Instagram Agent Skill: Claude skills for reviewable Instagram content
A Claude skill pack for Instagram creators that drafts Reels through DMs without logging in or posting.
GitHub Jakeschincariol/instagram-agent-skill Updated 2026-11-20 Branch main Stars 7.4K Forks 1.4K
Python Instagram content automation Claude Skills Claude Code

🧭 Decision Guide

Try it if you

  • You use Claude Code for Instagram content and need /ig-reel to generate scripts and beat sheets.
    The README's “The thirteen” section says /ig-reel generates hooks, scripts, on-screen text, and a timed beat sheet.
  • You want text and the five Python tools to stay local without external network services.
    The README's “The five tools that actually run” section says “No dependencies, no network, nothing uploaded”.
  • You are willing to publish manually and want automation limited to drafts and pre-approval checks.
    The README says “Nothing gets posted until you say yes” and explicitly states that these skills do not post to Instagram.
  • You need to screen obviously weak hooks with 74 short-form examples rather than predict viral winners.
    The README's “What I actually measured” section reports AUC 0.83 for deliberately bad hooks but only 0.56 AUC for separating hits from misses.

Skip it if you

  • You need the tool to automatically publish, comment, follow, or send Instagram DMs at scale.
    The README's “The fine print” says the skills do not post and that browser automation and third-party tools violate Instagram's Terms of Use.
  • You need native Instagram Reel view data to train or validate the hook evaluator.
    The README says the evaluation corpus uses 74 YouTube Shorts because Instagram view counts cannot be collected without logging in.
  • You need an official verdict from GPTZero, Originality, or Turnitin.
    The README says the five checks are local heuristics, not GPTZero, Originality, Copyleaks, Winston, or Turnitin.
  • You need mature release and multi-maintainer signals.
    Project metadata shows 1 contributor, 0 releases, and the latest version as “No releases”.

Requirements

  • Claude is required; the README provides a “Paste this into Claude” installation method.
  • With Claude Code, use the README's git clone and cp commands.
  • The five tools run locally; the README says “No dependencies, no network, nothing uploaded”.
  • Prepare templates/voice.md and copy it to ~/.claude/instagram/voice.md; every skill reads this file.
  • Without Claude Code, a single SKILL.md can be pasted at the top of a chat, but the five Python tools are lost.

First step (verbatim from README)

git clone https://github.com/Jakeschincariol/instagram-agent-skill.git

Watch out

  • Platform rules become stale: the hashtag cap changed from 30 to 5, while the old linter still used 30.
    The README's “The fine print” says the cap changed from 30 to 5 in December 2025 and the linter was updated after verification.
  • /ig-viral reads rather than scrapes: you manually browse 10 accounts with about 12 Reels each.
    The README says /ig-viral is driven through your browser, never asks for your password, and never logs in as you.
  • The hook classifier abstains on 51% of inputs, so an empty classification is not necessarily an error.
    The README reports 49% coverage after rewriting and says it abstains on the remaining 51%.
  • SPECIFICITY originally counted only digits, so spoken numbers such as “zero dollars” were treated as non-specific.
    The measurement section documents this fixed SPECIFICITY bug.
  • The project only creates drafts; Instagram publishing still requires copying, human approval, and manual action.
    The README says every skill ends with a copy-ready block that the user posts.

Alternatives

  • Instagram native tools and approved partners:They are better when you need keyword DM auto-replies triggered after someone comments; the README says Instagram native tools and approved partners support this case.
    The fine print

Not stated in the README

  • The README does not specify a required Claude version or minimum Claude Code version.
  • The README does not identify the five Python tools, their minimum Python version, or supported operating systems.
  • The README does not provide full implementation details for all 13 skills; sections such as Hooks, The humanizer, and Files are omitted.
  • The README does not state runtime, memory usage, or per-run scale limits.
  • The README does not say whether the complete 74-hook corpus is included in the repository.
  • The README does not state whether an Instagram account must be Professional for all content-generation features.
  • The README does not describe a long-term maintenance plan for the 1-contributor, 0-release project.

