ig-human
/ig-human
Usuwa ślady AI z tekstu: długie myślniki, typowe słowa AI i niewidoczne znaki znaczników wodnych, a przed publikacją ocenia go w pięciopunktowym panelu detekcji.
- Autor
- nategoldai ↗
- Repozytorium
- instagram-agent ↗
- Licencja
- MIT
- Ostatnia zmiana repozytorium
- 07.10.2026
Ocena i bezpieczeństwo
- Pliki w paczce
- 4
- Skrypty
- 2
- Komentarze HTML
- 0
- Adresy URL w SKILL.md
- 0
Brak sygnałów. Repozytorium zmienione 07.10.2026, bez odwołań do wycofanych modeli i API.
MIT, otwarta: można używać i zmieniać, z zachowaniem informacji o autorze.
- 2 skrypty w paczce. Przejrzyj je przed instalacją, bo Claude może je uruchomić.
Skrypty: detect.py, humanize.py
Co robi i kiedy
Usuwa ślady AI z tekstu: długie myślniki, typowe słowa AI i niewidoczne znaki znaczników wodnych, a przed publikacją ocenia go w pięciopunktowym panelu detekcji.
Wyzwalacze z opisu
Oryginalny opis (EN)
Remove the machine fingerprint from any draft - em dashes, AI slop words, invisible watermark characters - and score it on a five-check detection panel before it goes out. Use whenever text needs to sound human, when the user says humanize, "does this sound like AI", "remove the em dashes", "de-slop this", "this sounds like ChatGPT", or before showing the user any caption, script, comment, reply or DM.
Jak działa
- Why this matters more on Instagram than you'd think
- What gets fixed automatically
- What does NOT get fixed automatically
- The five checks
- Be honest about this
- Order of steps
Podgląd SKILL.md
--- name: ig-human description: >- Remove the machine fingerprint from any draft - em dashes, AI slop words, invisible watermark characters - and score it on a five-check detection panel before it goes out. Use whenever text needs to sound human, when the user says humanize, "does this sound like AI", "remove the em dashes", "de-slop this", "this sounds like ChatGPT", or before showing the user any caption, script, comment, reply or DM. --- # ig-human There are two tools in this folder and both of them really run. Use them. Don't judge this by eye. ```bash python3 humanize.py draft.txt --report # clean it, show what changed python3 detect.py draft.txt # score it, five checks python3 detect.py before.txt after.txt # show the improvement ``` Both read `slop.json`: 154 stock words and phrases with plain-English swaps, 18 classes of invisible character, 11 typographic substitutions and 16 structural tells. The last block of each list is Instagram-specific, the words that only turn up in captions and voiceovers. It's meant to be edited. If the user has a word they always say that the list strips out, remove it from the file. ## Why this matters more on Instagram than you'd think Captions are short and scripts are spoken. One written-sounding line in a 600-character caption is a bigger chunk of the text than the same line in an essay, and a voiceover nobody could say naturally gives itself away on the first take. The giveaway isn't a detector flagging the post. It's someone scrolling past something that sounds like a brand, or a creator tripping over their own script. ## What gets fixed automatically **1. Invisible characters.** Zero-width spaces and joiners, word joiners, soft hyphens, byte-order marks, Unicode tag characters, invisible separators, non-breaking and narrow spaces. Keyboards don't type these. They survive copy-paste, no editor shows them, and they're the most mechanical trace in generated text. `humanize.py` removes all of them, plus any other Unicode format character it doesn't have a name for. **2. Typography.** Em dash to comma, en dash to hyphen, curly quotes to straight, ellipsis to three dots, bullet character to hyphen. The em dash pass is the important one: it turns the dash into a comma, then tidies up the double punctuation and stray periods that leaves behind. **3. The slop list.