&lt;/p&gt;

&lt;h2&gt;
  
  
  FAQ
&lt;/h2&gt;

&lt;h3&gt;
  
  
  Do I need to know Remotion to use this skill?
&lt;/h3&gt;

&lt;p&gt;No. The templates are done and the agent writes a JSON file. If you want to change the design, the project is a normal Remotion project: run &lt;code&gt;npm run studio&lt;/code&gt; and edit &lt;code&gt;src/Short.tsx&lt;/code&gt;.&lt;/p&gt;

&lt;h3&gt;
  
  
  Can Claude Code make videos with Remotion?
&lt;/h3&gt;

&lt;p&gt;Yes. Claude Code can run commands on your machine, so with this skill it writes the script, runs the checks and renders the MP4 locally with Node. Rendering needs no paid API.&lt;/p&gt;

&lt;h3&gt;
  
  
  Does the skill edit existing video?
&lt;/h3&gt;

&lt;p&gt;No. It makes new text-led shorts. You can put a photo or video behind any scene or the whole short; it is dimmed under the theme colour so the text stays readable.&lt;/p&gt;

&lt;h3&gt;
  
  
  Does it work in Codex, Cursor and OpenClaw?
&lt;/h3&gt;

&lt;p&gt;Yes, any agent that can run commands on your computer. It needs Node 20+ and macOS 15+, Windows x64 or Linux with glibc 2.35+ (Remotion's own requirements).&lt;/p&gt;

&lt;h3&gt;
  
  
  Is Remotion free for commercial use?
&lt;/h3&gt;

&lt;p&gt;The skill is MIT. Remotion has its own licence: free for individuals, non-profits and companies of up to 3 people; larger companies need a Remotion company licence, and rendering from scripts counts as automation. The skill says so before the first render. Current terms: &lt;a href="https://www.remotion.pro/license" rel="noopener noreferrer"&gt;remotion.pro/license&lt;/a&gt;.&lt;/p&gt;

&lt;h2&gt;
  
  
  Try the Remotion Shorts Skill
&lt;/h2&gt;



&lt;div class="highlight js-code-highlight"&gt;
&lt;pre class="highlight shell"&gt;&lt;code&gt;npx skills add social-media-skills/remotion-shorts-skill &lt;span class="nt"&gt;-g&lt;/span&gt; &lt;span class="nt"&gt;-a&lt;/span&gt; claude-code &lt;span class="nt"&gt;-y&lt;/span&gt;
&lt;/code&gt;&lt;/pre&gt;

&lt;/div&gt;



&lt;p&gt;Then ask your agent for a short about something you know well. Repo, both examples with their review sheets, and every rule with its source: &lt;a href="https://github.com/social-media-skills/remotion-shorts-skill" rel="noopener noreferrer"&gt;github.com/social-media-skills/remotion-shorts-skill&lt;/a&gt;. Issues and pull requests are welcome.&lt;/p&gt;

&lt;p&gt;More free skills from the same team at &lt;a href="https://social-media-skills.com/?utm_source=remotion-shorts-skill&amp;amp;utm_medium=devto" rel="noopener noreferrer"&gt;social-media-skills.com&lt;/a&gt;.&lt;/p&gt;

</description>
      <category>remotion</category>
      <category>claudecode</category>
      <category>showdev</category>
      <category>video</category>
    </item>
    <item>
      <title>YouTube Skill for Claude Code: How It Picks the Search Phrase, Counts YouTube's Limits and Scores Thumbnail Frames</title>
      <dc:creator>Frank Heijdenrijk</dc:creator>
      <pubDate>Mon, 02 Nov 2026 09:38:04 +0000</pubDate>
      <link>https://dev.to/fheijdenrijk/youtube-skill-for-claude-code-how-it-picks-the-search-phrase-counts-youtubes-limits-and-scores-ac0</link>
      <guid>https://dev.to/fheijdenrijk/youtube-skill-for-claude-code-how-it-picks-the-search-phrase-counts-youtubes-limits-and-scores-ac0</guid>
      <description>&lt;p&gt;YouTube skill for Claude Code, Codex, Cursor and OpenClaw: a free, MIT-licensed agent skill that takes a finished video and builds everything the upload needs to be found. It transcribes the video, reads what people type into YouTube search, writes 10 titles, a description, chapters, tags and hashtags, checks every field against YouTube's limits, renders thumbnail options from the video's own frames, writes a caption file, and uploads only after you say yes. This post skips the pitch and walks through the parts that are code: what the scripts count, how they count it, and why the defaults are what they are.&lt;/p&gt;

&lt;p&gt;Repo: &lt;a href="https://github.com/social-media-skills/youtube-skill" rel="noopener noreferrer"&gt;github.com/social-media-skills/youtube-skill&lt;/a&gt;&lt;/p&gt;

&lt;p&gt;&lt;a href="https://media2.dev.to/dynamic/image/width=800%2Cheight=%2Cfit=scale-down%2Cgravity=auto%2Cformat=auto/https%3A%2F%2Fdev-to-uploads.s3.us-east-2.amazonaws.com%2Fuploads%2Farticles%2Fp3vpobl862e0ukcyzvo2.png" class="article-body-image-wrapper"&gt;&lt;img src="https://media2.dev.to/dynamic/image/width=800%2Cheight=%2Cfit=scale-down%2Cgravity=auto%2Cformat=auto/https%3A%2F%2Fdev-to-uploads.s3.us-east-2.amazonaws.com%2Fuploads%2Farticles%2Fp3vpobl862e0ukcyzvo2.png" alt="A raw upload with a filename title next to the package the skill built: search-first title, thumbnail from the video, description opening and 7 chapters" width="800" height="502"&gt;&lt;/a&gt;&lt;/p&gt;

&lt;p&gt;&lt;em&gt;The repo's worked example. The video is "Earth Water Budget Animation", courtesy NASA/JPL-Caltech. The title, description, chapters and thumbnails were written by an AI agent with the skill; no endorsement by NASA, JPL or Caltech is implied.&lt;/em&gt;&lt;/p&gt;

&lt;h2&gt;
  
  
  Six scripts that count, one agent that chooses
&lt;/h2&gt;

&lt;p&gt;The skill is a &lt;code&gt;SKILL.md&lt;/code&gt; workflow plus six small Python scripts. The scripts handle everything with a right answer: byte counts, timestamps, ffmpeg calls, pixel work. The agent handles everything that needs reading: which search phrase the video actually answers, which title is honest, which frame shows what the video promises. Neither does the other's job.&lt;br&gt;
&lt;/p&gt;

&lt;div class="highlight js-code-highlight"&gt;
&lt;pre class="highlight plaintext"&gt;&lt;code&gt;final.mp4
  -&amp;gt; check_setup.py           Python, ffmpeg, Pillow, transcription routes, OpenCV (optional)
  -&amp;gt; transcribe.py            work/transcript.json + transcript.txt
       --from captions.srt | GROQ_API_KEY | OPENAI_API_KEY | local faster-whisper
  -&amp;gt; suggest.py               pack/suggestions.md: what people type into YouTube search
  -&amp;gt; agent reads the whole transcript, picks one phrase, writes upload.json
  -&amp;gt; youtube_pack.py build    checks every field, snaps chapters, writes description.txt,
                              captions.srt, post.json, manifest.json, upload-pack.md
  -&amp;gt; youtube_pack.py frames   frames.jpg: 12 scored candidate frames
  -&amp;gt; agent writes thumbs.json (2 or 3 different ideas)
  -&amp;gt; thumbnail.py             thumb-a.jpg, thumb-b.jpg ... and preview.png at feed size
  -&amp;gt; you approve
  -&amp;gt; WoopSocial               upload session, posts_validate, posts_create
&lt;/code&gt;&lt;/pre&gt;

&lt;/div&gt;



&lt;p&gt;Apart from Pillow, the scripts use only the standard library and ffmpeg. No paid API is required anywhere in the chain.&lt;/p&gt;

&lt;h2&gt;
  
  
  Finding the phrase: autocomplete, used lightly
&lt;/h2&gt;

&lt;p&gt;&lt;code&gt;suggest.py&lt;/code&gt; asks YouTube's search suggestions about the topic and its question forms (&lt;code&gt;how to&lt;/code&gt;, &lt;code&gt;how does&lt;/code&gt;, &lt;code&gt;why&lt;/code&gt;, &lt;code&gt;what is&lt;/code&gt;, &lt;code&gt;is&lt;/code&gt;, &lt;code&gt;best&lt;/code&gt;, &lt;code&gt;vs&lt;/code&gt;). With &lt;code&gt;--deep&lt;/code&gt; it also tries the topic followed by every letter. It writes the suggestions in YouTube's order and says plainly that the order is not a search volume.&lt;/p&gt;

&lt;p&gt;That endpoint has no official API and no published terms, so the script treats it with care: a handful of requests per video, results saved to &lt;code&gt;suggestions.json&lt;/code&gt;, and a docstring telling the agent never to loop it over hundreds of topics. One portability detail: Python from python.org on a Mac often ships without a certificate store, so &lt;code&gt;fetch()&lt;/code&gt; tries &lt;code&gt;urllib&lt;/code&gt; first (with &lt;code&gt;certifi&lt;/code&gt; when installed) and falls back to &lt;code&gt;curl&lt;/code&gt;, which every current Mac, Windows and Linux machine has.&lt;/p&gt;

&lt;p&gt;The example shows why a phrase beats a topic. The broad "earth water" returns noise, such as the chemical elements and water parks. "how much water is on earth" is YouTube's top suggestion for the topic, and the video answers exactly that question, so it became the main phrase. Other suggestions such as "how much drinkable water is there on earth" went into the description body and the tags, where they're true.&lt;/p&gt;

&lt;p&gt;With your own free &lt;code&gt;YOUTUBE_API_KEY&lt;/code&gt;, &lt;code&gt;--top&lt;/code&gt; also lists the 10 videos ranking for the phrase with views per day. As of October 2026, &lt;code&gt;search.list&lt;/code&gt; has its own quota bucket of 100 calls a day at 1 unit each, so one run costs 1 of your 100 daily searches.&lt;/p&gt;

&lt;h2&gt;
  
  
  Counting the way YouTube counts
&lt;/h2&gt;

&lt;p&gt;The YouTube Data API and the Help Center don't always agree. The Help Center describes a 5,000-character description; the API docs say 5,000 bytes. &lt;code&gt;youtube_pack.py&lt;/code&gt; uses the stricter one, which matters as soon as a description contains emoji or accented letters.&lt;/p&gt;

&lt;div class="table-wrapper-paragraph"&gt;&lt;table&gt;
&lt;thead&gt;
&lt;tr&gt;
&lt;th&gt;Field&lt;/th&gt;
&lt;th&gt;What the script measures&lt;/th&gt;
&lt;th&gt;Warns when&lt;/th&gt;
&lt;/tr&gt;
&lt;/thead&gt;
&lt;tbody&gt;
&lt;tr&gt;
&lt;td&gt;Title&lt;/td&gt;
&lt;td&gt;characters&lt;/td&gt;
&lt;td&gt;over 100 (hard), over 70 (usually cut), phrase missing, mostly capitals, &lt;code&gt;&amp;lt;&lt;/code&gt; or &lt;code&gt;&amp;gt;&lt;/code&gt;
&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;Description&lt;/td&gt;
&lt;td&gt;UTF-8 bytes&lt;/td&gt;
&lt;td&gt;over 5,000 bytes, phrase not in the first 150 characters, &lt;code&gt;&amp;lt;&lt;/code&gt; or &lt;code&gt;&amp;gt;&lt;/code&gt;
&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;Tags&lt;/td&gt;
&lt;td&gt;characters as the API counts them&lt;/td&gt;
&lt;td&gt;over 500, or a tag contains &lt;code&gt;&amp;lt;&lt;/code&gt;, &lt;code&gt;&amp;gt;&lt;/code&gt; or a comma&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;Hashtags&lt;/td&gt;
&lt;td&gt;count and characters&lt;/td&gt;
&lt;td&gt;a space or symbol breaks one, or more than 60 (all are then ignored)&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;Chapters&lt;/td&gt;
&lt;td&gt;seconds&lt;/td&gt;
&lt;td&gt;not at 0:00, fewer than 3, any under 10 s, any after the end&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;Made for kids&lt;/td&gt;
&lt;td&gt;presence&lt;/td&gt;
&lt;td&gt;missing; it's a legal setting the user must declare&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;Schedule&lt;/td&gt;
&lt;td&gt;ISO time&lt;/td&gt;
&lt;td&gt;no timezone, or in the past&lt;/td&gt;
&lt;/tr&gt;
&lt;/tbody&gt;
&lt;/table&gt;&lt;/div&gt;

&lt;p&gt;The tag budget is the least obvious one. The API counts the whole list against 500 characters, commas included, and a tag with a space is wrapped in quotes that count too:&lt;br&gt;
&lt;/p&gt;

&lt;div class="highlight js-code-highlight"&gt;
&lt;pre class="highlight python"&gt;&lt;code&gt;&lt;span class="k"&gt;def&lt;/span&gt; &lt;span class="nf"&gt;tags_length&lt;/span&gt;&lt;span class="p"&gt;(&lt;/span&gt;&lt;span class="n"&gt;tags&lt;/span&gt;&lt;span class="p"&gt;):&lt;/span&gt;
    &lt;span class="k"&gt;return&lt;/span&gt; &lt;span class="nf"&gt;sum&lt;/span&gt;&lt;span class="p"&gt;(&lt;/span&gt;&lt;span class="nf"&gt;len&lt;/span&gt;&lt;span class="p"&gt;(&lt;/span&gt;&lt;span class="n"&gt;t&lt;/span&gt;&lt;span class="p"&gt;)&lt;/span&gt; &lt;span class="o"&gt;+&lt;/span&gt; &lt;span class="p"&gt;(&lt;/span&gt;&lt;span class="mi"&gt;2&lt;/span&gt; &lt;span class="k"&gt;if&lt;/span&gt; &lt;span class="sh"&gt;"&lt;/span&gt;&lt;span class="s"&gt; &lt;/span&gt;&lt;span class="sh"&gt;"&lt;/span&gt; &lt;span class="ow"&gt;in&lt;/span&gt; &lt;span class="n"&gt;t&lt;/span&gt; &lt;span class="k"&gt;else&lt;/span&gt; &lt;span class="mi"&gt;0&lt;/span&gt;&lt;span class="p"&gt;)&lt;/span&gt; &lt;span class="k"&gt;for&lt;/span&gt; &lt;span class="n"&gt;t&lt;/span&gt; &lt;span class="ow"&gt;in&lt;/span&gt; &lt;span class="n"&gt;tags&lt;/span&gt;&lt;span class="p"&gt;)&lt;/span&gt; &lt;span class="o"&gt;+&lt;/span&gt; &lt;span class="nf"&gt;max&lt;/span&gt;&lt;span class="p"&gt;(&lt;/span&gt;&lt;span class="mi"&gt;0&lt;/span&gt;&lt;span class="p"&gt;,&lt;/span&gt; &lt;span class="nf"&gt;len&lt;/span&gt;&lt;span class="p"&gt;(&lt;/span&gt;&lt;span class="n"&gt;tags&lt;/span&gt;&lt;span class="p"&gt;)&lt;/span&gt; &lt;span class="o"&gt;-&lt;/span&gt; &lt;span class="mi"&gt;1&lt;/span&gt;&lt;span class="p"&gt;)&lt;/span&gt;
&lt;/code&gt;&lt;/pre&gt;

