instagram-agent-skill: 13 Claude skills for Reels, captions and planning

Most of the hard work in an Instagram account happens in two places: the first few seconds of a Reel, and the first line of a caption, which is all most people see before they tap “more”. instagram-agent-skill is a free Instagram agent skill made of 13 Claude skills. They cover Reels, captions, carousels, Stories, profile audits, weekly plans, comments and DMs, with a few standard-library Python tools that score opening lines, estimate spoken length and lint captions. GitHubDaily recommended it on X on November 10, 2026. On the same day, the GitHub repository showed 5,743 stars and an MIT license.

Core features

• Opening-line scoring: /ig-reel drafts three openings from 26 hook formulas listed in the README. hookscore.py then scores each on five properties and names the weakest one. In our test in a scratch clone, an opening with a concrete dollar figure scored 70.0 (STRONG). A sample that began “Hey guys” and talked about content strategy scored 9.6 (WEAK). The tool’s reason was the greeting: nobody scrolls the feed to be greeted.
• Spoken timing: beats.py estimates how long each line takes to say, stacks the lines into timecodes, and flags the problems the README lists: an opening that runs past three seconds, a beat long enough to lose the viewer, several lines with nothing concrete, and an ending that never returns to the opening. These come from the README; we did not run this script on its own.
• Caption preview and lint: caption.py shows the roughly 125 characters the feed displays before the tap on “more”. It then checks length, whether the first line is cut mid-sentence, the hashtag count, the search terms you pass in, and the call to action. The README sets the hashtag limit at five. Our six-tag sample came back as FIX: it went over the limit, and neither of the two search terms appeared.
• Humanizer and detector: humanize.py strips zero-width and other invisible characters, converts em dashes and curly quotes to plain punctuation, and replaces 154 stock phrases listed in slop.json, a file you can edit. detect.py scores text on five local heuristics: burstiness, specificity, stock-phrase density, character fingerprint and voice. The README is explicit that these are not calls to GPTZero, Originality or any other detector, and that no detection result is promised. On a sample we wrote to be full of stock phrasing, detect.py gave a human score of 15.1 and flagged it.
• The rest of the set: /ig-profile scores a profile out of 100 against a 12-part rubric and rewrites it in fix-first order. /ig-plan builds the week. /ig-viral ranks what works in a niche by each account’s own median, names the hook formula and writes a swipe file. /ig-audit ranks your past posts by outlier multiple and sends per reach, not raw views. /ig-repurpose turns one video, podcast or newsletter into a week of reels and carousels. /ig-comment, /ig-reply and /ig-dm draft comments, replies and direct messages. The skills write drafts; they do not send anything.

Typical use cases

• A solo creator teaching a niche: plan the week with /ig-plan, get three openings from /ig-reel, sort them with hookscore.py, and check the runtime with beats.py before filming. You still decide what goes live.
• An operator studying a niche: /ig-viral works through about ten accounts, roughly a dozen reels each, in your own browser at human speed. The README says it reads and does not scrape, and that it never logs in as you.
• Someone whose drafts read like templates: run detect.py for a baseline, then humanize.py --report to see every change. Patterns such as “It’s not just X, it’s Y” are flagged for a person to rewrite, not swapped out by a regex.

Quick start

The README’s install commands:

git clone https://github.com/Jakeschincariol/instagram-agent-skill.git
cp -r instagram-agent-skill/skills/ig-* ~/.claude/skills/

The README then asks you to copy templates/voice.md to ~/.claude/instagram/voice.md and fill it in, because every skill reads that file. The Python tools need no dependencies. The README’s first-run example assumes you are in skills/ig-reel/, with one candidate opening per line in hooks.txt:

python3 hookscore.py hooks.txt

The output is a ranked list. Each line carries STRONG, OK or WEAK, and the line below it names the weakest property. Our test run printed the same format.

Benchmarks

The README publishes the author’s own hook test. We did not reproduce it. The corpus is 74 opening lines from YouTube Shorts, taken from the auto-captions of the first three seconds. It is not Instagram Reels data, because Instagram does not expose view counts without logging in. The README gives no test date, and the measurement script is not in the repository. An AUC of 0.5 is a coin flip; closer to 1 means better separation.

Test	Result	The README’s reading
Ten deliberately weak openings	AUC 0.83; nine scored below the corpus median	Flags clearly weak openings
Same creator’s hits vs misses	AUC 0.56	Cannot tell which of two decent openings will travel
Formula classifier on real hooks	8% matched, then 49% after rewrites	The first regexes were written from the author’s own templates

Source: the README. Hardware and date: not stated. In this corpus, only specificity (SPECIFICITY) separated the hits from the misses. The medians for STAKES and ADDRESS were identical in both groups.

Summary

Use it if you want Claude to draft, check and plan Instagram content while you keep the decision to publish. Skip it if you need automatic posting, commenting, following or direct messaging. The README says the Content Publishing API is open to Professional accounts and needs a Meta developer app, a linked Page, a long-lived token and app review. It also says other automation, whether browser automation or third-party tools, breaks Instagram’s Terms of Use and gets accounts action-blocked. That is why the skills stop at a draft you can copy.

Keep these limits in mind:

• It does not publish. The output is text you copy and post. The README treats this as the design, not a setting.
• The hook test only screens out weak openings. The author’s own numbers show it can flag clearly weak lines, but it cannot tell two decent openings apart.
• The detection checks are local heuristics. They make no promise about external detectors, as the README says.
• Platform numbers go stale. Hashtag limits and similar rules change, and the README says to trust Instagram’s current rules. The repository was created and last pushed on September 13, 2026, with no push since. On November 10 it had 50 open issues and pull requests combined.

It is MIT-licensed and written in Python. The tools use only the standard library and make no network calls. If you want a tool that turns a topic into a finished video, look at MoneyPrinterTurbo, which renders video. This set produces scripts, captions and drafts; it does not edit video or publish. The same account also publishes replica-skill, a separate set of Claude skills for rebuilding an app.