** delve, leverage, robust, seamless, crucial, testament to, "in today's fast-paced world", plus the Instagram block: "stop scrolling", "in today's video", "follow for more", "tag someone who needs this", "the algorithm loves", "run don't walk". Each one gets swapped for a plain word or removed, with capitals kept and URLs left alone. ## What does NOT get fixed automatically Structural tells get **flagged, not rewritten**, because reshaping a sentence takes judgement: - "It's not just X, it's Y" and "not only X but also Y" - Groups of three - One-word rhetorical question lines: "The result?" - The video intro: "in this video I'm going to show you" - Emoji bullet lists - Three or more shouted words in a row - Walls of hashtags - Reflex bait: "follow for more", "tag someone who", "double tap if" That list is on you. Rewrite every flagged line by hand, keep the meaning, then run `detect.py` again. This is what moves the score from REVIEW to PASS, and it's the part no script can do. ## The five checks `detect.py` scores five signals from 0 to 100, where higher is more human: | check | what it measures | machine-like looks like | | --- | --- | --- | | BURSTINESS | variation in sentence length | every sentence the same length | | SPECIFICITY | numbers, names, concrete details per 100 words | abstract nouns, no figures | | SLOP DENSITY | list hits per 100 words | stock vocabulary | | FINGERPRINT | invisible chars, em dashes, curly quotes per 1k chars | typographically perfect | | VOICE | contractions, person, structural tells | no contractions, staged reveals | The verdict is 60% the average and 40% the **single weakest check**, because one bad signal is enough to give it away. PASS needs 70+ overall and no check under 55. ## Be honest about this These are five local heuristics based on the signals public detectors look for. They run entirely on the user's computer and nothing is uploaded. They are **not** GPTZero, Originality, Copyleaks, Winston or Turnitin, they don't call those services, and they can't guarantee those results. Fixing what they measure usually does move those scores, because they're measuring the same underlying things. That's the claim. Don't make a bigger one for the user, and never tell them their text is undetectable. ## Order of steps 1. `humanize.py draft.txt -o clean.txt --report` 2. Read the structural flags. Rewrite those lines yourself. 3. `detect.py draft.txt clean.txt` to show before and after. 4. If it's not PASS, fix the weakest check named in the output and go again. Two rounds is normal. Five means the draft was written to a formula, and the answer is a new draft, not more rounds. 5. Show the user the cleaned text and the score together. Never just the score.
Pliki w paczce (4)
- SKILL.md 5 KB
- detect.py 8 KB
- humanize.py 11 KB
- slop.json 21 KB
Wersje
| Repozytorium | Wynik | Licencja |
|---|---|---|
| Ta strona · nategoldai/instagram-agent | 100 | MIT |
| ma-91-02/babili-instagram-agent-multilingual | 100 | MIT |
| jakeschincariol/instagram-agent-skill | 100 | MIT |
Migawka bazy z 13.11.2026 · wersja paczki e4f4f4f11bed (skrót SHA-256 całej paczki)
Pobrania w ostatnich 30 dniach: 0 · ranking
Zgłoś problem z tym skillem: szkodliwa treść, błędny opis albo prośba autora o usunięcie.
Podobne skille
- 100li-human
Usuwa ślady AI z tekstu: długie myślniki, typowe słowa AI i niewidoczne znaki wodne, a potem ocenia tekst w panelu pięciu testów wykrywania. - 100x-human
Usuwa z szkicu maszynowe ślady, takie jak długie myślniki, słowa typowe dla AI i niewidoczne znaki, i ocenia tekst przed publikacją. - 100im-human
Usuwa ze szkicu ślady pisania przez maszynę, takie jak wzorce AI i niewidoczne znaki, i ocenia tekst przed wysłaniem. - 100li-human
Usuwa z szkicu maszynowe ślady: myślniki, banalne słowa i niewidoczne znaki, a potem ocenia tekst w pięciu sprawdzeniach. - 100instagram-humanizer
Redaguje teksty na Instagram napisane przez AI: usuwa typowe zwroty i niewidoczne znaki, a potem ocenia, czy brzmią jak pisane przez człowieka. - 100fb-human
Usuwa z szkicu posta na Facebooku ślady AI: znaki niewidoczne, myślniki, wytarte zwroty i wyrażenia wabiące, a potem ocenia wynik.