&lt;/div&gt;



&lt;p&gt;The example's 8 tags come to 187 of 500, and its description to 739 bytes. Those numbers land in &lt;code&gt;manifest.json&lt;/code&gt; so the agent can report them without guessing.&lt;/p&gt;

&lt;p&gt;The phrase check is deliberately loose about grammar and strict about words. &lt;code&gt;phrase_in()&lt;/code&gt; drops function words ("is", "on", "the"), then looks for the start of each remaining word, so "Earth's" satisfies "earth" and "waters" satisfies "water". The title check also renders where a phone line usually ends, about 40 characters, which &lt;code&gt;upload-pack.md&lt;/code&gt; shows as "How Much Water Is on Earth? Only 0.01% I...". The 40 and 70 are rules of thumb; YouTube publishes no number, and the bundled &lt;code&gt;references/youtube-seo.md&lt;/code&gt; says so next to every limit it does publish, with sources.&lt;/p&gt;

&lt;h2&gt;
  
  
  Chapters snapped to the sentence they fall in
&lt;/h2&gt;

&lt;p&gt;A chapter start read off a timestamp can land a moment after the sentence begins, and then the chapter opens mid-word. The pack moves each start to the nearest sentence start within 3 seconds, minus a 0.3-second lead-in:&lt;br&gt;
&lt;/p&gt;

&lt;div class="highlight js-code-highlight"&gt;
&lt;pre class="highlight python"&gt;&lt;code&gt;&lt;span class="k"&gt;for&lt;/span&gt; &lt;span class="n"&gt;t&lt;/span&gt;&lt;span class="p"&gt;,&lt;/span&gt; &lt;span class="n"&gt;name&lt;/span&gt; &lt;span class="ow"&gt;in&lt;/span&gt; &lt;span class="n"&gt;chapters&lt;/span&gt;&lt;span class="p"&gt;:&lt;/span&gt;
    &lt;span class="n"&gt;s&lt;/span&gt; &lt;span class="o"&gt;=&lt;/span&gt; &lt;span class="nf"&gt;secs&lt;/span&gt;&lt;span class="p"&gt;(&lt;/span&gt;&lt;span class="n"&gt;t&lt;/span&gt;&lt;span class="p"&gt;)&lt;/span&gt;
    &lt;span class="k"&gt;if&lt;/span&gt; &lt;span class="n"&gt;sentences&lt;/span&gt; &lt;span class="ow"&gt;and&lt;/span&gt; &lt;span class="n"&gt;s&lt;/span&gt; &lt;span class="o"&gt;&amp;gt;&lt;/span&gt; &lt;span class="mi"&gt;0&lt;/span&gt;&lt;span class="p"&gt;:&lt;/span&gt;
        &lt;span class="n"&gt;near&lt;/span&gt; &lt;span class="o"&gt;=&lt;/span&gt; &lt;span class="nf"&gt;min&lt;/span&gt;&lt;span class="p"&gt;(&lt;/span&gt;&lt;span class="n"&gt;sentences&lt;/span&gt;&lt;span class="p"&gt;,&lt;/span&gt; &lt;span class="n"&gt;key&lt;/span&gt;&lt;span class="o"&gt;=&lt;/span&gt;&lt;span class="k"&gt;lambda&lt;/span&gt; &lt;span class="n"&gt;x&lt;/span&gt;&lt;span class="p"&gt;:&lt;/span&gt; &lt;span class="nf"&gt;abs&lt;/span&gt;&lt;span class="p"&gt;(&lt;/span&gt;&lt;span class="n"&gt;x&lt;/span&gt;&lt;span class="p"&gt;[&lt;/span&gt;&lt;span class="sh"&gt;"&lt;/span&gt;&lt;span class="s"&gt;start&lt;/span&gt;&lt;span class="sh"&gt;"&lt;/span&gt;&lt;span class="p"&gt;]&lt;/span&gt; &lt;span class="o"&gt;-&lt;/span&gt; &lt;span class="n"&gt;s&lt;/span&gt;&lt;span class="p"&gt;))&lt;/span&gt;
        &lt;span class="k"&gt;if&lt;/span&gt; &lt;span class="nf"&gt;abs&lt;/span&gt;&lt;span class="p"&gt;(&lt;/span&gt;&lt;span class="n"&gt;near&lt;/span&gt;&lt;span class="p"&gt;[&lt;/span&gt;&lt;span class="sh"&gt;"&lt;/span&gt;&lt;span class="s"&gt;start&lt;/span&gt;&lt;span class="sh"&gt;"&lt;/span&gt;&lt;span class="p"&gt;]&lt;/span&gt; &lt;span class="o"&gt;-&lt;/span&gt; &lt;span class="n"&gt;s&lt;/span&gt;&lt;span class="p"&gt;)&lt;/span&gt; &lt;span class="o"&gt;&amp;lt;=&lt;/span&gt; &lt;span class="mf"&gt;3.0&lt;/span&gt;&lt;span class="p"&gt;:&lt;/span&gt;
            &lt;span class="n"&gt;s&lt;/span&gt; &lt;span class="o"&gt;=&lt;/span&gt; &lt;span class="nf"&gt;max&lt;/span&gt;&lt;span class="p"&gt;(&lt;/span&gt;&lt;span class="mf"&gt;0.0&lt;/span&gt;&lt;span class="p"&gt;,&lt;/span&gt; &lt;span class="n"&gt;near&lt;/span&gt;&lt;span class="p"&gt;[&lt;/span&gt;&lt;span class="sh"&gt;"&lt;/span&gt;&lt;span class="s"&gt;start&lt;/span&gt;&lt;span class="sh"&gt;"&lt;/span&gt;&lt;span class="p"&gt;]&lt;/span&gt; &lt;span class="o"&gt;-&lt;/span&gt; &lt;span class="mf"&gt;0.3&lt;/span&gt;&lt;span class="p"&gt;)&lt;/span&gt;
&lt;/code&gt;&lt;/pre&gt;

&lt;/div&gt;



&lt;p&gt;In the example, the agent wrote 1:30, 1:59 and 3:05; the pack published 1:29, 1:58 and 3:04. Then it applies YouTube's rules, which are all or nothing: first chapter at 0:00 (if missing, it adds one called "Intro" and warns you to rename it), at least 3 chapters, each at least 10 seconds, none after the end. Break any of them and YouTube shows no chapters at all, so every break is a warning the agent has to fix before the pack goes to you.&lt;/p&gt;

&lt;h2&gt;
  
  
  Scoring frames for thumbnails
&lt;/h2&gt;

&lt;p&gt;&lt;code&gt;youtube_pack.py frames&lt;/code&gt; samples the video evenly at three times the number of candidates wanted (36 frames for the default 12), skipping the first and last 2 seconds. Each frame, scaled to 640 pixels wide, gets a score:&lt;br&gt;
&lt;/p&gt;

&lt;div class="highlight js-code-highlight"&gt;
&lt;pre class="highlight python"&gt;&lt;code&gt;&lt;span class="n"&gt;score&lt;/span&gt; &lt;span class="o"&gt;=&lt;/span&gt; &lt;span class="p"&gt;(&lt;/span&gt;&lt;span class="nf"&gt;min&lt;/span&gt;&lt;span class="p"&gt;(&lt;/span&gt;&lt;span class="n"&gt;sharp&lt;/span&gt;&lt;span class="p"&gt;,&lt;/span&gt; &lt;span class="mi"&gt;2500&lt;/span&gt;&lt;span class="p"&gt;)&lt;/span&gt; &lt;span class="o"&gt;/&lt;/span&gt; &lt;span class="mi"&gt;2500&lt;/span&gt;&lt;span class="p"&gt;)&lt;/span&gt; &lt;span class="o"&gt;*&lt;/span&gt; &lt;span class="mf"&gt;0.5&lt;/span&gt; \
      &lt;span class="o"&gt;+&lt;/span&gt; &lt;span class="p"&gt;(&lt;/span&gt;&lt;span class="nf"&gt;min&lt;/span&gt;&lt;span class="p"&gt;(&lt;/span&gt;&lt;span class="n"&gt;contrast&lt;/span&gt;&lt;span class="p"&gt;,&lt;/span&gt; &lt;span class="mi"&gt;70&lt;/span&gt;&lt;span class="p"&gt;)&lt;/span&gt; &lt;span class="o"&gt;/&lt;/span&gt; &lt;span class="mi"&gt;70&lt;/span&gt;&lt;span class="p"&gt;)&lt;/span&gt; &lt;span class="o"&gt;*&lt;/span&gt; &lt;span class="mf"&gt;0.3&lt;/span&gt; \
      &lt;span class="o"&gt;+&lt;/span&gt; &lt;span class="p"&gt;(&lt;/span&gt;&lt;span class="nf"&gt;min&lt;/span&gt;&lt;span class="p"&gt;(&lt;/span&gt;&lt;span class="n"&gt;hsv&lt;/span&gt;&lt;span class="p"&gt;,&lt;/span&gt; &lt;span class="mi"&gt;120&lt;/span&gt;&lt;span class="p"&gt;)&lt;/span&gt; &lt;span class="o"&gt;/&lt;/span&gt; &lt;span class="mi"&gt;120&lt;/span&gt;&lt;span class="p"&gt;)&lt;/span&gt; &lt;span class="o"&gt;*&lt;/span&gt; &lt;span class="mf"&gt;0.2&lt;/span&gt;
&lt;span class="k"&gt;if&lt;/span&gt; &lt;span class="n"&gt;mean&lt;/span&gt; &lt;span class="o"&gt;&amp;lt;&lt;/span&gt; &lt;span class="mi"&gt;25&lt;/span&gt; &lt;span class="ow"&gt;or&lt;/span&gt; &lt;span class="n"&gt;mean&lt;/span&gt; &lt;span class="o"&gt;&amp;gt;&lt;/span&gt; &lt;span class="mi"&gt;235&lt;/span&gt;&lt;span class="p"&gt;:&lt;/span&gt;
    &lt;span class="n"&gt;score&lt;/span&gt; &lt;span class="o"&gt;*=&lt;/span&gt; &lt;span class="mf"&gt;0.2&lt;/span&gt;          &lt;span class="c1"&gt;# black or blown-out frames
&lt;/span&gt;&lt;span class="k"&gt;if&lt;/span&gt; &lt;span class="n"&gt;faces&lt;/span&gt;&lt;span class="p"&gt;:&lt;/span&gt;
    &lt;span class="n"&gt;score&lt;/span&gt; &lt;span class="o"&gt;+=&lt;/span&gt; &lt;span class="mf"&gt;0.15&lt;/span&gt; &lt;span class="o"&gt;+&lt;/span&gt; &lt;span class="nf"&gt;min&lt;/span&gt;&lt;span class="p"&gt;(&lt;/span&gt;&lt;span class="n"&gt;face_share&lt;/span&gt; &lt;span class="o"&gt;*&lt;/span&gt; &lt;span class="mi"&gt;2&lt;/span&gt;&lt;span class="p"&gt;,&lt;/span&gt; &lt;span class="mf"&gt;0.2&lt;/span&gt;&lt;span class="p"&gt;)&lt;/span&gt;
&lt;/code&gt;&lt;/pre&gt;

&lt;/div&gt;



&lt;p&gt;Sharpness is the variance of an edge filter, contrast is the standard deviation of brightness, and the last term is colour saturation. Faces come from OpenCV's YuNet model, which ships in the repo, with the older Haar cascade as a fallback; without OpenCV, faces simply aren't scored.&lt;/p&gt;

&lt;p&gt;A high score alone would return twelve versions of the same shot, so each candidate gets a 64-bit difference hash and is kept only if it differs from every picked frame by more than 6 bits. A fixed-camera video can look identical all the way through; then the script fills the remaining slots with the frames furthest apart in time. The result is &lt;code&gt;frames.jpg&lt;/code&gt;, a numbered contact sheet with times and scores.&lt;/p&gt;

&lt;p&gt;The scores rank; they don't choose. The agent looks at the sheet and picks frames that carry the video's promise. When it renders, &lt;code&gt;thumbnail.py&lt;/code&gt; doesn't trust the exact timestamp either: it pulls six full-resolution frames across half a second around it and keeps the sharpest, so a moving shot doesn't produce a blurred thumbnail.&lt;/p&gt;

&lt;p&gt;The example's three ideas are deliberately different, not three crops of one frame: Earth drained dry next to "All of it in one drop", a surfer with "96.5% is salt water", and the full 370-quintillion figure with no words at all. Different ideas are what make an A/B test result mean something.&lt;/p&gt;

&lt;h2&gt;
  
  
  4K when the source has it, under 2 MB always
&lt;/h2&gt;

&lt;p&gt;YouTube now recommends 3840x2160 thumbnails. It allows 50 MB from a computer but only 2 MB from a phone. &lt;code&gt;thumbnail.py&lt;/code&gt; renders at the source's own size up to 3840x2160 (never upscaling past the source) and keeps every file under 2 MB, so it uploads from either:&lt;br&gt;
&lt;/p&gt;

&lt;div class="highlight js-code-highlight"&gt;
&lt;pre class="highlight python"&gt;&lt;code&gt;&lt;span class="k"&gt;def&lt;/span&gt; &lt;span class="nf"&gt;save_jpeg&lt;/span&gt;&lt;span class="p"&gt;(&lt;/span&gt;&lt;span class="n"&gt;im&lt;/span&gt;&lt;span class="p"&gt;,&lt;/span&gt; &lt;span class="n"&gt;path&lt;/span&gt;&lt;span class="p"&gt;):&lt;/span&gt;
    &lt;span class="k"&gt;while&lt;/span&gt; &lt;span class="bp"&gt;True&lt;/span&gt;&lt;span class="p"&gt;:&lt;/span&gt;
        &lt;span class="k"&gt;for&lt;/span&gt; &lt;span class="n"&gt;q&lt;/span&gt; &lt;span class="ow"&gt;in&lt;/span&gt; &lt;span class="p"&gt;(&lt;/span&gt;&lt;span class="mi"&gt;92&lt;/span&gt;&lt;span class="p"&gt;,&lt;/span&gt; &lt;span class="mi"&gt;88&lt;/span&gt;&lt;span class="p"&gt;,&lt;/span&gt; &lt;span class="mi"&gt;84&lt;/span&gt;&lt;span class="p"&gt;,&lt;/span&gt; &lt;span class="mi"&gt;80&lt;/span&gt;&lt;span class="p"&gt;,&lt;/span&gt; &lt;span class="mi"&gt;76&lt;/span&gt;&lt;span class="p"&gt;,&lt;/span&gt; &lt;span class="mi"&gt;72&lt;/span&gt;&lt;span class="p"&gt;):&lt;/span&gt;
            &lt;span class="n"&gt;im&lt;/span&gt;&lt;span class="p"&gt;.&lt;/span&gt;&lt;span class="nf"&gt;save&lt;/span&gt;&lt;span class="p"&gt;(&lt;/span&gt;&lt;span class="n"&gt;path&lt;/span&gt;&lt;span class="p"&gt;,&lt;/span&gt; &lt;span class="sh"&gt;"&lt;/span&gt;&lt;span class="s"&gt;JPEG&lt;/span&gt;&lt;span class="sh"&gt;"&lt;/span&gt;&lt;span class="p"&gt;,&lt;/span&gt; &lt;span class="n"&gt;quality&lt;/span&gt;&lt;span class="o"&gt;=&lt;/span&gt;&lt;span class="n"&gt;q&lt;/span&gt;&lt;span class="p"&gt;,&lt;/span&gt; &lt;span class="n"&gt;optimize&lt;/span&gt;&lt;span class="o"&gt;=&lt;/span&gt;&lt;span class="bp"&gt;True&lt;/span&gt;&lt;span class="p"&gt;,&lt;/span&gt; &lt;span class="n"&gt;progressive&lt;/span&gt;&lt;span class="o"&gt;=&lt;/span&gt;&lt;span class="bp"&gt;True&lt;/span&gt;&lt;span class="p"&gt;)&lt;/span&gt;
            &lt;span class="k"&gt;if&lt;/span&gt; &lt;span class="n"&gt;os&lt;/span&gt;&lt;span class="p"&gt;.&lt;/span&gt;&lt;span class="n"&gt;path&lt;/span&gt;&lt;span class="p"&gt;.&lt;/span&gt;&lt;span class="nf"&gt;getsize&lt;/span&gt;&lt;span class="p"&gt;(&lt;/span&gt;&lt;span class="n"&gt;path&lt;/span&gt;&lt;span class="p"&gt;)&lt;/span&gt; &lt;span class="o"&gt;&amp;lt;=&lt;/span&gt; &lt;span class="n"&gt;MAX_BYTES&lt;/span&gt;&lt;span class="p"&gt;:&lt;/span&gt;
                &lt;span class="k"&gt;return&lt;/span&gt; &lt;span class="n"&gt;q&lt;/span&gt;&lt;span class="p"&gt;,&lt;/span&gt; &lt;span class="n"&gt;im&lt;/span&gt;&lt;span class="p"&gt;.&lt;/span&gt;&lt;span class="n"&gt;size&lt;/span&gt;&lt;span class="p"&gt;,&lt;/span&gt; &lt;span class="bp"&gt;True&lt;/span&gt;
        &lt;span class="c1"&gt;# still too big: step down 2160 -&amp;gt; 1440 -&amp;gt; 1080 -&amp;gt; 720, then lower quality at 720
&lt;/span&gt;&lt;/code&gt;&lt;/pre&gt;

