Skip to content
Back to search

instagram-agent-skill

Thirteen Claude skills that run an Instagram account: Reels off 26 scored hook formulas, a viral research skill, a caption linter, and a humanizer. Free, MIT.

Safety scan

A static text scan, not a guarantee. Select the badge to see what was scanned and every finding.

Checked today
Badge for your README
findsafeskills safety rating

Markdown

[![findsafeskills safety rating](/api/listings/jakeschincariol--instagram-agent-skill/badge.svg)](/listings/jakeschincariol--instagram-agent-skill)

HTML

<a href="/listings/jakeschincariol--instagram-agent-skill"><img src="/api/listings/jakeschincariol--instagram-agent-skill/badge.svg" alt="findsafeskills safety rating" /></a>

Install

i

Install instructions from the README

As written by the repo's author. findsafeskills has not verified these steps.

Paste this into Claude:

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

Install this skill, then confirm /ig-reel works.

Or do it yourself, in Claude Code:

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

Or as a plugin:

/plugin marketplace add Jakeschincariol/instagram-agent-skill
/plugin install instagram-agent

Project-local instead of global: copy the same folders into your repo's `.claude/skills/`. No Claude Code at all? Paste any single `SKILL.md` at the top of a chat and it runs as a mode. You lose the five Python tools, which is most of the point of `/ig-reel` and `/ig-human`, but the rest works.

Then spend ten minutes on `templates/voice.md`. Copy it to `~/.claude/instagram/voice.md` and fill it in, or send Claude three of your own reels and say "write my voice.md from these". Every skill reads that file. It matters more here than on other platforms, because you have to say the words out loud.

What was scanned

The files the safety scan read, as written by the repository's author and shown as plain text. Lines with a finding are marked.

.claude-plugin/marketplace.json · Manifest · 18 lines

View this file on GitHub (read at the repository's current version: the scan did not record which commit it read)

  1. {
  2. "$schema": "https://anthropic.com/claude-code/marketplace.schema.json",
  3. "name": "instagram-agent-skill",
  4. "description": "The Instagram agent skill pack for Claude",
  5. "owner": {
  6. "name": "Jake Schincariol",
  7. "url": "https://opusjake.ai"
  8. },
  9. "plugins": [
  10. {
  11. "name": "instagram-agent",
  12. "description": "Thirteen Claude skills that run an Instagram account, including a scored hook panel, a viral research skill, a caption linter and a humanizer.",
  13. "source": "./",
  14. "category": "productivity",
  15. "homepage": "https://github.com/Jakeschincariol/instagram-agent-skill"
  16. }
  17. ]
  18. }
README.md · README · 338 lines

View this file on GitHub (read at the repository's current version: the scan did not record which commit it read)

