Instagram Agent Skill is a free set of 13 Claude skills that write Reels, captions and weekly plans for an Instagram account, for creators who post themselves.

What makes it different: it writes, you post

The repo is by Jake Schincariol and is MIT licensed, with no signup, no API key and nothing to connect. Its README says nothing gets posted until you say yes, and that every skill ends the same way: a copy-ready block that you post yourself. The author explains why. The official Content Publishing API needs a Meta developer app, a linked Page, a long-lived token and app review, and the other ways people automate posting, commenting, following or DMing are browser automation or third-party tools that break Instagram's Terms of Use.

The one exception it names is keyword auto-replies in DMs, which Instagram supports through its own tools and approved partners and which only fire after someone comments first. /ig-dm says where that line is.

The 13 commands

/ig-reel: turns one idea into three hooks from 26 formulas, scored, then a script, on-screen text and a timed beat sheet.
/ig-viral: finds what is working in your niche and ranks it by each account's multiple over its own median, then writes a swipe file. It reads ten accounts at human speed while you drive your own browser, and never asks for your password.
/ig-caption: writes the caption and shows the 125 characters the feed displays before the "... more" tap.
/ig-carousel, /ig-story and /ig-plan: swipe posts with 1080x1350 files, the daily story sequence with a DM funnel, and a weekly plan with 10 accounts to engage with.
/ig-profile: scores your profile out of 100 against a 12-part rubric and rewrites it in fix-first order.
/ig-comment, /ig-reply and /ig-dm: comments on other people's posts in nine types, sorting and answering the thread under your own post, and DM keyword delivery, first message, collab pitch and two follow-ups.
/ig-repurpose and /ig-audit: one video, podcast or newsletter into a week of posts, and a post-mortem that ranks by outlier multiple and sends per reach rather than views.
/ig-human: the humanizer, described below.

Five scripts that run on your own machine

hookscore.py ranks hook options on a five-property panel and names dealbreakers such as opening with "stop scrolling" or a greeting. beats.py times a script at about 165 words per minute against a target length and flags a hook past three seconds, a beat long enough to lose the viewer, and no loop back to the first line.
caption.py prints the feed truncation as a box and checks length, first line, concrete hook, hashtags, one ask and search terms. It enforces the hashtag cap of five per post, which the author says Instagram cut from 30 on December 18, 2025.
humanize.py removes invisible characters such as zero-width spaces and soft hyphens, normalizes em dashes and curly quotes, and swaps 154 stock phrases for plain wording. detect.py then scores the draft on five checks. Structures that need judgment, like "It's not just X, it's Y" and rule-of-three lists, are flagged for a rewrite, not auto-fixed.
swipe.py ranks collected reels by the multiple over each account's own median and names the hook formula.
They use no dependencies and no network, and nothing is uploaded. Without Claude Code you can paste a single SKILL.md at the top of a chat, but you lose the five Python tools.

What the author measured, and what he admits

The hook scorer was tested on 74 real hooks taken from YouTube Shorts, not Reels, because Instagram view counts cannot be collected without logging into someone's account. The author states this as a limitation.
It catches bad hooks: against ten deliberately bad ones the AUC was 0.83. It does not pick winners: separating one creator's hits from their misses gave an AUC of 0.56, where 0.50 is a coin flip. Only the specificity check separated the two groups, and the stakes and address checks showed identical medians.
The README's advice is to kill the obviously weak hooks before shooting, then trust your own retention graph. It also reports that the formula classifier first named only 8% of real hooks and reached 49% after rewrite.
The detection checks are local heuristics, not GPTZero, Originality, Copyleaks, Winston or Turnitin. The author says nobody can honestly sell "undetectable". The invisible-character pass removes a checkable fingerprint and does not claim to defeat cryptographic watermarking.

Install and set up

In Claude Code: /plugin marketplace add Jakeschincariol/instagram-agent-skill then /plugin install instagram-agent. Or clone the repo and copy skills/ig-* into ~/.claude/skills/. Copying into a repo's .claude/skills/ makes it project-local.
Spend about ten minutes on templates/voice.md: copy it to ~/.claude/instagram/voice.md and fill it in, or give Claude three of your own reels and ask it to write the file. Every skill reads it, and the author says it matters because you have to say the words out loud.
If a draft needs a number you did not give, the output keeps a {{your number}} placeholder and a flag instead of inventing one.

Who is behind it, and how fresh it is

The repository was created on September 13, 2026, and had about 5,200 stars and 1,060 forks on November 7. Its last push was also September 13, so it has had no update since launch day. 35 issues were open on the GitHub card.
The author's GitHub profile reads "Always vibes, never slop". The README points to opusjake.ai and a sibling repo, linkedin-agent-skill, for LinkedIn.
Another repository has the same name: postzen-dev/instagram-agent-skill from the PostZen company, with 14 skills. It connects the PostZen MCP server so the agent can publish, schedule, reply to comments and DMs and read analytics through the official API. That is a separate project from this one, and Seailife did not check PostZen's pricing.
Seailife installed neither version and did not run the scripts. The platform rules in the README can go stale, and the author says that if the repo contradicts what Instagram does today, Instagram is right.