&lt;/div&gt;



&lt;p&gt;The example's source is 1080p, so its three thumbnails came out at 1920x1080 and quality 92, between about 150 and 190 KB each.&lt;/p&gt;

&lt;p&gt;Layout numbers are written once for a 1280-wide frame and scaled. Two of them do real work: words smaller than 80 pixels at 1280 wide trigger a warning (they vanish in the small card), and a 300x140 zone in the bottom-right corner stays clear, because the duration badge covers it.&lt;/p&gt;

&lt;h2&gt;
  
  
  Judged at the size people see it
&lt;/h2&gt;

&lt;p&gt;A 4K thumbnail viewed full screen tells you nothing. &lt;code&gt;preview.png&lt;/code&gt; draws each option, with your title and a duration badge, as a desktop home-feed card (360 pixels wide) and as the small card used for "up next" and phone search (168 pixels wide), on YouTube's dark theme. The workflow tells the agent to reject an option when its words can't be read in the small card or its subject is unclear at that size.&lt;/p&gt;

&lt;h2&gt;
  
  
  Shorts: 9:16, or YouTube replaces it
&lt;/h2&gt;

&lt;p&gt;A vertical or square video of 3 minutes or less is a Short, and the pack spots it from the probe (rotation metadata included, so a phone clip is measured the way it plays). For a Short it drops chapters and doesn't ask for &lt;code&gt;#shorts&lt;/code&gt;, since YouTube defines a Short by length and shape.&lt;/p&gt;

&lt;p&gt;Thumbnails change shape too. YouTube replaces a 16:9 custom thumbnail on a vertical video with an auto-generated frame, so &lt;code&gt;thumbnail.py&lt;/code&gt; renders 9:16 up to 2160x3840, centres the words in the upper part of the frame (the bottom third sits under the title and buttons), and swaps the preview for Shorts-shelf cards. Custom Shorts thumbnails can only be added in Studio on a computer, and Test &amp;amp; Compare doesn't run on Shorts; the pack's checklist says both.&lt;/p&gt;

&lt;h2&gt;
  
  
  Captions from the same transcript
&lt;/h2&gt;

&lt;p&gt;&lt;code&gt;captions.srt&lt;/code&gt; comes from the transcript the rest of the package was written from: fillers out, at most 2 lines of 42 characters, and a &lt;code&gt;replace&lt;/code&gt; map for words the transcriber got wrong (the example uses it to capitalise "Earth" throughout). The example's transcript came from the video's existing caption file through &lt;code&gt;--from&lt;/code&gt;, so word timing is estimated, and the pack prints that as a note rather than hiding it.&lt;/p&gt;

&lt;h2&gt;
  
  
  Mac, Windows and Linux
&lt;/h2&gt;

&lt;p&gt;The requirements are Python 3.8 or newer, any ffmpeg build and Pillow. The skill tells the agent to use &lt;code&gt;py&lt;/code&gt; instead of &lt;code&gt;python3&lt;/code&gt; on Windows, &lt;code&gt;check_setup.py&lt;/code&gt; prints the install command for the system it finds, and the fonts (Poppins ExtraBold for thumbnail words, Inter for the preview) are bundled so text renders the same everywhere. A CI workflow builds a pack and renders thumbnails on Ubuntu, Windows and macOS. OpenCV and the YouTube API key are optional extras; everything else works without them.&lt;/p&gt;

&lt;h2&gt;
  
  
  Uploading, and what stays in YouTube Studio
&lt;/h2&gt;

&lt;p&gt;&lt;code&gt;post.json&lt;/code&gt; is the post body for WoopSocial, a social media API and MCP server for AI agents. The agent creates an upload session with the byte size from &lt;code&gt;manifest.json&lt;/code&gt;, sends the parts with &lt;code&gt;upload_parts.py&lt;/code&gt;, fills in the two IDs, runs &lt;code&gt;posts_validate&lt;/code&gt;, and shows you the title, the description's first lines, chapters, tags, made for kids, privacy and time before it calls &lt;code&gt;posts_create&lt;/code&gt;. In a pre-release trial, the worked example's full YouTube post body passed &lt;code&gt;posts_validate&lt;/code&gt; against a real connected channel; it was not posted.&lt;/p&gt;

&lt;p&gt;The API sets the title, description, tags, category, made for kids, privacy and schedule. WoopSocial doesn't write any of that text and doesn't edit or crop the video; the skill builds the package locally. It also doesn't set the thumbnail, caption files, playlists, end screens or the AI-use label, so &lt;code&gt;upload-pack.md&lt;/code&gt; ends with those steps for YouTube Studio. After 2 and 7 days the agent reads views back and asks you for click-through rate and average view duration from Studio, and it never estimates a number it wasn't given.&lt;/p&gt;

&lt;p&gt;The free plan covers 2 social accounts, with everything from posting to analytics included, and setup for each agent is at &lt;a href="https://woopsocial.com/agents?utm_source=youtube-skill&amp;amp;utm_medium=devto" rel="noopener noreferrer"&gt;woopsocial.com/agents&lt;/a&gt;. Disclosure: the skill is maintained by Frank Heijdenrijk and the team behind &lt;a href="https://woopsocial.com/?utm_source=youtube-skill&amp;amp;utm_medium=devto" rel="noopener noreferrer"&gt;WoopSocial&lt;/a&gt;. Uploading is optional; without it the skill hands you the pack to upload yourself.&lt;/p&gt;

&lt;h2&gt;
  
  
  Try it
&lt;/h2&gt;



&lt;div class="highlight js-code-highlight"&gt;
&lt;pre class="highlight shell"&gt;&lt;code&gt;npx skills add social-media-skills/youtube-skill &lt;span class="nt"&gt;-g&lt;/span&gt; &lt;span class="nt"&gt;-a&lt;/span&gt; claude-code &lt;span class="nt"&gt;-y&lt;/span&gt;
&lt;/code&gt;&lt;/pre&gt;

&lt;/div&gt;



&lt;p&gt;OpenClaw: &lt;code&gt;openclaw skills install @social-media-skills/youtube-skill&lt;/code&gt;. Codex, Cursor and other agents: run &lt;code&gt;npx skills add social-media-skills/youtube-skill&lt;/code&gt; and pick your agent. You need an agent that can run ffmpeg on your machine.&lt;/p&gt;

&lt;p&gt;Then hand it a finished video: "put final.mp4 on my channel and make it findable." Issues and pull requests are welcome on &lt;a href="https://github.com/social-media-skills/youtube-skill" rel="noopener noreferrer"&gt;GitHub&lt;/a&gt;, and more free skills for social media work are at &lt;a href="https://social-media-skills.com" rel="noopener noreferrer"&gt;social-media-skills.com&lt;/a&gt;.&lt;/p&gt;

</description>
      <category>showdev</category>
      <category>claudecode</category>
      <category>youtube</category>
      <category>python</category>
    </item>
    <item>
      <title>Video Editing Skill for Claude Code: An Agent That Plans Every Cut, Then Renders With ffmpeg</title>
      <dc:creator>Frank Heijdenrijk</dc:creator>
      <pubDate>Fri, 30 Oct 2026 10:42:28 +0000</pubDate>
      <link>https://dev.to/fheijdenrijk/video-editing-skill-for-claude-code-an-agent-that-plans-every-cut-then-renders-with-ffmpeg-518a</link>
      <guid>https://dev.to/fheijdenrijk/video-editing-skill-for-claude-code-an-agent-that-plans-every-cut-then-renders-with-ffmpeg-518a</guid>
      <description>&lt;p&gt;Video editing skill is a literal name: it's a free, MIT-licensed agent skill for Claude Code, Codex, Cursor and OpenClaw that does the first edit of a talking-head recording. It transcribes with word timestamps, writes a cut list where every cut has a reason, waits for your approval, then renders a frame-exact edit with punch-ins on the jump cuts, levelled audio, an SRT caption file and YouTube chapters. Three Python scripts do the deterministic work with ffmpeg and Pillow on your machine; the agent does the reading and judging.&lt;/p&gt;

&lt;p&gt;Repo: &lt;a href="https://github.com/social-media-skills/video-editing-skill" rel="noopener noreferrer"&gt;github.com/social-media-skills/video-editing-skill&lt;/a&gt;&lt;/p&gt;

&lt;p&gt;&lt;a href="https://media2.dev.to/dynamic/image/width=800%2Cheight=%2Cfit=scale-down%2Cgravity=auto%2Cformat=auto/https%3A%2F%2Fdev-to-uploads.s3.us-east-2.amazonaws.com%2Fuploads%2Farticles%2F7ersh0d9kzqrf7f7desq.png" class="article-body-image-wrapper"&gt;&lt;img src="https://media2.dev.to/dynamic/image/width=800%2Cheight=%2Cfit=scale-down%2Cgravity=auto%2Cformat=auto/https%3A%2F%2Fdev-to-uploads.s3.us-east-2.amazonaws.com%2Fuploads%2Farticles%2F7ersh0d9kzqrf7f7desq.png" alt="Before and after frames of a NASA interview above a timeline of kept and cut time" width="799" height="353"&gt;&lt;/a&gt;&lt;/p&gt;

&lt;p&gt;&lt;em&gt;The repo's worked example: a 6:39 public-domain NASA Goddard interview edited to 5:57, question cards kept. Footage courtesy of NASA; not endorsed by NASA.&lt;/em&gt;&lt;/p&gt;

&lt;h2&gt;
  
  
  The split: scripts render, the agent decides
&lt;/h2&gt;

&lt;p&gt;Finding silence is a solved problem in ffmpeg. What makes a first edit usable is everything around it: knowing that "um" is safe to cut and "like" often isn't, telling a retake from a sentence that just happens to start the same way, spotting a tangent, keeping an interviewer's question on screen, and not clipping the last consonant of a word. Those calls need something that reads the transcript. So the skill is a &lt;code&gt;SKILL.md&lt;/code&gt; workflow plus scripts that never guess:&lt;br&gt;
&lt;/p&gt;

&lt;div class="highlight js-code-highlight"&gt;
&lt;pre class="highlight plaintext"&gt;&lt;code&gt;video.mp4
  -&amp;gt; check_setup.py     Python, ffmpeg, Pillow, transcription routes, install commands
  -&amp;gt; transcribe.py      work/transcript.json (every word, start, end) + transcript.txt
       --terms "OSIRIS-REx, Bennu"   names and jargon spelled right
  -&amp;gt; agent reads the whole transcript, writes edit.json
       style, extra fillers, cut ranges with reasons, keep ranges, replace map, chapters
  -&amp;gt; edit_video.py --plan   edited/edit-plan.md: summary, removals, possible retakes
  -&amp;gt; you approve
  -&amp;gt; edit_video.py      edited.mp4, captions.srt, chapters.txt, cuts.jpg,
                        edit-report.md, manifest.json (with warnings)
  -&amp;gt; agent checks cuts.jpg and the warnings; re-renders if needed
  -&amp;gt; WoopSocial MCP     upload, validate, confirm, post
&lt;/code&gt;&lt;/pre&gt;

&lt;/div&gt;



&lt;p&gt;&lt;code&gt;transcript.txt&lt;/code&gt; is one timestamped line per sentence, compact enough for an agent to read an hour of speech in one go. &lt;code&gt;transcript.json&lt;/code&gt; holds the word timings the editor actually cuts on.&lt;/p&gt;

&lt;h2&gt;
  
  
  Getting the fillers and the names into the transcript
&lt;/h2&gt;

&lt;p&gt;Transcript-driven editing quietly breaks when the transcript is too clean. OpenAI's speech-to-text guide notes that the model may leave out filler words unless the prompt contains some, and an "um" that was never written down can't be cut. &lt;code&gt;transcribe.py&lt;/code&gt; sends a disfluent prompt on both the API and the local route, and appends whatever you pass with &lt;code&gt;--terms&lt;/code&gt;:&lt;br&gt;
&lt;/p&gt;

&lt;div class="highlight js-code-highlight"&gt;
&lt;pre class="highlight python"&gt;&lt;code&gt;&lt;span class="n"&gt;FILLER_PROMPT&lt;/span&gt; &lt;span class="o"&gt;=&lt;/span&gt; &lt;span class="sh"&gt;"&lt;/span&gt;&lt;span class="s"&gt;Umm, let me think, like, hmm... Uh, okay, so, um, here&lt;/span&gt;&lt;span class="sh"&gt;'&lt;/span&gt;&lt;span class="s"&gt;s what I, uh, mean.&lt;/span&gt;&lt;span class="sh"&gt;"&lt;/span&gt;
&lt;span class="n"&gt;prompt&lt;/span&gt; &lt;span class="o"&gt;=&lt;/span&gt; &lt;span class="n"&gt;FILLER_PROMPT&lt;/span&gt; &lt;span class="o"&gt;+&lt;/span&gt; &lt;span class="p"&gt;(&lt;/span&gt;&lt;span class="sa"&gt;f&lt;/span&gt;&lt;span class="sh"&gt;"&lt;/span&gt;&lt;span class="s"&gt; Names and terms: &lt;/span&gt;&lt;span class="si"&gt;{&lt;/span&gt;&lt;span class="n"&gt;a&lt;/span&gt;&lt;span class="p"&gt;.&lt;/span&gt;&lt;span class="n"&gt;terms&lt;/span&gt;&lt;span class="si"&gt;}&lt;/span&gt;&lt;span class="s"&gt;.&lt;/span&gt;&lt;span class="sh"&gt;"&lt;/span&gt; &lt;span class="k"&gt;if&lt;/span&gt; &lt;span class="n"&gt;a&lt;/span&gt;&lt;span class="p"&gt;.&lt;/span&gt;&lt;span class="n"&gt;terms&lt;/span&gt; &lt;span class="k"&gt;else&lt;/span&gt; &lt;span class="sh"&gt;""&lt;/span&gt;&lt;span class="p"&gt;)&lt;/span&gt;
&lt;/code&gt;&lt;/pre&gt;