  1. # The Instagram agent skill
  2. Thirteen Claude skills that run an Instagram account. Free, MIT, no signup, no
  3. API key, nothing to connect.
  4. One of them writes your Reels off 26 hook formulas and scores the hook before
  5. you waste a take on it. One goes and finds the reels that are actually working
  6. in your niche and ranks them by how far each beat its own account. One writes
  7. the caption and shows you exactly what the feed shows before the "... more".
  8. One scores your profile out of 100 and rewrites what lost points. One plans the
  9. week.
  10. And one is the humanizer, which is the reason the rest are usable. It strips
  11. the em dashes, the slop vocabulary and the invisible watermark characters out
  12. of a draft, then scores what is left against a five-check panel before you ever
  13. see it.
  14. **Nothing gets posted until you say yes.** These skills write. You post.
  15. ## Install
  16. Paste this into Claude:
  17. ```
  18. https://github.com/Jakeschincariol/instagram-agent-skill
  19. Install this skill, then confirm /ig-reel works.
  20. ```
  21. Or do it yourself, in Claude Code:
  22. ```bash
  23. git clone https://github.com/Jakeschincariol/instagram-agent-skill.git
  24. cp -r instagram-agent-skill/skills/ig-* ~/.claude/skills/
  25. ```
  26. Or as a plugin:
  27. ```
  28. /plugin marketplace add Jakeschincariol/instagram-agent-skill
  29. /plugin install instagram-agent
  30. ```
  31. Project-local instead of global: copy the same folders into your repo's
  32. `.claude/skills/`. No Claude Code at all? Paste any single `SKILL.md` at the top
  33. of a chat and it runs as a mode. You lose the five Python tools, which is most
  34. of the point of `/ig-reel` and `/ig-human`, but the rest works.
  35. Then spend ten minutes on `templates/voice.md`. Copy it to
  36. `~/.claude/instagram/voice.md` and fill it in, or send Claude three of your own
  37. reels and say "write my voice.md from these". Every skill reads that file. It
  38. matters more here than on other platforms, because you have to say the words
  39. out loud.
  40. ## The thirteen
  41. | command | what it does |
  42. | --- | --- |
  43. | `/ig-reel` | One idea into a Reel. Three hooks from [26 formulas](skills/ig-reel/hooks.json), scored, then the script, the on-screen text and a timed beat sheet. |
  44. | `/ig-viral` | Goes and finds what is working in your niche, ranks it by multiple over each account's own median, names the formula, writes the swipe file. |
  45. | `/ig-caption` | The caption, linted. Shows you the 125 characters the feed actually shows before the tap. |
  46. | `/ig-carousel` | Swipe posts. The cover that earns the swipe, slide copy, and the 1080x1350 files. |
  47. | `/ig-story` | The daily story sequence, which sticker does which job, and the DM funnel that starts with them moving first. |
  48. | `/ig-profile` | Scores your profile against a [12-part rubric](skills/ig-profile/rubric.json) out of 100, then rewrites in fix-first order. |
  49. | `/ig-plan` | The week. What to post, which format, when, and the 10 accounts to engage with. |
  50. | `/ig-human` | The humanizer. Two scripts that actually run. See below. |
  51. | `/ig-comment` | Comments on other people's posts. Nine types, picked by what the post actually is. Never "🔥🔥🔥". |
  52. | `/ig-reply` | The thread under your own post. Sorts into keyword / lead / substance / question / support / noise, then writes in that order. |
  53. | `/ig-dm` | The keyword delivery, the first message, the collab pitch, and the two follow-ups. Two. |
  54. | `/ig-repurpose` | One video, podcast or newsletter into a week of reels and carousels that each stand alone. |
  55. | `/ig-audit` | Post-mortem on what you already posted. Ranks by outlier multiple and sends per reach, not views. |
  56. ## The five tools that actually run
  57. No dependencies, no network, nothing uploaded. They run on your machine, on
  58. your text.
  59. ### Hooks
  60. ```bash
  61. python3 hookscore.py hooks.txt # rank your options
  62. python3 beats.py script.txt --target 30 # time it before you shoot it
  63. ```
  64. ```
  65. HOOK RANKING
  66. ==============================================================================
  67. -> 85.6 STRONG Nobody tells you that your first 30 reels are supposed to f...
