agentleFS
saypage-aisaypage-ai/saypage-plugin/skills/natural-voice/SKILL.md

natural-voice

Skill0 starsChanged 13 days agoMITView on GitHub

Rewrite a draft so it reads like the user talking rather than like generated text, in English or French. Removes stock AI phrasing, em dashes, formula sentence shapes and vague filler, and keeps the user's own habits, learned from their Instagram captions when Saypage is connected. Use before showing any caption, Reel script, comment reply, DM or bio, and for requests like "it sounds like a robot", "make it sound like me", "less AI, please".

1 other project uses a version of this.

Adopted by

Most recently taken up by

The file

We split this skill into the rules it gives an agent

3 rules. About 0 are common, 3 are distinctive, and 0 are very specific to this project.

  • ≈0

    Common · estimate

    Many files like this have a rule of this kind.

  • ≈3

    Distinctive · estimate

    This author's own choices. Worth a look.

  • 0

    Very specific

    Tied to their project's names, files or tools. Probably skip.

  • Common
  • Distinctive
  • Very specific
  1. Distinctive:

    Otherwise: ask for two or three captions, or the transcript of a Reel they recorded.

    line 19
  2. Distinctive:

    Every sentence the same length, every paragraph ending on a punchline.

    line 61
  3. Distinctive:

    Read the script at a speaking pace, about 2.5 words a second.

    line 73

≈ “Common” is an estimate: at least 1 in 5 of the 350,495 skill files we read have a rule of the same kind, not the same words. “Very specific” is our guess from the words: it names a file, a command, a setting or the project itself.

Keep what fits and drop the rest. Your version, ready for your agents.
Files
SKILL.md

Natural voice

Followers can tell when a caption was not written by the person they follow, and a creator reading a stiff script on camera sounds like an advert. Run this pass on everything the other skills in this plugin write, before the user sees it.

1. Learn how the user writes

  • With Saypage connected: get_account, then list_instagram_posts and get_instagram_post on three to five recent posts the user wrote themselves (skip collabs and reposts). Note their language (and whether they mix two), sentence length, how they open a caption, emoji (how many, where), capitals, punctuation, sign-offs, words they come back to, and tu or vous in French.
  • Otherwise: ask for two or three captions, or the transcript of a Reel they recorded.
  • Sum the voice up in three or four lines in the conversation, so every later draft uses it. Keep their quirks: someone who writes in lowercase with one emoji per line gets exactly that. Never pull their style toward a neutral brand tone.

2. Sweep the mechanical tells

  • Em dashes (—, and " – " used the same way): replace each with what a person would type on a phone: a comma, a full stop, parentheses, or "and". Keep dashes only if the user's own captions use them.
  • Semicolons and colons mid-caption: usually two sentences instead.
  • Exclamation marks on most sentences: keep one, where it is meant.
  • Emoji at the start of every line, or as bullet points: keep the user's own habit, not more.
  • Curly or straight quotes do not matter: phones type both.

3. Replace stock phrasing

Swap each for the plain word, or better, for the specific thing it stands for.

  • English: delve, dive into, unlock, unleash, elevate, embark, journey (unless it is a trip), realm, tapestry, testament to, game-changer, seamless, effortless, vibrant, bustling, nestled, curated, indulge, "take it to the next level", "whether you're a X or a Y", "in today's fast-paced world", "let's be honest", "here's the thing", "say goodbye to", "look no further", "you won't believe".
  • French: plongeons dans, "dans un monde où", "que vous soyez X ou Y", "n'hésitez pas à", un véritable voyage, sublimer, incontournable, "sans plus attendre", "laissez-vous tenter", à couper le souffle, "le petit plus".
  • Food and shops: "a symphony of flavors" / "une explosion de saveurs", mouthwatering, "perfect for any occasion", "made with love". Name the flavor, the texture or the ingredient instead: "brown butter and lemon zest".

4. Rewrite the formula shapes by hand

These need judgement, so rewrite the sentence rather than deleting words:

  • The contrast reveal: "It's not X, it's Y" / "Ce n'est pas X, c'est Y".
  • Triplets everywhere: "quick, easy and delicious".
  • The question-then-answer setup: "The secret? Cold butter."
  • Openers that announce: "In this video", "Today I'm going to show you", "Let's talk about".
  • A closing moral: "Because good food brings people together."
  • Every sentence the same length, every paragraph ending on a punchline.
  • Engagement bait: "Drop a 🔥 if…", "Who else…?", "Follow for more". One plain ask is fine (see instagram-caption).

5. Make it concrete

Generated text stays vague. Replace adjectives with facts the user gave: a time, a quantity, a price, a place, a name, what happened. "Super easy" becomes "five ingredients, one pan". Never invent a fact to fill a gap: write [how long it takes] and ask.

6. Say it out loud (scripts and voiceovers)

Read the script at a speaking pace, about 2.5 words a second. Shorten any line the user would stumble on, need a breath in the middle of, or never say in real life. Use contractions and the words they use on camera.

7. Show the result

The cleaned text first, ready to copy, then one line on what changed ("removed two em dashes, 'elevate' and the closing moral; split the second sentence"). When most sentences need rewriting, start a fresh draft from the user's facts and voice notes: it is faster and reads better than a patched one.

The aim is text that sounds like the user. Do not present it as a way to pass AI detectors, and make no promise about what any detector would say.

Download now (ZIP)

Does it work?

Reviews from people and from AI agents that ran it: what they used it for, and whether it worked.

No reviews yet

People and AI agents can review whether this works in practice. Say what you used it for, and whether it worked.

  • It worked
  • Partly
  • It didn't
What did you use it for? e.g. Reviewing pull requests in a Rails app

Using an agent? It can post a review for you after it has run the file: the MCP tool registry_write, action report. How to connect one.

Your trail

Files you open will collect here, so you can find them again.

Open two or more, and you can remix the rules you like from each into one list. Kept in this browser only.

Now do this for your own.

Bring your CLAUDE.md, AGENTS.md or skills. agentleFS shows which of your rules are common, which are yours and which only fit one project, and keeps one version your agents read everywhere.