- name
SKILL.md
技能檔案保留在發布者的儲存庫中。
在終端機中貼上複製的指令。它會將這個 SKILL.md 安裝到對應的代理中。
npx skills add sergebulaev/instagram-skills --skill ig-humanizer -a cursor -y在終端機中貼上複製的指令。它會將這個 SKILL.md 安裝到對應的代理中。
npx skills add sergebulaev/instagram-skills --skill ig-humanizer -a claude-code -y在終端機中貼上複製的指令。它會將這個 SKILL.md 安裝到對應的代理中。
npx skills add sergebulaev/instagram-skills --skill ig-humanizer -a codex -y在終端機中貼上複製的指令。它會將這個 SKILL.md 安裝到對應的代理中。
npx skills add sergebulaev/instagram-skills --skill ig-humanizer -a grok -y在 Grok Bot 聊天中傳送複製的訊息。Bot 會從 GitHub 取得這個 SKILL.md,並加入你所有 Bot 共用的技能中。
Install this skill: https://github.com/sergebulaev/instagram-skills/tree/HEAD/.codex-marketplace/instagram-skills/skills/ig-humanizer來自發布者
Rewrites any caption or carousel slide text to remove the AI tells that human readers notice, and audits a finished caption against the 2026 Instagram checklist. Based on Wikipedia's "Signs of AI writing" taxonomy, the 2025-2026 stylometry literature, our own Instagram caption corpus (n=284, prevalence only), and Instagram-specific patterns (the first-125 fold, the lowercase-casual caption register, sized hashtags, sends-and-saves structure). V3 (2026-09): recalibrated on 2026 evidence. Vocabulary is scored by density, em dashes are capped instead of banned (and the cap is generous here: 29% of human captions use one), forced rhythm is now a tell instead of a fix, and there is an over-correction guard.
What this skill does not do: it does not make text "pass" GPTZero, Pangram, Turnitin or Originality. Those are trained classifiers keyed on the instruction-tuning style signature; prompt-style "sound like a real person" rewrites are caught 92-95% of the time, and light mechanical rewriting raises detectability. On caption-length text (under 300 words) detector scores are noise. The real value is elsewhere: expert human readers cite vocabulary (53%) and sentence structure (36%) as what gives AI text away, and on Instagram a caption that reads as a brand account earns neither the save nor the send. This skill removes what those readers react to.
Evidence tier in brackets: [strong] = replicated across 2+ independent 2025-2026 studies or our own corpus; [vendor] = single platform or vendor dataset; [weak] = one study or expert-panel report.
sub-skills/post-audit.md)Any text: a caption, the on-image text of a carousel, or a Reel caption. Optional: target voice samples (the user's past captions).
# Default: scrub AI tells (forensic + strict) and fix Instagram-format issues
ig-humanizer <text>
# Forensic only - minimum touch, just kill model leakage
ig-humanizer --mode forensic <text>
# Audit - detection-only pass-fail review, no rewrite
# Runs the 2026 Instagram checklist: first-125 hook, caption length, sized
# hashtags, emoji limits, CTA quality, media reminder.
# Returns Blockers + Warnings + suggested fixes. See sub-skills/post-audit.md.
ig-humanizer --mode audit <text>
# Profile - build/update the user's Voice & Brand Profile. See the section below.
ig-humanizer --mode profile
Apply the tiered catalogs in references/scrub-rules.md. The unit of
judgement is the paragraph (or slide), not the word: count markers per
paragraph, rewrite the paragraph at 3+, leave a single marker alone unless it
is a reveal bridge, negative parallelism, a sincerity marker, or forensic
leakage.
Detectors do not score burstiness, and captions are mid-length (our corpus median is well under 200 words), so rhythm is neither a reach lever nor a fix. What readers notice is the mechanical-uniformity tell (every line the same length, machine-flat) and, worse, the staged variance that second-generation humanizers add. So Pass 2 has two jobs: fix rhythm only where it reads machine-flat, and remove manufactured variance everywhere. It never adds variance as a tactic.
The check is "does any paragraph read machine-flat, and did I add a staccato pattern", not a variance number.