&lt;/div&gt;



&lt;p&gt;Names that still come out wrong get fixed in &lt;code&gt;edit.json&lt;/code&gt; before any caption is written. In the example, the transcriber heard the mission name as "Cyrus Rex":&lt;br&gt;
&lt;/p&gt;

&lt;div class="highlight js-code-highlight"&gt;
&lt;pre class="highlight json"&gt;&lt;code&gt;&lt;span class="nl"&gt;"replace"&lt;/span&gt;&lt;span class="p"&gt;:&lt;/span&gt;&lt;span class="w"&gt; &lt;/span&gt;&lt;span class="p"&gt;{&lt;/span&gt;&lt;span class="nl"&gt;"Cyrus Rex"&lt;/span&gt;&lt;span class="p"&gt;:&lt;/span&gt;&lt;span class="w"&gt; &lt;/span&gt;&lt;span class="s2"&gt;"OSIRIS-REx"&lt;/span&gt;&lt;span class="p"&gt;,&lt;/span&gt;&lt;span class="w"&gt; &lt;/span&gt;&lt;span class="nl"&gt;"gimbled"&lt;/span&gt;&lt;span class="p"&gt;:&lt;/span&gt;&lt;span class="w"&gt; &lt;/span&gt;&lt;span class="s2"&gt;"gimbaled"&lt;/span&gt;&lt;span class="p"&gt;}&lt;/span&gt;&lt;span class="w"&gt;
&lt;/span&gt;&lt;/code&gt;&lt;/pre&gt;

&lt;/div&gt;



&lt;p&gt;Three transcription routes, picked in this order by &lt;code&gt;--engine auto&lt;/code&gt;:&lt;/p&gt;

&lt;div class="table-wrapper-paragraph"&gt;&lt;table&gt;
&lt;thead&gt;
&lt;tr&gt;
&lt;th&gt;Route&lt;/th&gt;
&lt;th&gt;Flag or env&lt;/th&gt;
&lt;th&gt;Trade-off&lt;/th&gt;
&lt;/tr&gt;
&lt;/thead&gt;
&lt;tbody&gt;
&lt;tr&gt;
&lt;td&gt;Existing transcript&lt;/td&gt;
&lt;td&gt;&lt;code&gt;--from captions.srt&lt;/code&gt;&lt;/td&gt;
&lt;td&gt;Free, instant; word timing estimated, most fillers missing&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;Groq or OpenAI&lt;/td&gt;
&lt;td&gt;
&lt;code&gt;GROQ_API_KEY&lt;/code&gt; / &lt;code&gt;OPENAI_API_KEY&lt;/code&gt;
&lt;/td&gt;
&lt;td&gt;Fast; Groq is about $0.04 per audio hour; long audio is sent in 20-minute chunks&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;Local&lt;/td&gt;
&lt;td&gt;&lt;code&gt;pip install faster-whisper&lt;/code&gt;&lt;/td&gt;
&lt;td&gt;Free, private; roughly 0.5 to 1x realtime with &lt;code&gt;--model small&lt;/code&gt;
&lt;/td&gt;
&lt;/tr&gt;
&lt;/tbody&gt;
&lt;/table&gt;&lt;/div&gt;

&lt;h2&gt;
  
  
  Detection heuristics, and what is deliberately not automatic
&lt;/h2&gt;

&lt;p&gt;Everything runs over the word list after lowercasing and stripping punctuation. Each rule returns a reason per word, and that reason ends up in the plan.&lt;/p&gt;

&lt;p&gt;&lt;strong&gt;Fillers.&lt;/strong&gt; The default set is &lt;code&gt;um, uh, uhm, umm, er, erm, ah, hmm&lt;/code&gt;, cut only when the word has a small gap on at least one side; one fused to both neighbours would take part of them with it. "Mm hmm" as agreement stays. Habit words ("like", "so", "you know", "basically") are opt-in through &lt;code&gt;extra_fillers&lt;/code&gt;, and single words are cut only next to a comma.&lt;/p&gt;

&lt;p&gt;&lt;strong&gt;Stutters.&lt;/strong&gt; The same word twice with under 0.25 s between and no comma, or the same two words twice. A stop list (&lt;code&gt;very, really, that, had, is, no, yeah, so, well&lt;/code&gt; and a few more) is never touched, because "had had" and "no, no" are usually meant.&lt;/p&gt;

&lt;p&gt;&lt;strong&gt;Retake cues.&lt;/strong&gt; "let me say that again", "let me start over", "let me try that again", "scratch that", "take two". The cue goes. The sentence before it goes too only when the next sentence repeats at least half of its words, so a cue on its own can't delete a good sentence.&lt;/p&gt;

&lt;p&gt;&lt;strong&gt;Restarts.&lt;/strong&gt; An unfinished fragment (no end punctuation, eight words or fewer) followed by a sentence that opens with the same words is cut; the last take stays.&lt;/p&gt;

&lt;p&gt;&lt;strong&gt;Possible retakes.&lt;/strong&gt; Two finished sentences that open the same way ("It was great. It was fun.") are never cut automatically. They're listed in the plan, and the agent decides each one with you, adding real retakes to &lt;code&gt;cut&lt;/code&gt;.&lt;/p&gt;

&lt;p&gt;&lt;strong&gt;Agent picks.&lt;/strong&gt; Tangents, "can you hear me?", a long set-up before the point go in &lt;code&gt;edit.json&lt;/code&gt; as &lt;code&gt;"cut": [[61.2, 74.8, "tangent about the weather"]]&lt;/code&gt;. &lt;code&gt;"keep"&lt;/code&gt; ranges override every automatic rule, which is how the example keeps about 3 seconds of each interview question card.&lt;/p&gt;

&lt;h2&gt;
  
  
  Pauses become beats, and small cuts are skipped
&lt;/h2&gt;

&lt;p&gt;Deleting every gap makes a speaker sound breathless. Each pause over the style's limit shrinks to a beat that depends on context:&lt;/p&gt;

&lt;div class="table-wrapper-paragraph"&gt;&lt;table&gt;
&lt;thead&gt;
&lt;tr&gt;
&lt;th&gt;Style&lt;/th&gt;
&lt;th&gt;Shorten gaps over&lt;/th&gt;
&lt;th&gt;Mid-sentence&lt;/th&gt;
&lt;th&gt;After a sentence&lt;/th&gt;
&lt;th&gt;Long gap or card&lt;/th&gt;
&lt;th&gt;Skip removals under&lt;/th&gt;
&lt;/tr&gt;
&lt;/thead&gt;
&lt;tbody&gt;
&lt;tr&gt;
&lt;td&gt;youtube&lt;/td&gt;
&lt;td&gt;0.50 s&lt;/td&gt;
&lt;td&gt;0.30 s&lt;/td&gt;
&lt;td&gt;0.45 s&lt;/td&gt;
&lt;td&gt;0.90 s&lt;/td&gt;
&lt;td&gt;0.40 s&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;tight&lt;/td&gt;
&lt;td&gt;0.30 s&lt;/td&gt;
&lt;td&gt;0.22 s&lt;/td&gt;
&lt;td&gt;0.25 s&lt;/td&gt;
&lt;td&gt;0.50 s&lt;/td&gt;
&lt;td&gt;0.25 s&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;podcast&lt;/td&gt;
&lt;td&gt;0.80 s&lt;/td&gt;
&lt;td&gt;0.40 s&lt;/td&gt;
&lt;td&gt;0.60 s&lt;/td&gt;
&lt;td&gt;1.00 s&lt;/td&gt;
&lt;td&gt;0.60 s&lt;/td&gt;
&lt;/tr&gt;
&lt;/tbody&gt;
&lt;/table&gt;&lt;/div&gt;

&lt;p&gt;The last column matters more than it looks. Every cut inside one continuous shot is a visible jump, so a removal that saves less than the minimum is left alone: a dozen tiny cuts save a few seconds and make the video look nervous. Fillers, stutters and retakes are always cut regardless. A kept piece under 0.4 s joins a touching neighbour so nothing flickers, and the kept beat is the real room tone, never digital silence.&lt;/p&gt;

&lt;h2&gt;
  
  
  Cut edges placed by ear and by eye
&lt;/h2&gt;

&lt;p&gt;Transcribers often put a word's end 50 to 150 ms early, so cutting on raw timestamps clips word endings. The script measures the audio's level in 10 ms windows and estimates the room's noise floor, then moves each edge:&lt;/p&gt;

&lt;ul&gt;
&lt;li&gt;
&lt;strong&gt;Out-points&lt;/strong&gt; move forward until the audio has stayed below the noise floor plus 6 dB for 30 ms, up to 250 ms.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;In-points&lt;/strong&gt; move back to where the sound starts, up to 200 ms, and never closer than 100 ms before the word.&lt;/li&gt;
&lt;li&gt;Edges never move into a neighbouring word.&lt;/li&gt;
&lt;/ul&gt;

&lt;p&gt;The picture gets one pass too. ffmpeg's &lt;code&gt;blackdetect&lt;/code&gt; and a scene-change &lt;code&gt;select&lt;/code&gt; on a 160-pixel-wide copy find title cards and shot changes. A card inside a removed gap goes entirely, and a shot change within half a second of a kept edge pushes the edge past it, so no stray frames of a card or a second camera angle flash by.&lt;/p&gt;

&lt;h2&gt;
  
  
  Frame-exact rendering in groups
&lt;/h2&gt;

&lt;p&gt;Cutting hundreds of pieces as separate files and concatenating them lets audio and video drift by a few milliseconds per join, which adds up. Here every edge is rounded to the frame grid, and each group of up to 60 pieces is trimmed and joined inside one filter graph:&lt;br&gt;
&lt;/p&gt;

&lt;div class="highlight js-code-highlight"&gt;
&lt;pre class="highlight plaintext"&gt;&lt;code&gt;[0:v]trim=start=S:end=E,setpts=PTS-STARTPTS,&amp;lt;crop/scale&amp;gt;[v0]
[0:a]atrim=start=S:end=E,asetpts=PTS-STARTPTS,afade=t=in:d=0.004,afade=t=out:...[a0]
...
[v0][a0][v1][a1]...concat=n=N:v=1:a=1[cv][ca]
&lt;/code&gt;&lt;/pre&gt;

&lt;/div&gt;



&lt;p&gt;The 4 ms fades stop clicks without dipping the room tone. After the render the script compares the audio and video stream durations and adds a warning to &lt;code&gt;manifest.json&lt;/code&gt; if they differ by more than a frame.&lt;/p&gt;

&lt;h2&gt;
  
  
  Punch-ins where a jump cut needs hiding
&lt;/h2&gt;

&lt;p&gt;Alternating between the full frame and a 1.1x crop makes a jump cut read as an edit. The script switches framing at every cut that removed a second or more, and otherwise only once 4 seconds (2.5 in tight) have passed on the same framing. It never punches in at a cut onto a new camera shot, which already changes the picture, and never on a title card. The crop is anchored to an eye line at 38% of the frame height so the eyes stay level between framings, the zoom is capped at 1.2x to keep 1080p sharp, and &lt;code&gt;"punch_in": 0&lt;/code&gt; turns it off for screen recordings.&lt;/p&gt;

&lt;h2&gt;
  
  
  The audio chain
&lt;/h2&gt;

&lt;p&gt;Every filter here ships in standard ffmpeg builds, so no special build is needed on any OS:&lt;/p&gt;

&lt;ol&gt;
&lt;li&gt;
&lt;code&gt;volume&lt;/code&gt; to bring the recording to about -20 LUFS, a working level.&lt;/li&gt;
&lt;li&gt;
&lt;code&gt;highpass=f=80&lt;/code&gt; for rumble.&lt;/li&gt;
&lt;li&gt;
&lt;code&gt;afftdn&lt;/code&gt; with its noise floor set from the measured room noise, skipped entirely when the room is already under -60 dB, so quiet word endings aren't eaten.&lt;/li&gt;
&lt;li&gt;
&lt;code&gt;acompressor&lt;/code&gt;, 3:1 from -24 dB, 5 ms attack, 120 ms release.&lt;/li&gt;
&lt;li&gt;
&lt;code&gt;alimiter&lt;/code&gt;, set so peaks end at -2 dB after the final gain, leaving headroom for the AAC encoder.&lt;/li&gt;
&lt;li&gt;One linear &lt;code&gt;volume&lt;/code&gt; gain to -14 LUFS, measured on the processed audio, so there's no pumping.&lt;/li&gt;
&lt;/ol&gt;

&lt;p&gt;-14 LUFS is the level YouTube and the short-form apps are widely measured to play at; the podcast style uses -16.&lt;/p&gt;

&lt;h2&gt;
  
  
  Captions and chapters that survive the cuts
&lt;/h2&gt;

&lt;p&gt;Every word keeps its source timestamps, and &lt;code&gt;to_out()&lt;/code&gt; maps a source time onto the edited timeline by walking the kept spans. Captions are built from kept words only, with &lt;code&gt;replace&lt;/code&gt; applied: at most 2 lines of 42 characters, 1 to 7 seconds each, gaps under 0.3 s closed, and no line break after a small word ("the", "of", "my") or inside a name. They go to &lt;code&gt;captions.srt&lt;/code&gt; for upload in YouTube Studio.&lt;/p&gt;

&lt;p&gt;With &lt;code&gt;"burn": true&lt;/code&gt;, Pillow renders each caption as a transparent PNG at 4.2% of the frame height and ffmpeg overlays the sequence, so ffmpeg's text filters aren't needed. Vertical versions burn by default, one line of up to 32 characters. The skill tells the agent not to burn captions into a YouTube video that also gets the SRT, since viewers with captions on would see both.&lt;/p&gt;

&lt;p&gt;Chapters follow the same mapping. The agent writes them in original-video times, named for what the viewer gets; the script moves them through the cut list and writes &lt;code&gt;chapters.txt&lt;/code&gt; with the first at 0:00, warning if any chapter drops under YouTube's 10-second minimum.&lt;/p&gt;

&lt;h2&gt;
  
  
  The plan and the check sheet
&lt;/h2&gt;

&lt;p&gt;&lt;code&gt;--plan&lt;/code&gt; writes &lt;code&gt;edit-plan.md&lt;/code&gt; without rendering. The example's summary line:&lt;br&gt;
&lt;/p&gt;

&lt;div class="highlight js-code-highlight"&gt;
&lt;pre class="highlight plaintext"&gt;&lt;code&gt;6:39 -&amp;gt; 5:57 (11% shorter; 9 s from inside the talking, 34 s from long gaps,
cards, opening and ending). 12 cuts, pace 183 words per minute.
&lt;/code&gt;&lt;/pre&gt;

&lt;/div&gt;



&lt;p&gt;Splitting time removed inside the talking from time removed in gaps, cards and the ends keeps the headline honest: in the example most of the saving is a 24-second ending and long gaps around the question cards, while the talking itself lost about 9 seconds, because the interview was already well delivered. The plan then lists every removal over a second with its reason and the possible retakes. The report flags a pace over 190 words per minute and any edit that removes more than 45%.&lt;/p&gt;

