instagram-skills: A Claude Code Skill Pack for Instagram Captions, Carousels, and Hooks

sergebulaev/instagram-skills has been on GitHub since early July, but the number worth paying attention to is its growth over the last few days: 100 stars on September 19, 105 on the 20th, 120 on the 21st, and 138 as of this week, a sustained climb rather than a single viral spike. That kind of steady, multi-day build is a better signal than a one-day spike for a Claude Code Instagram skill, a category this specific. It is free, MIT licensed, and it does something narrower and more useful than most content-generation tools: it writes for Instagram specifically, not for social media in general.

What instagram-skills actually does

The repo packages nine separate skills, each one scoped to a single Instagram task rather than one do-everything prompt. A Caption Writer drafts captions with the hook placed in the first 125 characters, the exact character count before Instagram truncates a caption in feed. A Carousel Planner lays out multi-slide sequences with each slide positioned for a specific job, the hook slide, the value slides, the close. A Hook Extractor takes a Reel URL or pasted transcript and reverse-engineers the opening line that made it work, so you can study a competitor's hook mechanics instead of guessing at them. A Hashtag Strategist builds sets of three to five tags sized for realistic reach rather than the old spray-thirty-tags approach. And a Humanizer runs a pre-publish audit specifically to strip AI tells before anything goes out.

The remaining four skills handle the operational side: a Content Planner that builds a weekly schedule with posting times and a goal mix (awareness versus conversion versus engagement), a Repurposer that converts a LinkedIn post, blog, YouTube script, or X thread into Instagram format, a Profile Optimizer that audits bios, highlights, grid layout, and pinned posts, and an Audience Insights skill that pulls niche and competitor data through Apify.

What using it actually looks like

Skills activate automatically inside Claude Code or Codex when you ask an Instagram-shaped question, rather than requiring you to invoke a specific command. Ask for a carousel about a product launch, and the Carousel Planner skill picks it up and drafts slide-by-slide copy. The underlying library is plain Python, with a url_parser.py for pulling data out of Instagram links and a publora_client.py that handles an optional draft-to-upload-to-schedule flow if you connect a free Publora account. Every skill's output is a draft, not a publish. Nothing goes live without you approving it first, which matters more here than it sounds: a tool that writes in your voice is only useful if a human still checks the tone before it reaches an audience.

Media is the one piece it doesn't touch. These skills write the text, the hooks, the hashtags, and the plan. You still need the photo or video. That is a reasonable division of labor for a text-generation skill, but worth knowing going in if you were hoping for an end-to-end pipeline.

Where it fits next to the rest of the family

Instagram-skills is one of a matching set from the same author, alongside standalone x-skills and threads-skills packages built the same way, one skill pack per platform rather than a single generic social-media tool trying to cover all three. That platform-specific design is the actual value proposition here. A caption that works on X rewards brevity and a strong single line. An Instagram caption has a completely different job: it has to survive the truncation point, carry a hook for someone scrolling fast, and set up whatever's in the carousel or Reel it's attached to. Treating those as the same writing problem is why a lot of AI-generated social copy reads generic regardless of platform. Splitting the skill by platform is a small architectural choice that produces a meaningfully different output.

For anyone who is already comfortable running Claude Code or Codex and wants Instagram copy that accounts for the platform's actual mechanics, this clears a real bar. The caption hook length, the carousel slide sequencing, and the hashtag sizing are all specific, testable claims about how Instagram works, not vague AI-marketing promises, and they hold up against how the platform actually behaves.

Who it's not for

If you don't already have Claude Code, Codex, or a compatible agent set up, this isn't a standalone app, and installing an entire coding agent just to get Instagram captions is a heavier lift than it needs to be for a casual poster. The optional integrations, Publora for scheduling and Pixfaro for AI carousel illustrations, both require separate signups, so the fully automated path has more moving parts than the core caption and planning skills alone. And because every skill still needs a human to review drafts before anything ships, this replaces the blank-page problem, not the judgment call about whether a given caption is actually good.

The growth curve is worth watching rather than treating as proof by itself, and it's worth noting the code itself was last updated on September 17, so the recent star growth is word of mouth, not a new release driving it. A skill pack can climb for a week and then plateau once the initial audience of early adopters runs out. What makes this one worth trying now, rather than waiting to see if it sticks, is that the underlying mechanic, platform-specific skills instead of one generic prompt, is sound regardless of how the star count moves from here.

Source: sergebulaev/instagram-skills on GitHub.