  68. weakest: SPECIFICITY (75)
  69. 81.4 STRONG I lost $18,000 because of one missing contract.
  70. weakest: ADDRESS (70)
  71. 54.4 OK Stop scrolling if you want to grow on Instagram in 2026 🔥
  72. weakest: STAKES (70)
  73. dealbreaker: Opens with "stop scrolling". Asking for attention proves
  74. you have not earned it.
  75. 9.6 WEAK Hey guys, today I wanted to talk about content strategy
  76. weakest: FRONTLOAD (0)
  77. dealbreaker: Greeting. Nobody came to the feed to be greeted.
  78. ```
  79. `beats.py` estimates how long each line takes to say, stacks them into
  80. timecodes, and flags the four things that kill a Reel in the edit: a hook that
  81. runs past three seconds, a beat long enough for the viewer to leave, a run of
  82. lines with nothing concrete in them, and no loop back to the first line.
  83. ```
  84. BEAT SHEET · 73 words · ~26.6s at 165 wpm · target 30.0s
  85. ========================================================================
  86. 0:00.0 2.9s HOOK I lost $18,000 because of one missing contract.
  87. 0:02.9 14.6s MID Here is the exact clause I now put in every single...
  88. ^ 14.6s on one beat
  89. 0:17.4 2.9s Section 4. Payment on delivery, not on approval.
  90. 0:20.4 2.9s Approval is a feeling. Delivery is a date.
  91. 0:23.3 3.3s CTA That one word change is worth $18,000 to me.
  92. ------------------------------------------------------------------------
  93. - Beat 2 runs past 4s. Either split the line or change what is on screen
  94. inside it. A static frame is where people leave.
  95. - Loops: the last beat repeats "$18,000" from the hook. Second watches are
  96. free reach.
  97. - 3.5s under target. Either add 9 words or shoot it short. Short is usually
  98. right.
  99. ```
  100. ### The caption
  101. ```bash
  102. python3 caption.py caption.txt --keywords "client contracts,freelance pricing"
  103. ```
  104. Instagram gives a caption about 125 characters in the feed and hides the rest
  105. behind a tap. Almost every caption that fails, fails there. So the first thing
  106. this prints is that window, as a box, the way a scrolling stranger reads it:
  107. ```
  108. WHAT THE FEED SHOWS
  109. +------------------------------------------------------+
  110. | I lost $18,000 because of one missing clause in a |
  111. | contract, and the worst part is that I had read the |
  112. | thing twice before I si |
  113. +-------------------------------------------- ... more +
  114. PASS LENGTH 389 / 2200 characters
  115. WARN FIRST LINE 133 characters, so it gets cut at 125 mid-thought
  116. PASS HOOK IS CONCRETE 2 numbers or names in the visible window
  117. PASS HASHTAGS 3 tags: #freelance #contracts #agencyowner
  118. PASS ONE ASK one call to action: comment a keyword
  119. WARN SEARCH TERMS 1/2 present. Missing: client contracts
  120. ```
  121. It enforces the current hashtag cap, which is **five per post**, not thirty.
  122. Instagram cut it on 18 December 2025.
  123. ### The humanizer
  124. ```bash
  125. python3 humanize.py draft.txt --report # clean it, show every change
  126. python3 detect.py draft.txt # score it, five checks
  127. python3 detect.py before.txt after.txt # prove the delta
  128. ```
  129. **What comes out automatically:**
  130. - **Invisible characters.** Zero-width spaces and joiners, word joiners, soft
  131. hyphens, byte-order marks, Unicode tag characters, invisible separators,
  132. non-breaking and narrow spaces. Your keyboard does not make these. They
  133. survive copy-paste and they are invisible in every editor you own.
  134. - **Typography.** Em dash to comma, en dash to hyphen, curly quotes to
  135. straight, ellipsis to three dots, and the orphaned punctuation that leaves.
  136. - **The lexicon.** 154 stock words and phrases with plain-English replacements.
  137. The last block of it is Instagram-specific: "stop scrolling", "in today's
  138. video", "follow for more", "tag someone who needs this", "the algorithm
  139. loves", "run don't walk". It lives in
  140. [`slop.json`](skills/ig-human/slop.json) and it is meant to be edited.
  141. **What gets flagged instead of fixed:** "It's not just X, it's Y", rule-of-three
  142. triads, the video preamble, emoji bullet lists, three shouted words in a row,
  143. hashtag walls, reflex follow bait, uniform sentence length. Changing the shape