&lt;p&gt;A render also writes &lt;code&gt;cuts.jpg&lt;/code&gt;: frames just before and after the biggest cuts, side by side. The agent opens it to look for a flash of a card, a jump into another shot or a frozen expression, instead of claiming it watched the whole video.&lt;/p&gt;

&lt;h2&gt;
  
  
  Posting
&lt;/h2&gt;

&lt;p&gt;After the render, the skill can post through WoopSocial, a social media API and MCP server for AI agents. The agent creates an upload session with the byte size from &lt;code&gt;manifest.json&lt;/code&gt;, uploads the parts with &lt;code&gt;upload_parts.py&lt;/code&gt;, builds a YouTube post (title, privacy, a description that starts with what the viewer gets and then the chapters), runs &lt;code&gt;posts_validate&lt;/code&gt;, and shows you the whole post before calling &lt;code&gt;posts_create&lt;/code&gt;. Caption files aren't sent through the API; you upload &lt;code&gt;captions.srt&lt;/code&gt; in YouTube Studio. Setup per agent is at &lt;a href="https://woopsocial.com/agents?utm_source=video-editing-skill&amp;amp;utm_medium=devto" rel="noopener noreferrer"&gt;woopsocial.com/agents&lt;/a&gt;, and the free plan covers 2 social accounts. Disclosure: the skill is maintained by Frank Heijdenrijk and the team behind &lt;a href="https://woopsocial.com/?utm_source=video-editing-skill&amp;amp;utm_medium=devto" rel="noopener noreferrer"&gt;WoopSocial&lt;/a&gt;. Posting is optional; the edited files are yours either way.&lt;/p&gt;

&lt;h2&gt;
  
  
  Try it
&lt;/h2&gt;



&lt;div class="highlight js-code-highlight"&gt;
&lt;pre class="highlight shell"&gt;&lt;code&gt;npx skills add social-media-skills/video-editing-skill &lt;span class="nt"&gt;-g&lt;/span&gt; &lt;span class="nt"&gt;-a&lt;/span&gt; claude-code &lt;span class="nt"&gt;-y&lt;/span&gt;
&lt;/code&gt;&lt;/pre&gt;

&lt;/div&gt;



&lt;p&gt;Other routes: &lt;code&gt;openclaw skills install @social-media-skills/video-editing-skill&lt;/code&gt;, &lt;code&gt;gh skill install social-media-skills/video-editing-skill video-editing-skill&lt;/code&gt;, or &lt;code&gt;gemini extensions install https://github.com/social-media-skills/video-editing-skill&lt;/code&gt;. You need Python 3.8+, ffmpeg and Pillow, and an agent that can run ffmpeg locally; CI runs a transcribe, plan and render on Ubuntu, Windows and macOS.&lt;/p&gt;

&lt;p&gt;Then point it at a recording: "edit lesson3.mp4 for YouTube, cut the silences and ums, add chapters." Issues and pull requests are welcome on &lt;a href="https://github.com/social-media-skills/video-editing-skill" rel="noopener noreferrer"&gt;GitHub&lt;/a&gt;. More free skills for social media work are at &lt;a href="https://social-media-skills.com" rel="noopener noreferrer"&gt;social-media-skills.com&lt;/a&gt;.&lt;/p&gt;

</description>
      <category>showdev</category>
      <category>claudecode</category>
      <category>opensource</category>
      <category>python</category>
    </item>
    <item>
      <title>Video to Shorts Skill for Claude Code: Transcribe, Score, Cut and Post Vertical Clips From Your Agent</title>
      <dc:creator>Frank Heijdenrijk</dc:creator>
      <pubDate>Thu, 29 Oct 2026 13:57:03 +0000</pubDate>
      <link>https://dev.to/fheijdenrijk/video-to-shorts-skill-for-claude-code-transcribe-score-cut-and-post-vertical-clips-from-your-34c6</link>
      <guid>https://dev.to/fheijdenrijk/video-to-shorts-skill-for-claude-code-transcribe-score-cut-and-post-vertical-clips-from-your-34c6</guid>
      <description>&lt;p&gt;Video to shorts, done by your coding agent: the Video to Shorts Skill is a free, MIT-licensed agent skill for Claude Code, Codex, Cursor and OpenClaw that takes a long recording, finds the moments worth clipping, renders them as 1080x1920 vertical clips with word-by-word captions, and posts them to TikTok, Instagram Reels, YouTube Shorts and Facebook Reels once you say yes. The heavy lifting is three Python scripts, ffmpeg and Pillow, all running on your machine.&lt;/p&gt;

&lt;p&gt;Repo: &lt;a href="https://github.com/social-media-skills/video-to-shorts-skill" rel="noopener noreferrer"&gt;github.com/social-media-skills/video-to-shorts-skill&lt;/a&gt;&lt;/p&gt;

&lt;p&gt;&lt;a href="https://media2.dev.to/dynamic/image/width=800%2Cheight=%2Cfit=scale-down%2Cgravity=auto%2Cformat=auto/https%3A%2F%2Fdev-to-uploads.s3.us-east-2.amazonaws.com%2Fuploads%2Farticles%2Fbg0nsa4qycwhrtuescfo.png" class="article-body-image-wrapper"&gt;&lt;img src="https://media2.dev.to/dynamic/image/width=800%2Cheight=%2Cfit=scale-down%2Cgravity=auto%2Cformat=auto/https%3A%2F%2Fdev-to-uploads.s3.us-east-2.amazonaws.com%2Fuploads%2Farticles%2Fbg0nsa4qycwhrtuescfo.png" alt="A 9-minute NASA interview turned into three vertical shorts, each with a hook on a white box and word-by-word captions" width="799" height="333"&gt;&lt;/a&gt;&lt;/p&gt;

&lt;p&gt;&lt;em&gt;Frames from the repo's worked example, cut from a public-domain NASA interview (jsc2020m000132). Footage courtesy of NASA; not endorsed by NASA.&lt;/em&gt;&lt;/p&gt;

&lt;h2&gt;
  
  
  Why a skill and not a single script
&lt;/h2&gt;

&lt;p&gt;Cutting a clip is the easy part. The judgment calls are where clips fail: which 40 seconds out of an hour, where the first word lands, whether a stranger can follow it, whether the crop still has the speaker in it. Those calls need something that can read a transcript and look at a frame, which is what an agent already does. So the skill splits the work: deterministic scripts for transcription and rendering, and a &lt;code&gt;SKILL.md&lt;/code&gt; that tells the agent how to choose, check and confirm.&lt;/p&gt;

&lt;h2&gt;
  
  
  The pipeline
&lt;/h2&gt;



&lt;div class="highlight js-code-highlight"&gt;
&lt;pre class="highlight plaintext"&gt;&lt;code&gt;video.mp4
  -&amp;gt; check_setup.py        what this machine can do, with install commands
  -&amp;gt; transcribe.py         work/transcript.txt + work/transcript.json (word timings)
  -&amp;gt; agent reads transcript, scores candidates with the rubric, you pick
  -&amp;gt; clips.json            start, end, hook, title, caption, framing per clip
  -&amp;gt; make_shorts.py --preview   three frames per clip with the planned crop
  -&amp;gt; make_shorts.py        shorts/clip-01.mp4 ..., contact-sheet.jpg, manifest.json
  -&amp;gt; WoopSocial MCP        upload, validate, confirm, post, read results
&lt;/code&gt;&lt;/pre&gt;

&lt;/div&gt;



&lt;h2&gt;
  
  
  Install
&lt;/h2&gt;



&lt;div class="highlight js-code-highlight"&gt;
&lt;pre class="highlight shell"&gt;&lt;code&gt;npx skills add social-media-skills/video-to-shorts-skill &lt;span class="nt"&gt;-g&lt;/span&gt; &lt;span class="nt"&gt;-a&lt;/span&gt; claude-code &lt;span class="nt"&gt;-y&lt;/span&gt;
&lt;/code&gt;&lt;/pre&gt;

&lt;/div&gt;



&lt;p&gt;Other routes: &lt;code&gt;openclaw skills install @social-media-skills/video-to-shorts-skill&lt;/code&gt; (ClawHub), &lt;code&gt;gh skill install social-media-skills/video-to-shorts-skill video-to-shorts-skill&lt;/code&gt; (GitHub CLI), or &lt;code&gt;gemini extensions install https://github.com/social-media-skills/video-to-shorts-skill&lt;/code&gt; (Gemini CLI). For Codex or Cursor, run &lt;code&gt;npx skills add social-media-skills/video-to-shorts-skill&lt;/code&gt; and pick the agent.&lt;/p&gt;

&lt;p&gt;Dependencies are deliberately small: Python 3.8+, ffmpeg and Pillow.&lt;/p&gt;

&lt;div class="table-wrapper-paragraph"&gt;&lt;table&gt;
&lt;thead&gt;
&lt;tr&gt;
&lt;th&gt;OS&lt;/th&gt;
&lt;th&gt;ffmpeg&lt;/th&gt;
&lt;/tr&gt;
&lt;/thead&gt;
&lt;tbody&gt;
&lt;tr&gt;
&lt;td&gt;macOS&lt;/td&gt;
&lt;td&gt;&lt;code&gt;brew install ffmpeg&lt;/code&gt;&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;Windows&lt;/td&gt;
&lt;td&gt;&lt;code&gt;winget install ffmpeg&lt;/code&gt;&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;Debian, Ubuntu&lt;/td&gt;
&lt;td&gt;&lt;code&gt;sudo apt install ffmpeg&lt;/code&gt;&lt;/td&gt;
&lt;/tr&gt;
&lt;/tbody&gt;
&lt;/table&gt;&lt;/div&gt;

&lt;p&gt;Then &lt;code&gt;pip install pillow&lt;/code&gt;. The skill needs an agent that can run ffmpeg locally; it won't render real video inside a browser chat sandbox.&lt;/p&gt;

&lt;h2&gt;
  
  
  Step 1: check_setup.py
&lt;/h2&gt;



&lt;div class="highlight js-code-highlight"&gt;
&lt;pre class="highlight shell"&gt;&lt;code&gt;python3 scripts/check_setup.py
&lt;/code&gt;&lt;/pre&gt;

&lt;/div&gt;



&lt;p&gt;It prints ok, missing or optional for ffmpeg, Pillow, each transcription route and OpenCV, plus the install command for the OS it's running on. The agent runs it first so it never tries to render on a machine that can't.&lt;/p&gt;

&lt;h2&gt;
  
  
  Step 2: transcribe.py and the four routes
&lt;/h2&gt;



&lt;div class="highlight js-code-highlight"&gt;
&lt;pre class="highlight shell"&gt;&lt;code&gt;python3 scripts/transcribe.py VIDEO &lt;span class="nt"&gt;--out&lt;/span&gt; work/                 &lt;span class="c"&gt;# best route available&lt;/span&gt;
python3 scripts/transcribe.py VIDEO &lt;span class="nt"&gt;--engine&lt;/span&gt; &lt;span class="nb"&gt;local&lt;/span&gt; &lt;span class="nt"&gt;--model&lt;/span&gt; small
python3 scripts/transcribe.py VIDEO &lt;span class="nt"&gt;--from&lt;/span&gt; captions.srt         &lt;span class="c"&gt;# existing SRT or VTT&lt;/span&gt;
&lt;/code&gt;&lt;/pre&gt;

&lt;/div&gt;



&lt;div class="table-wrapper-paragraph"&gt;&lt;table&gt;
&lt;thead&gt;
&lt;tr&gt;
&lt;th&gt;Route&lt;/th&gt;
&lt;th&gt;Trigger&lt;/th&gt;
&lt;th&gt;Trade-off&lt;/th&gt;
&lt;/tr&gt;
&lt;/thead&gt;
&lt;tbody&gt;
&lt;tr&gt;
&lt;td&gt;Existing transcript&lt;/td&gt;
&lt;td&gt;&lt;code&gt;--from file.srt&lt;/code&gt;&lt;/td&gt;
&lt;td&gt;Free and instant; word timing estimated from segments&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;Groq&lt;/td&gt;
&lt;td&gt;&lt;code&gt;GROQ_API_KEY&lt;/code&gt;&lt;/td&gt;
&lt;td&gt;About $0.04 per audio hour, a minute or two per hour; audio leaves your machine&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;OpenAI&lt;/td&gt;
&lt;td&gt;&lt;code&gt;OPENAI_API_KEY&lt;/code&gt;&lt;/td&gt;
&lt;td&gt;Uses whisper-1, which OpenAI plans to retire in February 2027&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;Local&lt;/td&gt;
&lt;td&gt;&lt;code&gt;pip install faster-whisper&lt;/code&gt;&lt;/td&gt;
&lt;td&gt;Free and private; roughly 0.5x to 1x the video's length on a laptop with &lt;code&gt;small&lt;/code&gt;
&lt;/td&gt;
&lt;/tr&gt;
&lt;/tbody&gt;
&lt;/table&gt;&lt;/div&gt;

&lt;p&gt;Auto mode tries them in that order. Outside ffmpeg, the script uses only the standard library.&lt;/p&gt;

&lt;p&gt;The output is built for an agent's context window. &lt;code&gt;transcript.txt&lt;/code&gt; has one timestamped line per sentence, compact enough to read an hour in one go. &lt;code&gt;transcript.json&lt;/code&gt; holds every word with start and end in seconds, which &lt;code&gt;make_shorts.py&lt;/code&gt; uses for captions and for snapping cuts to word edges.&lt;br&gt;
&lt;/p&gt;

&lt;div class="highlight js-code-highlight"&gt;
&lt;pre class="highlight plaintext"&gt;&lt;code&gt;[03:33.0-03:46.6] I think another piece of it is kind of coming up with a concept of how for a highly automated vehicle, you don't lose sight of kind of the need for the crew to always have a safe path home.
&lt;/code&gt;&lt;/pre&gt;

&lt;/div&gt;



&lt;h2&gt;
  
  
  Step 3: scoring candidates with the clip rubric
&lt;/h2&gt;

&lt;p&gt;The agent reads the whole transcript and lists 8 to 12 candidates. Each passes four hard gates or is dropped: the first sentence needs no context, it holds one complete idea, it contains no intro, outro, sponsor read or crosstalk, and it has no music or footage the user doesn't own. Survivors are scored out of 100:&lt;/p&gt;

&lt;div class="table-wrapper-paragraph"&gt;&lt;table&gt;
&lt;thead&gt;
&lt;tr&gt;
&lt;th&gt;Criterion&lt;/th&gt;
&lt;th&gt;Points&lt;/th&gt;
&lt;/tr&gt;
&lt;/thead&gt;
&lt;tbody&gt;
&lt;tr&gt;
&lt;td&gt;Hook in the first 3 seconds&lt;/td&gt;
&lt;td&gt;25&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;Stands alone&lt;/td&gt;
&lt;td&gt;20&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;Payoff or turn&lt;/td&gt;
&lt;td&gt;20&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;Emotion and energy&lt;/td&gt;
&lt;td&gt;15&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;Specifics&lt;/td&gt;
&lt;td&gt;10&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;Ending&lt;/td&gt;
&lt;td&gt;10&lt;/td&gt;
&lt;/tr&gt;
&lt;/tbody&gt;
&lt;/table&gt;&lt;/div&gt;