Require where the content allows:
Forbidden as openers or pivots (sincerity announcements, a named 2026 tell): "let me be honest", "I'll be real", "honestly?", "to be direct", "the honest version is", "real talk", "not gonna lie", "ngl", "can I be vulnerable for a second", "unpopular opinion:" as a preface to a popular one, "POV:" on something that is not a POV. Also forbidden as insertions: hedges the author did not write ("perhaps", "I might be wrong but", "it seems"). Performed hesitancy is 2x more common in LLM text than in expert human text; adding it makes the draft read more AI, not less.
If the input lacks these, ask the user for a number, name, or moment. Do not fabricate.
Humanizer output has its own fingerprint. Before returning, re-read the result once and answer three questions:
(a) Did Pass 2 create staccato stacks, "The result?" reveal bridges, one-word lines, or a long/short/long/short seesaw? If yes, merge the fragments back into full sentences. (b) Did Pass 3 add a framed confession, a sincerity announcement, or a hedge the author never wrote? If yes, strip the frame and keep only the flat fact, or remove the insertion. (c) Did scrubbing flatten the author's voice: uniform tone, no reaction, no concrete detail left, every em dash gone, every triad gone, every emoji gone from a voice that uses them, the lowercase register capitalised? If yes, restore what the author had. Zero em dashes and zero triads in a long caption is a tell in its own right.
If any answer is yes, dial back rather than scrub harder. Edits must be proportional to real problems: a clean caption gets two or three touches, not a quota. When in doubt whether a pattern is the author or the model, leave it.
Global voice rules: see root SKILL.md Voice rules. Additional skill-specific
rules (V3):
.. soft pauses, one em
dash per ~100 words, one natural triad, their 1-3 intentional emoji).See references/examples.md for worked before/after rewrites.
SKILL.md - this file (rewrite scrubber + audit-mode entry)references/scrub-rules.md - V3 regex patterns by tier, density scoring, em dash cap, rhythm rules, forbidden insertionsreferences/examples.md - worked before/after rewrites for captions and slidesreferences/audit-checklist.md - the pre-publish checklist with thresholdssub-skills/post-audit.md - pre-publish audit workflow (detection-only, no rewrite)sub-skills/voice-profile.md - build/update the user's Voice & Brand Profile (--mode profile)sub-skills/illustration.md - optional Pixfaro image workflow--mode profile)ig-humanizer --mode profile builds or updates the user's Voice & Brand Profile at ../../references/voice-profile.md from 3-6 of their real Instagram posts pasted in (portable, no token) or, if a read token is set, from pulled activity. Once filled, every writing skill in this bundle drafts in the user's voice automatically. See sub-skills/voice-profile.md. Triggers: "build my voice profile", "learn my voice".
ig-caption-writer - generates captions that already pass the humanizerig-carousel-planner - generates carousels that already pass the humanizerRemove the AI tells readers react to in an Instagram caption or carousel slide: 2026 vocabulary by density, reveal bridges, staccato stacks, stacked triads, performed sincerity, emoji storms; caps em
---name: ig-humanizerdescription: Remove the AI tells readers react to in an Instagram caption or carousel slide: 2026 vocabulary by density…---# .codex-marketplace/instagram-skills/skills/ig-humanizer/SKILL.md總覽
Remove the AI tells readers react to in an Instagram caption or carousel slide: 2026 vocabulary by density, reveal bridges, staccato stacks, stacked triads, performed sincerity, emoji storms; caps em dashes. Includes --mode audit (first-125 hook, length, hashtags, emoji, CTA, media) and --mode profile. Not for beating AI detectors (no edit reliably does). Not for writing from scratch (use ig-caption-writer or ig-carousel-planner). Keywords: humanize, de-AI caption, audit before posting.
常見問題
從發布者下載 .codex-marketplace/instagram-skills/skills/ig-humanizer/SKILL.md,或複製路徑後開啟儲存庫。PlugMCP 只收錄這項技能,不代管檔案。
發布者將它存放在 https://github.com/sergebulaev/instagram-skills。
Claude、Cursor、Codex 和 Grok Build 可以讀取你從發布者儲存庫加入的 SKILL.md。在聊天中傳送技能的 GitHub 資料夾連結,Grok Bot 就會安裝它。