💡 Deep Analysis

6
It depends I make Instagram Reels every week and want to remove greetings, long preambles, and vague hooks before filming. I will not accept a score being presented as a viral prediction. Do the `/ig-reel` test results support this decision?
For: An educator or professional creator who wants hook scoring to reduce wasted Reel takes but needs to distinguish filtering weak hooks from predicting viral winners

It depends: use it to reject obviously weak hooks, but not to predict which of two acceptable hooks will go viral.

  • The README’s test corpus contains 74 real short-form hooks from YouTube Shorts, not Instagram Reels, so there is a cross-platform limitation.
  • Against 10 deliberately bad hooks, the scorer achieved AUC 0.83, with 9 scoring below the real-corpus median, indicating that it can identify openings built on greetings, preambles, or vague content.
  • However, separating a creator’s hits from that creator’s ordinary posts produced AUC 0.56, close to the 0.50 coin-flip baseline; the README explicitly says it “does not pick winners.”
  • Facial delivery, editing, audio, and distribution are outside text scoring, so the score cannot replace retention data or post-publication evidence.

The README does not provide an Instagram Reels-specific test set or show that all 26 formulas work equally across languages, niches, or on-camera formats.

  • README, “What I actually measured”: the corpus contains 74 YouTube Shorts hooks rather than Reels
  • README, “What I actually measured”: 10 deliberately bad hooks produced AUC 0.83, with 9 of 10 below the real-corpus median
  • README, “What I actually measured”: separating hits from ordinary content by the same creator produced AUC 0.56, with 0.50 as the coin-flip baseline
  • README, “What I actually measured”: “It does not pick winners.”
https://github.com/Jakeschincariol/instagram-agent-skill

Install this skill, then confirm /ig-reel works.
Not stated in the README:The README does not provide scorer performance on native Instagram Reels data.;The README does not report scorer performance on Chinese or other non-English spoken hooks.;The README does not show whether the 26 formulas cover expression differences across niches.
Yes I do not allow tools to log into Instagram or scrape competitors in bulk, but I can research niche accounts manually in a browser. I want ranking by multiples over each account's own median rather than raw views. Is `/ig-viral` suitable for me?
For: An Instagram content planner who refuses account login or scraping, is willing to research niche Reels manually in a browser, and wants a swipe file ranked against each account's own median performance

Yes, but it is a manual research aid rather than an automated competitor crawler; that matches your constraints exactly.

  • /ig-viral finds Reels working in a niche, ranks them by multiples over each account’s own median, identifies formulas, and creates a swipe file.
  • The project insight says the user researches accounts manually in a browser and the tool does not scrape data in bulk, avoiding automated login or attempts to bypass Instagram rules.
  • The README explains that YouTube Shorts were used for hook evaluation because Instagram view counts cannot be collected without logging into someone else’s account, and scraping could violate the Terms.
  • It therefore fits a manually maintained competitor swipe file, but not automatic collection across many accounts, scheduled synchronization, or guaranteed sample completeness.

The README does not define the manual research table schema, sample size, import format, or how reusable records remain after Instagram page changes.

  • README, “The thirteen”: `/ig-viral` ranks by multiples over each account’s own median and creates a swipe file
  • Project insight, “common_pitfalls”: `ig-viral` depends on manual browser research and user-entered results and does not scrape in bulk
  • README, “What I actually measured”: Instagram views cannot be collected without logging into someone else’s account, and scraping may violate the Terms
  • Project insight, “innovation_points”: relative multiples over an account’s own median reduce account-size bias
https://github.com/Jakeschincariol/instagram-agent-skill

Install this skill, then confirm /ig-reel works.
Not stated in the README:The README does not specify the format for entering manual research results into `/ig-viral`.;The README does not say whether a long-lived swipe file can be stored, updated, or exported.
Yes I already use Claude Code to run one Instagram account. I want to connect `/ig-reel`, `/ig-caption`, `/ig-plan`, and `/ig-audit` into a weekly workflow, but I do not want to obtain an API key or connect my account to a third-party service. Is this project suitable for me?
For: An Instagram creator already using Claude Code who wants the 13 skills to cover Reels through weekly planning without configuring an API key

Yes, because it is designed for Claude Code, local execution, and manual Instagram publishing.