&lt;p&gt;70+ ships, 55 to 69 is a maybe, the rest goes. On the NASA example, a moment about training for zero surprises scored 88, "save the crew, not the mission" 82, a line about future crews cursing the designers' names 76, and an abstract stretch opening with "And so" scored 48 and was dropped. The agent shows the table and the user picks.&lt;/p&gt;

&lt;p&gt;Default length is 25 to 45 seconds, 60 at most, 90 only for a story scoring 85+. Clips carrying any third-party audio or footage stay at 60 seconds or less, since YouTube blocks longer Shorts with a Content ID claim.&lt;/p&gt;

&lt;h2&gt;
  
  
  Step 4: clips.json
&lt;/h2&gt;



&lt;div class="highlight js-code-highlight"&gt;
&lt;pre class="highlight json"&gt;&lt;code&gt;&lt;span class="p"&gt;{&lt;/span&gt;&lt;span class="w"&gt;
  &lt;/span&gt;&lt;span class="nl"&gt;"video"&lt;/span&gt;&lt;span class="p"&gt;:&lt;/span&gt;&lt;span class="w"&gt; &lt;/span&gt;&lt;span class="s2"&gt;"talk.mp4"&lt;/span&gt;&lt;span class="p"&gt;,&lt;/span&gt;&lt;span class="w"&gt;
  &lt;/span&gt;&lt;span class="nl"&gt;"style"&lt;/span&gt;&lt;span class="p"&gt;:&lt;/span&gt;&lt;span class="w"&gt; &lt;/span&gt;&lt;span class="p"&gt;{&lt;/span&gt;&lt;span class="nl"&gt;"accent"&lt;/span&gt;&lt;span class="p"&gt;:&lt;/span&gt;&lt;span class="w"&gt; &lt;/span&gt;&lt;span class="s2"&gt;"#FFD43B"&lt;/span&gt;&lt;span class="p"&gt;,&lt;/span&gt;&lt;span class="w"&gt; &lt;/span&gt;&lt;span class="nl"&gt;"words_per_line"&lt;/span&gt;&lt;span class="p"&gt;:&lt;/span&gt;&lt;span class="w"&gt; &lt;/span&gt;&lt;span class="mi"&gt;3&lt;/span&gt;&lt;span class="p"&gt;,&lt;/span&gt;&lt;span class="w"&gt; &lt;/span&gt;&lt;span class="nl"&gt;"uppercase"&lt;/span&gt;&lt;span class="p"&gt;:&lt;/span&gt;&lt;span class="w"&gt; &lt;/span&gt;&lt;span class="kc"&gt;false&lt;/span&gt;&lt;span class="p"&gt;,&lt;/span&gt;&lt;span class="w"&gt; &lt;/span&gt;&lt;span class="nl"&gt;"hook"&lt;/span&gt;&lt;span class="p"&gt;:&lt;/span&gt;&lt;span class="w"&gt; &lt;/span&gt;&lt;span class="kc"&gt;true&lt;/span&gt;&lt;span class="p"&gt;},&lt;/span&gt;&lt;span class="w"&gt;
  &lt;/span&gt;&lt;span class="nl"&gt;"clips"&lt;/span&gt;&lt;span class="p"&gt;:&lt;/span&gt;&lt;span class="w"&gt; &lt;/span&gt;&lt;span class="p"&gt;[&lt;/span&gt;&lt;span class="w"&gt;
    &lt;/span&gt;&lt;span class="p"&gt;{&lt;/span&gt;&lt;span class="nl"&gt;"id"&lt;/span&gt;&lt;span class="p"&gt;:&lt;/span&gt;&lt;span class="w"&gt; &lt;/span&gt;&lt;span class="s2"&gt;"01"&lt;/span&gt;&lt;span class="p"&gt;,&lt;/span&gt;&lt;span class="w"&gt; &lt;/span&gt;&lt;span class="nl"&gt;"start"&lt;/span&gt;&lt;span class="p"&gt;:&lt;/span&gt;&lt;span class="w"&gt; &lt;/span&gt;&lt;span class="mf"&gt;754.2&lt;/span&gt;&lt;span class="p"&gt;,&lt;/span&gt;&lt;span class="w"&gt; &lt;/span&gt;&lt;span class="nl"&gt;"end"&lt;/span&gt;&lt;span class="p"&gt;:&lt;/span&gt;&lt;span class="w"&gt; &lt;/span&gt;&lt;span class="mf"&gt;801.9&lt;/span&gt;&lt;span class="p"&gt;,&lt;/span&gt;&lt;span class="w"&gt;
     &lt;/span&gt;&lt;span class="nl"&gt;"hook"&lt;/span&gt;&lt;span class="p"&gt;:&lt;/span&gt;&lt;span class="w"&gt; &lt;/span&gt;&lt;span class="s2"&gt;"Text shown on screen for the first seconds"&lt;/span&gt;&lt;span class="p"&gt;,&lt;/span&gt;&lt;span class="w"&gt;
     &lt;/span&gt;&lt;span class="nl"&gt;"title"&lt;/span&gt;&lt;span class="p"&gt;:&lt;/span&gt;&lt;span class="w"&gt; &lt;/span&gt;&lt;span class="s2"&gt;"Title for YouTube Shorts"&lt;/span&gt;&lt;span class="p"&gt;,&lt;/span&gt;&lt;span class="w"&gt; &lt;/span&gt;&lt;span class="nl"&gt;"caption"&lt;/span&gt;&lt;span class="p"&gt;:&lt;/span&gt;&lt;span class="w"&gt; &lt;/span&gt;&lt;span class="s2"&gt;"Caption for TikTok and Reels"&lt;/span&gt;&lt;span class="p"&gt;,&lt;/span&gt;&lt;span class="w"&gt;
     &lt;/span&gt;&lt;span class="nl"&gt;"reframe"&lt;/span&gt;&lt;span class="p"&gt;:&lt;/span&gt;&lt;span class="w"&gt; &lt;/span&gt;&lt;span class="s2"&gt;"auto"&lt;/span&gt;&lt;span class="p"&gt;,&lt;/span&gt;&lt;span class="w"&gt; &lt;/span&gt;&lt;span class="nl"&gt;"focus_x"&lt;/span&gt;&lt;span class="p"&gt;:&lt;/span&gt;&lt;span class="w"&gt; &lt;/span&gt;&lt;span class="mf"&gt;0.5&lt;/span&gt;&lt;span class="p"&gt;,&lt;/span&gt;&lt;span class="w"&gt; &lt;/span&gt;&lt;span class="nl"&gt;"tighten"&lt;/span&gt;&lt;span class="p"&gt;:&lt;/span&gt;&lt;span class="w"&gt; &lt;/span&gt;&lt;span class="kc"&gt;true&lt;/span&gt;&lt;span class="p"&gt;}&lt;/span&gt;&lt;span class="w"&gt;
  &lt;/span&gt;&lt;span class="p"&gt;]&lt;/span&gt;&lt;span class="w"&gt;
&lt;/span&gt;&lt;span class="p"&gt;}&lt;/span&gt;&lt;span class="w"&gt;
&lt;/span&gt;&lt;/code&gt;&lt;/pre&gt;

&lt;/div&gt;



&lt;p&gt;&lt;code&gt;reframe&lt;/code&gt; takes &lt;code&gt;auto&lt;/code&gt; (face crop when OpenCV finds one speaker, otherwise fit), &lt;code&gt;face&lt;/code&gt;, &lt;code&gt;center&lt;/code&gt; or &lt;code&gt;fit&lt;/code&gt;. &lt;code&gt;focus_x&lt;/code&gt; from 0 to 1 places the crop by hand. &lt;code&gt;tighten&lt;/code&gt; removes pauses over 0.45 seconds.&lt;/p&gt;

&lt;h2&gt;
  
  
  Step 5: preview, then render
&lt;/h2&gt;



&lt;div class="highlight js-code-highlight"&gt;
&lt;pre class="highlight shell"&gt;&lt;code&gt;python3 scripts/make_shorts.py clips.json &lt;span class="nt"&gt;--transcript&lt;/span&gt; work/transcript.json &lt;span class="nt"&gt;--out&lt;/span&gt; shorts/ &lt;span class="nt"&gt;--preview&lt;/span&gt;
python3 scripts/make_shorts.py clips.json &lt;span class="nt"&gt;--transcript&lt;/span&gt; work/transcript.json &lt;span class="nt"&gt;--out&lt;/span&gt; shorts/
&lt;/code&gt;&lt;/pre&gt;

&lt;/div&gt;



&lt;p&gt;The preview writes a &lt;code&gt;preview-XX.jpg&lt;/code&gt; per clip with three frames and the planned crop box. The agent opens each one; if the box misses the speaker it sets &lt;code&gt;focus_x&lt;/code&gt;, and for two people far apart or a screen share it switches to &lt;code&gt;fit&lt;/code&gt;, which scales the whole frame onto a blurred, darkened copy of itself.&lt;/p&gt;

&lt;p&gt;The render produces 1080x1920 at the source frame rate, H.264 High with AAC 48 kHz, moov at the front. Audio gets an 80 Hz high-pass, loudness normalisation to -14 LUFS with a -1 dB true-peak ceiling, and 10 ms fades at every cut so removed pauses don't click. Pauses the transcriber hides inside a stretched word are removed too, along with stutters, "um", "uh" and "you know,", and every second segment after a cut punches in slightly so jump cuts look intended. Rendering takes about as long as the clips themselves; &lt;code&gt;--best&lt;/code&gt; is slower and slightly smaller, and &lt;code&gt;--only 02&lt;/code&gt; re-renders one clip.&lt;/p&gt;

&lt;h2&gt;
  
  
  Why captions are drawn with Pillow
&lt;/h2&gt;

&lt;p&gt;Burning text with ffmpeg usually means the &lt;code&gt;subtitles&lt;/code&gt; or &lt;code&gt;drawtext&lt;/code&gt; filter, and not every ffmpeg package ships with the libraries those need. Drawing the caption frames with Pillow and overlaying them keeps the requirement at "any ffmpeg build", which is what makes a one-line install hold on Mac, Windows and Linux. A GitHub Actions matrix renders a crop clip and a fit clip on all three and asserts the output is 1080x1920.&lt;/p&gt;

&lt;p&gt;The captions themselves: Poppins ExtraBold (bundled, SIL Open Font License) at a fixed 96 px, white with an 8 px black outline, up to three words at a time on one line, the spoken word in the accent colour; groups break at punctuation and pauses and never end on a small word like "the". Everything is centred on the frame and kept inside x 180 to 900 and y 290 to 1250, the area that TikTok, Reels and Shorts leave free of buttons, captions and tabs. The hook shows for the first 3.5 seconds, black on a white rounded box, and returns for 2 seconds halfway through clips over 40 seconds.&lt;/p&gt;

&lt;h2&gt;
  
  
  Step 6: the review loop
&lt;/h2&gt;

&lt;p&gt;After rendering, the agent watches the first 5 seconds of every clip and opens &lt;code&gt;contact-sheet.jpg&lt;/code&gt;. It re-renders when a clip opens mid-thought, a caption covers the face, the hook and the first spoken words repeat each other, or the ending trails off. &lt;code&gt;manifest.json&lt;/code&gt; lists byte sizes, durations, texts and warnings; every warning gets resolved.&lt;/p&gt;

&lt;h2&gt;
  
  
  Step 7: posting through MCP
&lt;/h2&gt;

&lt;p&gt;Posting runs through &lt;a href="https://woopsocial.com/?utm_source=video-to-shorts-skill&amp;amp;utm_medium=devto" rel="noopener noreferrer"&gt;WoopSocial&lt;/a&gt;, a social media API and MCP server for AI agents. Disclosure: the skill is maintained by Frank Heijdenrijk and the team behind WoopSocial. The free plan covers 2 social accounts with everything included; connection steps per agent are at &lt;a href="https://woopsocial.com/agents?utm_source=video-to-shorts-skill&amp;amp;utm_medium=devto" rel="noopener noreferrer"&gt;woopsocial.com/agents&lt;/a&gt;.&lt;/p&gt;

&lt;p&gt;The flow per clip: &lt;code&gt;media_uploads_create_session&lt;/code&gt; with the exact byte size from the manifest, &lt;code&gt;scripts/upload_parts.py&lt;/code&gt; to PUT each part to its presigned URL, &lt;code&gt;media_uploads_complete_session&lt;/code&gt; for the media ID. Then one post per clip per platform (TikTok &lt;code&gt;VIDEO&lt;/code&gt;, Instagram &lt;code&gt;REEL&lt;/code&gt;, YouTube with &lt;code&gt;title&lt;/code&gt; and &lt;code&gt;privacy&lt;/code&gt;, Facebook &lt;code&gt;REEL&lt;/code&gt;), each run through &lt;code&gt;posts_validate&lt;/code&gt;. The agent shows one table of clip, platform, account, text and time with timezone, and calls &lt;code&gt;posts_create&lt;/code&gt; only after a clear yes. With several accounts it varies clips or at least hooks and captions per account and staggers the times.&lt;/p&gt;

&lt;p&gt;WoopSocial doesn't write captions or crop media; the skill writes the text and renders the vertical file before anything is uploaded. Skip this step entirely and you still get finished MP4s plus titles and captions in the manifest.&lt;/p&gt;

&lt;h2&gt;
  
  
  Reading results back
&lt;/h2&gt;

&lt;p&gt;Three to seven days later, the agent calls &lt;code&gt;social_account_posts_list&lt;/code&gt; and logs views, likes, comments, shares and saves (where a platform reports them, with the collection date) in &lt;code&gt;clips-log.md&lt;/code&gt; next to the source timestamps and hook. Missing fields are recorded as not reported, never as zero. The next batch is pulled from the parts of the video that beat the account's usual numbers.&lt;/p&gt;

&lt;h2&gt;
  
  
  Guardrails in SKILL.md
&lt;/h2&gt;

&lt;ul&gt;
&lt;li&gt;Only footage the user owns or has permission to clip; otherwise the agent stops.&lt;/li&gt;
&lt;li&gt;No identical clip posted at the same time to many accounts.&lt;/li&gt;
&lt;li&gt;Hooks, titles and captions say only what the clip says.&lt;/li&gt;
&lt;li&gt;Nothing posts without a yes.&lt;/li&gt;
&lt;/ul&gt;

&lt;h2&gt;
  
  
  Try it
&lt;/h2&gt;



&lt;div class="highlight js-code-highlight"&gt;
&lt;pre class="highlight shell"&gt;&lt;code&gt;npx skills add social-media-skills/video-to-shorts-skill &lt;span class="nt"&gt;-g&lt;/span&gt; &lt;span class="nt"&gt;-a&lt;/span&gt; claude-code &lt;span class="nt"&gt;-y&lt;/span&gt;
&lt;/code&gt;&lt;/pre&gt;

&lt;/div&gt;



&lt;p&gt;Then ask your agent: "Turn talk.mp4 into three shorts for TikTok and YouTube." The worked example with transcript, &lt;code&gt;clips.json&lt;/code&gt;, contact sheet and manifest is in &lt;code&gt;examples/nasa-interview/&lt;/code&gt;. Issues and pull requests are welcome on &lt;a href="https://github.com/social-media-skills/video-to-shorts-skill" rel="noopener noreferrer"&gt;GitHub&lt;/a&gt;, and more free skills are at &lt;a href="https://social-media-skills.com" rel="noopener noreferrer"&gt;social-media-skills.com&lt;/a&gt;.&lt;/p&gt;