  144. of a sentence needs judgement, so those come back for a rewrite rather than
  145. getting mangled by a regex.
  146. Run against a caption written to be as bad as possible:
  147. ```
  148. BURSTINESS ###################..... 78.4
  149. SPECIFICITY #############........... 53.6 3.4 concrete markers per 100 words
  150. SLOP DENSITY ........................ 0.0 20 stock terms, 23.0 per 100 words
  151. FINGERPRINT ################........ 68.7 1 em dash
  152. VOICE #################....... 70.0 9 structural tells
  153. ------------------------------------------------------------
  154. HUMAN SCORE ########................ 32.5 FLAGGED
  155. ```
  156. After `humanize.py`, with the flagged structures still unrewritten:
  157. ```
  158. HUMAN SCORE ###################..... 79.7 PASS (+47.2)
  159. ```
  160. ### The swipe file
  161. ```bash
  162. python3 swipe.py captured.tsv --out ~/.claude/instagram/swipe.md
  163. ```
  164. Raw views are not evidence. A 2,000,000-follower account doing 400,000 views
  165. had a quiet Tuesday. A 4,000-follower account doing 400,000 views found
  166. something. `swipe.py` ranks on the multiple over each account's own median,
  167. names the hook formula, and prints what separates the top third from the
  168. bottom third.
  169. ```
  170. SWIPE FILE · 4 reels · 4 accounts · baseline: account median
  171. ==============================================================================
  172. 60.0x hook 57 #9 The Steal @c 180,000
  173. "steal this four line follow up it took me two years"
  174. 37.5x hook 86 #3 Nobody Tells You @a 412,000
  175. "nobody tells you that your first 30 reels are supposed to flop"
  176. 1.3x hook 13 - unclassified @b 1,200,000
  177. "in this video I am going to show you my morning routine"
  178. ```
  179. ## What I actually measured, which is the part worth reading
  180. I did not want to publish a hook scorer on the claim that it feels right, so I
  181. tested it. The corpus is **74 real short-form hooks**: the first three seconds
  182. of the auto-caption track from the top eight and bottom eight performing shorts
  183. on each of five channels, view counts from 931 to 550,000.
  184. They are YouTube Shorts rather than Reels, because Instagram does not hand you
  185. view counts you can collect without logging into somebody's account, and
  186. scraping it would violate the Terms this repo tells you not to violate. The
  187. hook grammar is the same and the sourcing is public. That is a real limitation
  188. and it is stated here rather than buried.
  189. **Three results, two of them uncomfortable:**
  190. **1. It catches bad hooks well.** Against ten hooks written deliberately badly,
  191. AUC 0.83, and nine of the ten scored below the median of the real corpus. If
  192. your hook opens on a greeting, a preamble or nothing concrete, this tells you.
  193. **2. It does not pick winners.** Separating a good creator's hits from that
  194. same creator's misses: **AUC 0.56, where 0.50 is a coin flip.** Of the five
  195. checks, only SPECIFICITY separated the bands meaningfully, by 60 points of
  196. median. STAKES and ADDRESS had identical medians in both bands, which means on
  197. this corpus they measured nothing.
  198. So the honest use is: kill the obviously weak hooks before you shoot them, then
  199. trust your own retention graph. Nothing that reads text can tell you which of
  200. two decent hooks will travel, because that is decided by your face, your edit,
  201. your audio and who Instagram shows it to.
  202. **3. The formula classifier was broken and the test is what caught it.** The
  203. `match` regexes in `hooks.json` were written off my own templates, and they
  204. named **8%** of real hooks. People do not speak in templates. Rewriting them
  205. against actual transcribed speech took it to **49%**, and four formulas went
  206. into the set because they kept appearing and were not there: Contrarian Flip,
  207. The Statistic, The Reveal, Someone Else's Result, plus The Superlative. On the
  208. other 51% it abstains, which is correct: a lot of short-form is podcast clips
  209. that have no hook formula at all.
  210. One fixed bug worth naming: `SPECIFICITY` only counted digits, so "zero
  211. dollars" and "three marketing books" scored as having nothing concrete in them.