  • The README says “no signup, no API key, nothing to connect,” and exposes 13 skills after installation.
  • /ig-reel handles 26 hook formulas, scripts, on-screen text, and beat sheets; /ig-caption checks the first 125 feed-visible characters; /ig-plan creates a weekly schedule; and /ig-audit reviews published posts.
  • The five tools “run on your machine, on your text,” with no network, dependencies, or upload requirement.
  • The boundary is explicit: “Nothing gets posted until you say yes”; publishing remains manual.

The README does not explain whether multiple accounts, persistent weekly task history, or a shared historical database are supported.

  • README, “The Instagram agent skill”: “Thirteen Claude skills that run an Instagram account. Free, MIT, no signup, no API key, nothing to connect.”
  • README, “The thirteen”: descriptions of `/ig-reel`, `/ig-caption`, `/ig-plan`, and `/ig-audit`
  • README, “The five tools that actually run”: “No dependencies, no network, nothing uploaded. They run on your machine, on your text.”
  • README, “The Instagram agent skill”: “Nothing gets posted until you say yes. These skills write. You post.”
git clone https://github.com/Jakeschincariol/instagram-agent-skill.git
cp -r instagram-agent-skill/skills/ig-* ~/.claude/skills/
Not stated in the README:The README does not say whether voice.md, history, and outputs can be safely isolated across multiple Instagram accounts.;The README does not say whether `/ig-audit` can automatically ingest Instagram native analytics.
Yes I run an Instagram personal brand and need Reels, captions, comments, and DMs to sound like me. I can provide three of my own Reels, but I do not want to repeat voice instructions for every skill. Is this project suitable for me?
For: A consultant or coach running an Instagram personal brand who needs consistent voice across Reels, captions, comments, and DMs and can provide three of their own Reels

Yes, provided that you are willing to capture your real speaking style in the shared voice.md, because the project uses it as the consistency layer for every skill.

  • The README instructs users to copy templates/voice.md to ~/.claude/instagram/voice.md and says, “Every skill reads that file.”
  • You can also give Claude three of your own Reels and ask it to write voice.md, which matches your constraint exactly.
  • The shared file influences Reels, captions, comments, replies, DMs, and Stories, reducing repeated style instructions.
  • The README also says to add audience, forbidden expressions, product facts, tone boundaries, and common CTAs; installing skills without configuring the file can produce generic copy.

The README does not provide the complete voice.md schema, a version-control workflow, or a way to switch among multiple brand identities in one Claude Code environment.

  • README, “Install”: “Then spend ten minutes on `templates/voice.md`.”
  • README, “Install”: “send Claude three of your own reels and say ‘write my voice.md from these’.”
  • README, “Install”: “Every skill reads that file.”
  • README, “Install”: voice.md should include audience, forbidden expressions, product facts, tone boundaries, and common CTAs
https://github.com/Jakeschincariol/instagram-agent-skill

Install this skill, then confirm /ig-reel works.
Not stated in the README:The README does not specify the complete voice.md schema, conflict rules, or change-tracking method.;The README does not say whether separate voice.md files for multiple brands or accounts are supported in one environment.
It depends Each week I have one video, podcast, or newsletter and want to turn it into standalone Reels and carousels. My team requires human confirmation for all comments, DMs, and publishing, with no automated login or scheduling. Can this project cover my workflow?
For: A solo operator who wants to turn a video, podcast, or newsletter into a week of Instagram content while requiring manual review and publishing and rejecting automated comments, DMs, or scheduling

It depends: it covers repurposing, copy, and interaction drafts, but not automated scheduling, login, or publishing. It fits a manual workflow, not an end-to-end automation team.

  • /ig-repurpose explicitly turns one video, podcast, or newsletter into a week of Reels and carousels, with each piece standing alone.
  • /ig-comment, /ig-reply, and /ig-dm generate comments, thread replies, and DM sequences, but the project insight says these actions are not executed automatically.
  • The README says, “These skills write. You post,” making human review and manual publishing a product boundary rather than an optional setting.
  • /ig-plan can plan formats, timing, and accounts to engage with, but the README does not say that it connects to an Instagram scheduler or triggers publication.