</description>
      <category>showdev</category>
      <category>claudecode</category>
      <category>opensource</category>
      <category>python</category>
    </item>
    <item>
      <title>Instagram Carousel Skill for Claude Code: Write, Render and Post Carousels From Your Agent</title>
      <dc:creator>Frank Heijdenrijk</dc:creator>
      <pubDate>Wed, 28 Oct 2026 09:31:35 +0000</pubDate>
      <link>https://dev.to/fheijdenrijk/instagram-carousel-skill-for-claude-code-write-render-and-post-carousels-from-your-agent-1odp</link>
      <guid>https://dev.to/fheijdenrijk/instagram-carousel-skill-for-claude-code-write-render-and-post-carousels-from-your-agent-1odp</guid>
      <description>&lt;p&gt;The Instagram carousel skill is a free, MIT-licensed agent skill that lets Claude Code, Claude, Codex, Cursor and OpenClaw plan, write, render and post Instagram carousels. It picks a format, writes the cover and the slides, renders 1080x1350 JPEGs on your machine with a small Python script, writes alt text and a caption, and posts the carousel once you say yes. No headless browser, no image API, no credits.&lt;/p&gt;

&lt;p&gt;Repo: &lt;a href="https://github.com/social-media-skills/instagram-carousel-skill" rel="noopener noreferrer"&gt;github.com/social-media-skills/instagram-carousel-skill&lt;/a&gt;&lt;/p&gt;

&lt;p&gt;&lt;a href="https://media2.dev.to/dynamic/image/width=800%2Cheight=%2Cfit=scale-down%2Cgravity=auto%2Cformat=auto/https%3A%2F%2Fdev-to-uploads.s3.us-east-2.amazonaws.com%2Fuploads%2Farticles%2Fjg6ya43n8d7j3s1fh56q.png" class="article-body-image-wrapper"&gt;&lt;img src="https://media2.dev.to/dynamic/image/width=800%2Cheight=%2Cfit=scale-down%2Cgravity=auto%2Cformat=auto/https%3A%2F%2Fdev-to-uploads.s3.us-east-2.amazonaws.com%2Fuploads%2Farticles%2Fjg6ya43n8d7j3s1fh56q.png" alt="Six slides rendered by the Instagram carousel skill: a serif cover, a numbered rule, a myth vs fact slide, a recap list, an app screenshot slide and a tweet-style card" width="800" height="662"&gt;&lt;/a&gt;&lt;/p&gt;

&lt;p&gt;&lt;em&gt;Slides from the two example carousels in the repo. The brand "plodd" and its app screenshot are made up.&lt;/em&gt;&lt;/p&gt;

&lt;h2&gt;
  
  
  What the skill fixes
&lt;/h2&gt;

&lt;p&gt;Most carousels that don't get saved fail in the same places, and each one maps to a rule in the skill:&lt;/p&gt;

&lt;ul&gt;
&lt;li&gt;
&lt;strong&gt;The cover names a topic&lt;/strong&gt; ("Habit tips") instead of a result or a tension. The skill writes 10 covers of 8 words or fewer and you pick one.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Slide 2 only works after slide 1.&lt;/strong&gt; Instagram may show a carousel again, starting at slide 2, to people who scrolled past it. The skill writes slide 2 so it works as a cover on its own.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Too many words.&lt;/strong&gt; One idea per slide, headlines of 8 words or fewer, body text of 30 words or fewer.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Template chrome.&lt;/strong&gt; No progress bars, no slide numbers, no logo on every slide. A small name and handle in the corner, and a small arrow on the cover only.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;A last slide with three asks.&lt;/strong&gt; One ask, matched to the goal: send, save, comment or link.&lt;/li&gt;
&lt;/ul&gt;

&lt;h2&gt;
  
  
  Install the Instagram carousel skill
&lt;/h2&gt;



&lt;div class="highlight js-code-highlight"&gt;
&lt;pre class="highlight shell"&gt;&lt;code&gt;npx skills add social-media-skills/instagram-carousel-skill &lt;span class="nt"&gt;-g&lt;/span&gt; &lt;span class="nt"&gt;-a&lt;/span&gt; claude-code &lt;span class="nt"&gt;-y&lt;/span&gt;
&lt;/code&gt;&lt;/pre&gt;

&lt;/div&gt;



&lt;p&gt;Other agents:&lt;/p&gt;

&lt;ul&gt;
&lt;li&gt;
&lt;strong&gt;OpenClaw (ClawHub):&lt;/strong&gt; &lt;code&gt;openclaw skills install @social-media-skills/instagram-carousel-skill&lt;/code&gt;
&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;GitHub CLI:&lt;/strong&gt; &lt;code&gt;gh skill install social-media-skills/instagram-carousel-skill instagram-carousel-skill&lt;/code&gt;
&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Gemini CLI:&lt;/strong&gt; &lt;code&gt;gemini extensions install https://github.com/social-media-skills/instagram-carousel-skill&lt;/code&gt;
&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Codex, Cursor and others:&lt;/strong&gt; &lt;code&gt;npx skills add social-media-skills/instagram-carousel-skill&lt;/code&gt; and pick your agent.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Claude (claude.ai):&lt;/strong&gt; download the zip from the latest GitHub release and upload it under Customize &amp;gt; Skills.&lt;/li&gt;
&lt;/ul&gt;

&lt;p&gt;Rendering needs Python 3.8+ and Pillow (&lt;code&gt;pip install pillow&lt;/code&gt;). The fonts ship in the repo.&lt;/p&gt;

&lt;h2&gt;
  
  
  What happens when you ask for a carousel
&lt;/h2&gt;

&lt;p&gt;A prompt like "Make an Instagram carousel for my habit tracker about not giving up after a missed day" starts a fixed sequence:&lt;/p&gt;

&lt;ol&gt;
&lt;li&gt;
&lt;strong&gt;Context.&lt;/strong&gt; The agent reads brand notes in the workspace if they exist, or asks once: who the account is for, whether this carousel should get saved, get sent or bring people to a link, what you know that your audience doesn't, which screenshots or photos you have, and what must not be said.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Format and cover.&lt;/strong&gt; It picks a format for the goal and writes 10 covers. You choose one; the others go into &lt;code&gt;covers.md&lt;/code&gt; for later tests.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Slides.&lt;/strong&gt; One idea per slide, the product shown once near the end with a real screenshot, a last slide with one ask, a keyword-first caption with at most 5 hashtags, and one sentence of alt text per slide.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Honesty check.&lt;/strong&gt; Every claim must be something you did, saw or can show. The agent never writes your first-person stories, user counts or results for you.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Render and review.&lt;/strong&gt; It writes &lt;code&gt;carousel.json&lt;/code&gt;, runs the renderer, opens every slide and the contact sheet, and fixes each warning.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Post.&lt;/strong&gt; It shows you the slides, caption, account and time, and posts only after a clear yes.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Read back.&lt;/strong&gt; A few days later it pulls the post's numbers and logs them next to the format and cover.&lt;/li&gt;
&lt;/ol&gt;

&lt;h2&gt;
  
  
  The renderer
&lt;/h2&gt;

&lt;p&gt;&lt;code&gt;scripts/render_carousel.py&lt;/code&gt; turns a JSON spec into slides. It uses Pillow only, renders at 2x and downscales for sharp text, and bundles Inter and Instrument Serif under the SIL Open Font License. This is a shortened version of the spec behind the editorial example:&lt;br&gt;
&lt;/p&gt;

&lt;div class="highlight js-code-highlight"&gt;
&lt;pre class="highlight json"&gt;&lt;code&gt;&lt;span class="p"&gt;{&lt;/span&gt;&lt;span class="w"&gt;
  &lt;/span&gt;&lt;span class="nl"&gt;"theme"&lt;/span&gt;&lt;span class="p"&gt;:&lt;/span&gt;&lt;span class="w"&gt; &lt;/span&gt;&lt;span class="p"&gt;{&lt;/span&gt;&lt;span class="nl"&gt;"preset"&lt;/span&gt;&lt;span class="p"&gt;:&lt;/span&gt;&lt;span class="w"&gt; &lt;/span&gt;&lt;span class="s2"&gt;"editorial"&lt;/span&gt;&lt;span class="p"&gt;,&lt;/span&gt;&lt;span class="w"&gt; &lt;/span&gt;&lt;span class="nl"&gt;"mode"&lt;/span&gt;&lt;span class="p"&gt;:&lt;/span&gt;&lt;span class="w"&gt; &lt;/span&gt;&lt;span class="s2"&gt;"light"&lt;/span&gt;&lt;span class="p"&gt;,&lt;/span&gt;&lt;span class="w"&gt; &lt;/span&gt;&lt;span class="nl"&gt;"headline_font"&lt;/span&gt;&lt;span class="p"&gt;:&lt;/span&gt;&lt;span class="w"&gt; &lt;/span&gt;&lt;span class="s2"&gt;"serif"&lt;/span&gt;&lt;span class="p"&gt;,&lt;/span&gt;&lt;span class="w"&gt;
            &lt;/span&gt;&lt;span class="nl"&gt;"name"&lt;/span&gt;&lt;span class="p"&gt;:&lt;/span&gt;&lt;span class="w"&gt; &lt;/span&gt;&lt;span class="s2"&gt;"plodd"&lt;/span&gt;&lt;span class="p"&gt;,&lt;/span&gt;&lt;span class="w"&gt; &lt;/span&gt;&lt;span class="nl"&gt;"handle"&lt;/span&gt;&lt;span class="p"&gt;:&lt;/span&gt;&lt;span class="w"&gt; &lt;/span&gt;&lt;span class="s2"&gt;"@plodd.app"&lt;/span&gt;&lt;span class="p"&gt;,&lt;/span&gt;&lt;span class="w"&gt; &lt;/span&gt;&lt;span class="nl"&gt;"avatar"&lt;/span&gt;&lt;span class="p"&gt;:&lt;/span&gt;&lt;span class="w"&gt; &lt;/span&gt;&lt;span class="s2"&gt;"avatar.png"&lt;/span&gt;&lt;span class="p"&gt;},&lt;/span&gt;&lt;span class="w"&gt;
  &lt;/span&gt;&lt;span class="nl"&gt;"slides"&lt;/span&gt;&lt;span class="p"&gt;:&lt;/span&gt;&lt;span class="w"&gt; &lt;/span&gt;&lt;span class="p"&gt;[&lt;/span&gt;&lt;span class="w"&gt;
    &lt;/span&gt;&lt;span class="p"&gt;{&lt;/span&gt;&lt;span class="nl"&gt;"type"&lt;/span&gt;&lt;span class="p"&gt;:&lt;/span&gt;&lt;span class="w"&gt; &lt;/span&gt;&lt;span class="s2"&gt;"cover"&lt;/span&gt;&lt;span class="p"&gt;,&lt;/span&gt;&lt;span class="w"&gt; &lt;/span&gt;&lt;span class="nl"&gt;"title"&lt;/span&gt;&lt;span class="p"&gt;:&lt;/span&gt;&lt;span class="w"&gt; &lt;/span&gt;&lt;span class="s2"&gt;"Missed a day?&lt;/span&gt;&lt;span class="se"&gt;\n&lt;/span&gt;&lt;span class="s2"&gt;Your habit is *fine*."&lt;/span&gt;&lt;span class="p"&gt;,&lt;/span&gt;&lt;span class="w"&gt;
     &lt;/span&gt;&lt;span class="nl"&gt;"kicker"&lt;/span&gt;&lt;span class="p"&gt;:&lt;/span&gt;&lt;span class="w"&gt; &lt;/span&gt;&lt;span class="s2"&gt;"4 rules for people who restart every Monday"&lt;/span&gt;&lt;span class="p"&gt;,&lt;/span&gt;&lt;span class="w"&gt; &lt;/span&gt;&lt;span class="nl"&gt;"alt"&lt;/span&gt;&lt;span class="p"&gt;:&lt;/span&gt;&lt;span class="w"&gt; &lt;/span&gt;&lt;span class="s2"&gt;"..."&lt;/span&gt;&lt;span class="p"&gt;},&lt;/span&gt;&lt;span class="w"&gt;
    &lt;/span&gt;&lt;span class="p"&gt;{&lt;/span&gt;&lt;span class="nl"&gt;"type"&lt;/span&gt;&lt;span class="p"&gt;:&lt;/span&gt;&lt;span class="w"&gt; &lt;/span&gt;&lt;span class="s2"&gt;"step"&lt;/span&gt;&lt;span class="p"&gt;,&lt;/span&gt;&lt;span class="w"&gt; &lt;/span&gt;&lt;span class="nl"&gt;"number"&lt;/span&gt;&lt;span class="p"&gt;:&lt;/span&gt;&lt;span class="w"&gt; &lt;/span&gt;&lt;span class="s2"&gt;"1"&lt;/span&gt;&lt;span class="p"&gt;,&lt;/span&gt;&lt;span class="w"&gt; &lt;/span&gt;&lt;span class="nl"&gt;"title"&lt;/span&gt;&lt;span class="p"&gt;:&lt;/span&gt;&lt;span class="w"&gt; &lt;/span&gt;&lt;span class="s2"&gt;"One missed day is noise. *Two* is a pattern."&lt;/span&gt;&lt;span class="p"&gt;,&lt;/span&gt;&lt;span class="w"&gt;
     &lt;/span&gt;&lt;span class="nl"&gt;"body"&lt;/span&gt;&lt;span class="p"&gt;:&lt;/span&gt;&lt;span class="w"&gt; &lt;/span&gt;&lt;span class="s2"&gt;"Protect one line: never miss twice in a row."&lt;/span&gt;&lt;span class="p"&gt;,&lt;/span&gt;&lt;span class="w"&gt; &lt;/span&gt;&lt;span class="nl"&gt;"alt"&lt;/span&gt;&lt;span class="p"&gt;:&lt;/span&gt;&lt;span class="w"&gt; &lt;/span&gt;&lt;span class="s2"&gt;"..."&lt;/span&gt;&lt;span class="p"&gt;},&lt;/span&gt;&lt;span class="w"&gt;
    &lt;/span&gt;&lt;span class="p"&gt;{&lt;/span&gt;&lt;span class="nl"&gt;"type"&lt;/span&gt;&lt;span class="p"&gt;:&lt;/span&gt;&lt;span class="w"&gt; &lt;/span&gt;&lt;span class="s2"&gt;"myth"&lt;/span&gt;&lt;span class="p"&gt;,&lt;/span&gt;&lt;span class="w"&gt; &lt;/span&gt;&lt;span class="nl"&gt;"myth"&lt;/span&gt;&lt;span class="p"&gt;:&lt;/span&gt;&lt;span class="w"&gt; &lt;/span&gt;&lt;span class="s2"&gt;"A habit takes 21 days"&lt;/span&gt;&lt;span class="p"&gt;,&lt;/span&gt;&lt;span class="w"&gt; &lt;/span&gt;&lt;span class="nl"&gt;"fact"&lt;/span&gt;&lt;span class="p"&gt;:&lt;/span&gt;&lt;span class="w"&gt; &lt;/span&gt;&lt;span class="s2"&gt;"Closer to *66 days*."&lt;/span&gt;&lt;span class="p"&gt;,&lt;/span&gt;&lt;span class="w"&gt;
     &lt;/span&gt;&lt;span class="nl"&gt;"source"&lt;/span&gt;&lt;span class="p"&gt;:&lt;/span&gt;&lt;span class="w"&gt; &lt;/span&gt;&lt;span class="s2"&gt;"Lally et al., European Journal of Social Psychology, 2010"&lt;/span&gt;&lt;span class="p"&gt;,&lt;/span&gt;&lt;span class="w"&gt; &lt;/span&gt;&lt;span class="nl"&gt;"alt"&lt;/span&gt;&lt;span class="p"&gt;:&lt;/span&gt;&lt;span class="w"&gt; &lt;/span&gt;&lt;span class="s2"&gt;"..."&lt;/span&gt;&lt;span class="p"&gt;},&lt;/span&gt;&lt;span class="w"&gt;
    &lt;/span&gt;&lt;span class="p"&gt;{&lt;/span&gt;&lt;span class="nl"&gt;"type"&lt;/span&gt;&lt;span class="p"&gt;:&lt;/span&gt;&lt;span class="w"&gt; &lt;/span&gt;&lt;span class="s2"&gt;"screenshot"&lt;/span&gt;&lt;span class="p"&gt;,&lt;/span&gt;&lt;span class="w"&gt; &lt;/span&gt;&lt;span class="nl"&gt;"title"&lt;/span&gt;&lt;span class="p"&gt;:&lt;/span&gt;&lt;span class="w"&gt; &lt;/span&gt;&lt;span class="s2"&gt;"plodd counts your *week*"&lt;/span&gt;&lt;span class="p"&gt;,&lt;/span&gt;&lt;span class="w"&gt;
     &lt;/span&gt;&lt;span class="nl"&gt;"body"&lt;/span&gt;&lt;span class="p"&gt;:&lt;/span&gt;&lt;span class="w"&gt; &lt;/span&gt;&lt;span class="s2"&gt;"Miss Wednesday. Still a good week."&lt;/span&gt;&lt;span class="p"&gt;,&lt;/span&gt;&lt;span class="w"&gt; &lt;/span&gt;&lt;span class="nl"&gt;"image"&lt;/span&gt;&lt;span class="p"&gt;:&lt;/span&gt;&lt;span class="w"&gt; &lt;/span&gt;&lt;span class="s2"&gt;"screens/plodd-week.png"&lt;/span&gt;&lt;span class="p"&gt;,&lt;/span&gt;&lt;span class="w"&gt; &lt;/span&gt;&lt;span class="nl"&gt;"alt"&lt;/span&gt;&lt;span class="p"&gt;:&lt;/span&gt;&lt;span class="w"&gt; &lt;/span&gt;&lt;span class="s2"&gt;"..."&lt;/span&gt;&lt;span class="p"&gt;},&lt;/span&gt;&lt;span class="w"&gt;
    &lt;/span&gt;&lt;span class="p"&gt;{&lt;/span&gt;&lt;span class="nl"&gt;"type"&lt;/span&gt;&lt;span class="p"&gt;:&lt;/span&gt;&lt;span class="w"&gt; &lt;/span&gt;&lt;span class="s2"&gt;"cta"&lt;/span&gt;&lt;span class="p"&gt;,&lt;/span&gt;&lt;span class="w"&gt; &lt;/span&gt;&lt;span class="nl"&gt;"title"&lt;/span&gt;&lt;span class="p"&gt;:&lt;/span&gt;&lt;span class="w"&gt; &lt;/span&gt;&lt;span class="s2"&gt;"Send this to the friend who restarts every Monday."&lt;/span&gt;&lt;span class="p"&gt;,&lt;/span&gt;&lt;span class="w"&gt; &lt;/span&gt;&lt;span class="nl"&gt;"alt"&lt;/span&gt;&lt;span class="p"&gt;:&lt;/span&gt;&lt;span class="w"&gt; &lt;/span&gt;&lt;span class="s2"&gt;"..."&lt;/span&gt;&lt;span class="p"&gt;}&lt;/span&gt;&lt;span class="w"&gt;
  &lt;/span&gt;&lt;span class="p"&gt;]&lt;/span&gt;&lt;span class="w"&gt;
&lt;/span&gt;&lt;span class="p"&gt;}&lt;/span&gt;&lt;span class="w"&gt;
&lt;/span&gt;&lt;/code&gt;&lt;/pre&gt;