  212. Spoken hooks say their numbers out loud.
  213. If you re-run this on a bigger or cleaner corpus and get a different answer, I
  214. would rather know. The measurement script is not in the repo because it depends
  215. on `yt-dlp`, but the method is four lines and is written out in
  216. [`/ig-viral`](skills/ig-viral/SKILL.md).
  217. ## The fine print, which is the honest part
  218. **These skills do not post to Instagram.** There is a real Content Publishing
  219. API for Professional accounts, and it needs a Meta developer app, a linked
  220. Page, a long-lived token and app review, which is not a thing a skill can hand
  221. you. Everything else people use to automate posting, commenting, following or
  222. DMing is browser automation or a third-party tool, and both violate
  223. [Instagram's Terms of Use](https://help.instagram.com/581066165581870) and get
  224. accounts action-blocked. So every skill here ends the same way: a copy-ready
  225. block, and you post it. That is not a limitation bolted on afterwards, it is
  226. the design, and it is why the approval gate is real rather than a setting.
  227. The one exception is keyword auto-replies in DMs, which Instagram supports
  228. through its own tools and approved partners, and which only fire after somebody
  229. comments first. `/ig-dm` says where that line is.
  230. **`/ig-viral` reads, it does not scrape.** Ten accounts, a dozen reels each, at
  231. human speed, with you driving your own browser. It never asks for your password
  232. and never logs in as you. Automated collection at volume is the thing that gets
  233. accounts restricted, and a crawler is not what this is.
  234. **The five detection checks are local heuristics, not detector APIs.** They are
  235. modelled on the signals public detectors key on and they run entirely on your
  236. machine. They are not GPTZero, Originality, Copyleaks, Winston or Turnitin,
  237. they do not call those services, and they cannot promise those verdicts. Fixing
  238. what they measure tends to move those numbers, because they are measuring the
  239. same underlying things. That is the whole claim. Nobody can honestly sell you
  240. "undetectable", and anybody who does is selling you something.
  241. **The invisible-character pass is real and it is narrow.** It removes the
  242. zero-width and format characters that end up in generated text and survive a
  243. copy-paste. That is a genuine, checkable fingerprint. It is not a claim about
  244. defeating a cryptographic watermarking scheme, and this repo does not make one.
  245. **Nothing here fabricates.** No invented metrics, clients or outcomes go under
  246. your name. If a draft needs a number you have not given, it comes back with
  247. `{{your number}}` in it and a flag, every time.
  248. **Platform numbers go stale.** The hashtag cap moved from 30 to 5 in December
  249. 2025 while this repo was being written, and the linter had the old number in it
  250. until the fact got checked. If something here contradicts what Instagram is
  251. doing when you read it, Instagram is right.
  252. ## Files
  253. ```
  254. skills/ig-reel/hooks.json 26 hook formulas: template, example, on-screen line,
  255. what it is for, how it gets ruined, and a match regex
  256. skills/ig-reel/hookscore.py the five-property hook panel
  257. skills/ig-reel/beats.py script to timed beat sheet
  258. skills/ig-caption/caption.py the truncation preview and the caption linter
  259. skills/ig-human/slop.json the lexicon: 154 terms, 18 invisible classes, 16 tells
  260. skills/ig-human/humanize.py the three cleaning passes
  261. skills/ig-human/detect.py the five-check panel
  262. skills/ig-viral/swipe.py outlier ranking and formula classification
  263. skills/ig-profile/rubric.json the 100-point profile score
  264. templates/voice.md your voice profile. Fill this in first.
  265. ```
  266. ## Credit
  267. Made by Jake Schincariol, [opusjake.ai](https://opusjake.ai).
  268. Sibling repo, same idea for a different platform:
  269. [linkedin-agent-skill](https://github.com/Jakeschincariol/linkedin-agent-skill).
  270. ## License
  271. MIT. Take it, change it, ship it.

Ratings and comments

Sign in to rate this skill.

Sign in

Free to use. If it’s helped, pass it on.