What makes it different: it writes, you post

The 13 commands

/ig-reel: turns one idea into three hooks from 26 formulas, scored, then a script, on-screen text and a timed beat sheet.

/ig-viral: finds what is working in your niche and ranks it by each account's multiple over its own median, then writes a swipe file. It reads ten accounts at human speed while you drive your own browser, and never asks for your password.

/ig-caption: writes the caption and shows the 125 characters the feed displays before the "... more" tap.

/ig-carousel, /ig-story and /ig-plan: swipe posts with 1080x1350 files, the daily story sequence with a DM funnel, and a weekly plan with 10 accounts to engage with.

/ig-profile: scores your profile out of 100 against a 12-part rubric and rewrites it in fix-first order.

/ig-comment, /ig-reply and /ig-dm: comments on other people's posts in nine types, sorting and answering the thread under your own post, and DM keyword delivery, first message, collab pitch and two follow-ups.

/ig-repurpose and /ig-audit: one video, podcast or newsletter into a week of posts, and a post-mortem that ranks by outlier multiple and sends per reach rather than views.

/ig-human: the humanizer, described below.

Five scripts that run on your own machine

hookscore.py ranks hook options on a five-property panel and names dealbreakers such as opening with "stop scrolling" or a greeting. beats.py times a script at about 165 words per minute against a target length and flags a hook past three seconds, a beat long enough to lose the viewer, and no loop back to the first line.

caption.py prints the feed truncation as a box and checks length, first line, concrete hook, hashtags, one ask and search terms. It enforces the hashtag cap of five per post, which the author says Instagram cut from 30 on December 18, 2025.

humanize.py removes invisible characters such as zero-width spaces and soft hyphens, normalizes em dashes and curly quotes, and swaps 154 stock phrases for plain wording. detect.py then scores the draft on five checks. Structures that need judgment, like "It's not just X, it's Y" and rule-of-three lists, are flagged for a rewrite, not auto-fixed.

swipe.py ranks collected reels by the multiple over each account's own median and names the hook formula.

They use no dependencies and no network, and nothing is uploaded. Without Claude Code you can paste a single SKILL.md at the top of a chat, but you lose the five Python tools.

What the author measured, and what he admits

The hook scorer was tested on 74 real hooks taken from YouTube Shorts, not Reels, because Instagram view counts cannot be collected without logging into someone's account. The author states this as a limitation.

It catches bad hooks: against ten deliberately bad ones the AUC was 0.83. It does not pick winners: separating one creator's hits from their misses gave an AUC of 0.56, where 0.50 is a coin flip. Only the specificity check separated the two groups, and the stakes and address checks showed identical medians.

The README's advice is to kill the obviously weak hooks before shooting, then trust your own retention graph. It also reports that the formula classifier first named only 8% of real hooks and reached 49% after rewrite.

The detection checks are local heuristics, not GPTZero, Originality, Copyleaks, Winston or Turnitin. The author says nobody can honestly sell "undetectable". The invisible-character pass removes a checkable fingerprint and does not claim to defeat cryptographic watermarking.

Install and set up

In Claude Code: /plugin marketplace add Jakeschincariol/instagram-agent-skill then /plugin install instagram-agent. Or clone the repo and copy skills/ig-* into ~/.claude/skills/. Copying into a repo's .claude/skills/ makes it project-local.

Spend about ten minutes on templates/voice.md: copy it to ~/.claude/instagram/voice.md and fill it in, or give Claude three of your own reels and ask it to write the file. Every skill reads it, and the author says it matters because you have to say the words out loud.

If a draft needs a number you did not give, the output keeps a {{your number}} placeholder and a flag instead of inventing one.

Who is behind it, and how fresh it is

The repository was created on September 13, 2026, and had about 5,200 stars and 1,060 forks on November 7. Its last push was also September 13, so it has had no update since launch day. 35 issues were open on the GitHub card.

The author's GitHub profile reads "Always vibes, never slop". The README points to opusjake.ai and a sibling repo, linkedin-agent-skill, for LinkedIn.

Another repository has the same name: postzen-dev/instagram-agent-skill from the PostZen company, with 14 skills. It connects the PostZen MCP server so the agent can publish, schedule, reply to comments and DMs and read analytics through the official API. That is a separate project from this one, and Seailife did not check PostZen's pricing.

Seailife installed neither version and did not run the scripts. The platform rules in the README can go stale, and the author says that if the repo contradicts what Instagram does today, Instagram is right.

Instagram Agent Skill

What makes it different: it writes, you post

The 13 commands

Five scripts that run on your own machine

What the author measured, and what he admits

Install and set up

Who is behind it, and how fresh it is

Comments

Instagram Agent Skill

What makes it different: it writes, you post

The 13 commands

Five scripts that run on your own machine

What the author measured, and what he admits

Install and set up

Who is behind it, and how fresh it is

Comments