Unknowns include whether repurposing handles transcription, image-asset generation, or multi-person approval; the README only establishes text and planning capabilities.

  • README, “The thirteen”: `/ig-repurpose` turns one video, podcast, or newsletter into a week of Reels and carousels
  • README, “The thirteen”: descriptions of `/ig-comment`, `/ig-reply`, and `/ig-dm`
  • README, “The Instagram agent skill”: “These skills write. You post.”
  • Project insight, “usage_limitations”: no automatic publishing, commenting, following, bulk DMs, or account login
  • README, “The thirteen”: `/ig-plan` plans what to post, which format, when, and which accounts to engage with
https://github.com/Jakeschincariol/instagram-agent-skill

Install this skill, then confirm /ig-reel works.
Not stated in the README:The README does not say whether `/ig-repurpose` automatically transcribes videos or podcasts.;The README does not clarify whether the 1080x1350 carousel files are merely output specifications or actual generated image files.;The README does not describe multi-person approval, comment assignment, or scheduler integrations.
Yes I create educational Instagram Reels and captions and cannot let a model invent clients, revenue, or performance numbers. If real data is missing, I want a placeholder rather than fabricated claims. Does this project fit that compliance constraint?
For: A professional service provider creating educational Instagram content who prioritizes factual accuracy and cannot invent client outcomes or marketing numbers

Yes, because the README explicitly uses placeholders and manual confirmation as safeguards instead of encouraging the model to invent outcomes.

  • The project insight says to use {{your number}} when clients, metrics, or results were not provided, rather than fabricating clients, numbers, or case studies.
  • The 13 skills cover Reels, captions, carousels, Stories, DMs, and audits, but the output remains a draft for review; the README says nothing is posted until the user says yes.
  • /ig-human removes em dashes, slop vocabulary, and invisible characters, then evaluates text with a five-check panel, making it useful as a pre-publishing quality gate.
  • It cannot prove that a marketing claim is true and is not professional compliance review; you still have to replace placeholders with verifiable data.

The README does not provide regulated-industry templates, citation fields, approval logs, or sector-specific rules, so this is a drafting and checking tool rather than a compliance system.

  • Project insight, “Fact and compliance protection”: use `{{your number}}` when numbers are not provided and do not invent clients, metrics, or results
  • README, “The Instagram agent skill”: “Nothing gets posted until you say yes. These skills write. You post.”
  • README, “The thirteen”: `/ig-human` removes dashes, slop vocabulary, and invisible characters and performs five checks
  • Project insight, “common_pitfalls”: retain `{{your number}}` when real numbers are unavailable and do not let the model fill it in
https://github.com/Jakeschincariol/instagram-agent-skill

Install this skill, then confirm /ig-reel works.
Not stated in the README:The README does not say whether outputs retain source references, citation links, or review records.;The README does not define additional compliance rules for regulated industries such as healthcare, finance, or law.

✨ Highlights

  • 13 Claude skills cover Reels, captions, DMs, and profiles
  • Reels use 26 hook formulas and output scores and scripts
  • Five Python tools run locally without network access or uploads
  • The hook test reached only 0.56 AUC for separating creator hits

🔧 Engineering

  • /ig-reel turns one idea into 3 hooks, a script, and a beat sheet
  • /ig-viral ranks niche Reels by multiples over each account's median
  • /ig-profile scores an account out of 100 with a 12-part rubric
  • /ig-human removes em dashes, slop vocabulary, and invisible characters with five checks

⚠️ Risks

  • The project does not post to Instagram; publishing remains manual
  • The hook evaluation used 74 YouTube Shorts, not Instagram Reels
  • The hook classifier improved from 8% to 49% and abstains on the other 51%
  • Maintenance signals are limited: 1 contributor, 0 releases, and 2 recent commits

👥 For who?

  • Individual creators using Claude or Claude Code for Instagram content
  • Teams needing local text processing without an API key or external network
  • Users who want manual publishing while automating Reel and caption drafts