&lt;/div&gt;





&lt;div class="highlight js-code-highlight"&gt;
&lt;pre class="highlight shell"&gt;&lt;code&gt;python3 scripts/render_carousel.py carousel.json &lt;span class="nt"&gt;--out&lt;/span&gt; carousel/
&lt;/code&gt;&lt;/pre&gt;

&lt;/div&gt;



&lt;p&gt;&lt;code&gt;*word*&lt;/code&gt; sets a word in the accent colour, &lt;code&gt;~~words~~&lt;/code&gt; strikes out a myth, and &lt;code&gt;\n&lt;/code&gt; places a line break where the headline reads best.&lt;/p&gt;

&lt;p&gt;&lt;strong&gt;Slide types:&lt;/strong&gt; &lt;code&gt;cover&lt;/code&gt; (text, photo or screenshot), &lt;code&gt;text&lt;/code&gt;, &lt;code&gt;step&lt;/code&gt;, &lt;code&gt;list&lt;/code&gt; (recap or checklist), &lt;code&gt;myth&lt;/code&gt;, &lt;code&gt;stat&lt;/code&gt; (needs a &lt;code&gt;source&lt;/code&gt;), &lt;code&gt;tweet&lt;/code&gt;, &lt;code&gt;photo&lt;/code&gt;, &lt;code&gt;screenshot&lt;/code&gt;, &lt;code&gt;beforeafter&lt;/code&gt;, &lt;code&gt;cta&lt;/code&gt;.&lt;/p&gt;

&lt;p&gt;&lt;strong&gt;Presets:&lt;/strong&gt;&lt;/p&gt;

&lt;div class="table-wrapper-paragraph"&gt;&lt;table&gt;
&lt;thead&gt;
&lt;tr&gt;
&lt;th&gt;Preset&lt;/th&gt;
&lt;th&gt;Look&lt;/th&gt;
&lt;th&gt;Good for&lt;/th&gt;
&lt;/tr&gt;
&lt;/thead&gt;
&lt;tbody&gt;
&lt;tr&gt;
&lt;td&gt;
&lt;code&gt;editorial&lt;/code&gt; light or dark&lt;/td&gt;
&lt;td&gt;Paper or near-black background, serif or heavy sans headlines, one accent&lt;/td&gt;
&lt;td&gt;Step-by-step, frameworks, lists&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;
&lt;code&gt;notes&lt;/code&gt; light or dark&lt;/td&gt;
&lt;td&gt;Plain phone-notes look&lt;/td&gt;
&lt;td&gt;Honest lists, lessons, lo-fi posts&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;
&lt;code&gt;tweet&lt;/code&gt; light or dark&lt;/td&gt;
&lt;td&gt;Tweet-card thread with your own name, handle and avatar&lt;/td&gt;
&lt;td&gt;Opinions, short lessons&lt;/td&gt;
&lt;/tr&gt;
&lt;/tbody&gt;
&lt;/table&gt;&lt;/div&gt;

&lt;p&gt;&lt;strong&gt;Output:&lt;/strong&gt; &lt;code&gt;slide-01.jpg&lt;/code&gt; and onward at 1080x1350 sRGB JPEG, plus:&lt;/p&gt;

&lt;ul&gt;
&lt;li&gt;
&lt;code&gt;contact-sheet.png&lt;/code&gt;: every slide on one image, for review.&lt;/li&gt;
&lt;li&gt;
&lt;code&gt;cover-grid.png&lt;/code&gt;: the cover cropped the way the 3:4 profile grid shows it.&lt;/li&gt;
&lt;li&gt;
&lt;code&gt;manifest.json&lt;/code&gt;: byte size, alt text and text per slide, plus warnings.&lt;/li&gt;
&lt;/ul&gt;

&lt;p&gt;The warnings are the part that saves review time: text below the size floor, too many words on a slide, a stat with no source, a missing alt text, and (with &lt;code&gt;opencv-python&lt;/code&gt; installed) text sitting on a face. When text doesn't fit at the floor, the fix is to split the slide, not to shrink it.&lt;/p&gt;

&lt;h3&gt;
  
  
  Why 1080x1350 and JPEG
&lt;/h3&gt;

&lt;p&gt;Meta's content publishing API accepts JPEG only, with an aspect ratio between 4:5 and 1.91:1, and crops every carousel item to the ratio of the first one. So 4:5 at 1080x1350 is the tallest size that posts through the API, and every slide is rendered at exactly that size. The Instagram app has accepted 3:4 (1080x1440) since May 2025 but the API still rejects it; set &lt;code&gt;"ratio": "3:4"&lt;/code&gt; only for slides you will upload by hand.&lt;/p&gt;

&lt;p&gt;The type scale is set for phone reading, not for a desktop preview: a 1080 px slide shows at roughly a third of its width on a phone, so body text sits at 54 px with a floor of 40, and slide headlines at 128 px serif or 96 px sans.&lt;/p&gt;

&lt;p&gt;The renderer has no agent dependency. You can call it from a script, a CI job or a cron task with your own spec.&lt;/p&gt;

&lt;h2&gt;
  
  
  Posting through WoopSocial
&lt;/h2&gt;

&lt;p&gt;Posting goes through the &lt;a href="https://woopsocial.com/?utm_source=instagram-carousel-skill&amp;amp;utm_medium=devto" rel="noopener noreferrer"&gt;WoopSocial&lt;/a&gt; MCP server, a social media API for AI agents. The skill is maintained by Frank Heijdenrijk and the team behind WoopSocial. The free plan covers 2 social accounts with posting, scheduling and analytics included. Setup for each agent is at &lt;a href="https://woopsocial.com/agents?utm_source=instagram-carousel-skill&amp;amp;utm_medium=devto" rel="noopener noreferrer"&gt;woopsocial.com/agents&lt;/a&gt;.&lt;/p&gt;

&lt;p&gt;The posting step, in tool calls:&lt;/p&gt;

&lt;ol&gt;
&lt;li&gt;
&lt;code&gt;social_accounts_list&lt;/code&gt; to check the connection. If Instagram isn't connected, &lt;code&gt;oauth_create_authorization&lt;/code&gt; returns a link for you to approve.&lt;/li&gt;
&lt;li&gt;For each slide, &lt;code&gt;media_uploads_create_session&lt;/code&gt; with the exact byte size from &lt;code&gt;manifest.json&lt;/code&gt;, then &lt;code&gt;python3 scripts/upload_parts.py session.json carousel/slide-01.jpg&lt;/code&gt;, then &lt;code&gt;media_uploads_complete_session&lt;/code&gt;. Media IDs stay in slide order.&lt;/li&gt;
&lt;li&gt;One post with all media IDs, &lt;code&gt;postType: POST&lt;/code&gt;, the caption, and &lt;code&gt;PUBLISH_NOW&lt;/code&gt;, &lt;code&gt;SCHEDULE_FOR_LATER&lt;/code&gt; or &lt;code&gt;DRAFT&lt;/code&gt;.&lt;/li&gt;
&lt;li&gt;
&lt;code&gt;posts_validate&lt;/code&gt;, then the confirmation: contact sheet, caption, account and time with timezone. &lt;code&gt;posts_create&lt;/code&gt; runs only after your yes, and &lt;code&gt;posts_get&lt;/code&gt; reports the result.&lt;/li&gt;
&lt;/ol&gt;

&lt;p&gt;For several Instagram accounts, each account gets its own cover and variant instead of the identical carousel.&lt;/p&gt;

&lt;p&gt;WoopSocial doesn't write captions, crop media or set alt text, collaborators or music on Instagram. The skill writes the caption and alt text itself, and after posting it tells you to paste each slide's alt text from &lt;code&gt;manifest.json&lt;/code&gt; into the app (Edit &amp;gt; Edit alt text). If you'd rather not connect anything, it hands over the slides, caption and alt text and stops there.&lt;/p&gt;

&lt;h2&gt;
  
  
  Reading the results
&lt;/h2&gt;

&lt;p&gt;After 3 to 7 days the agent calls &lt;code&gt;social_account_posts_list&lt;/code&gt;, which returns the stored post metrics Instagram reports (views, likes, comments, saves, shares) with their collection date. A missing number is logged as "not reported", never as zero. Instagram's API reports per post, not per slide, so per-slide views stay in the app.&lt;/p&gt;

&lt;p&gt;Each carousel goes into &lt;code&gt;carousels.md&lt;/code&gt; with its date, format and cover, and the next step comes from comparing it with the account's own recent posts:&lt;/p&gt;

&lt;div class="table-wrapper-paragraph"&gt;&lt;table&gt;
&lt;thead&gt;
&lt;tr&gt;
&lt;th&gt;Saves&lt;/th&gt;
&lt;th&gt;Shares&lt;/th&gt;
&lt;th&gt;Next carousel&lt;/th&gt;
&lt;/tr&gt;
&lt;/thead&gt;
&lt;tbody&gt;
&lt;tr&gt;
&lt;td&gt;High&lt;/td&gt;
&lt;td&gt;High&lt;/td&gt;
&lt;td&gt;Same format, new topic&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;High&lt;/td&gt;
&lt;td&gt;Low&lt;/td&gt;
&lt;td&gt;Useful but not shareable: try a send-style last slide&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;Low&lt;/td&gt;
&lt;td&gt;High&lt;/td&gt;
&lt;td&gt;Shareable but thin: add a step or a checklist slide&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;Low&lt;/td&gt;
&lt;td&gt;Low&lt;/td&gt;
&lt;td&gt;Change the cover first, then the format&lt;/td&gt;
&lt;/tr&gt;
&lt;/tbody&gt;
&lt;/table&gt;&lt;/div&gt;

&lt;h2&gt;
  
  
  What makes it different
&lt;/h2&gt;

&lt;ul&gt;
&lt;li&gt;It covers the full job in one skill: format, copy, render, alt text, caption, posting and the results log.&lt;/li&gt;
&lt;li&gt;It renders exact 1080x1350 JPEGs with Pillow, so there is no browser to install and no PNG for the API to reject.&lt;/li&gt;
&lt;li&gt;It checks the cover the way the profile grid shows it, not only at full size.&lt;/li&gt;
&lt;li&gt;It writes slide 2 for the second chance, and it refuses invented numbers, reviews and stories.&lt;/li&gt;
&lt;/ul&gt;

&lt;h2&gt;
  
  
  What it won't do
&lt;/h2&gt;

&lt;ul&gt;
&lt;li&gt;Promise reach, saves or followers.&lt;/li&gt;
&lt;li&gt;Invent results, testimonials, user counts or first-person stories.&lt;/li&gt;
&lt;li&gt;Fake likes or view counts on tweet cards, or make a card look like someone else posted it.&lt;/li&gt;
&lt;li&gt;Post anything before you approve the exact slides, caption and account.&lt;/li&gt;
&lt;/ul&gt;

&lt;p&gt;The skill is MIT-licensed: &lt;a href="https://github.com/social-media-skills/instagram-carousel-skill" rel="noopener noreferrer"&gt;github.com/social-media-skills/instagram-carousel-skill&lt;/a&gt;. More free skills for social media work are at &lt;a href="https://social-media-skills.com" rel="noopener noreferrer"&gt;social-media-skills.com&lt;/a&gt;.&lt;/p&gt;

&lt;p&gt;If you render a carousel with it, which preset did you start with, and what did the warnings catch?&lt;/p&gt;

</description>
      <category>showdev</category>
      <category>claudecode</category>
      <category>ai</category>
      <category>opensource</category>
    </item>
  </channel>
</rss>
