agentleFS
EvolutionUnleashedEvolutionUnleashed/elevate/SKILL.md

elevate

Skill7 starsChanged 15 days agoMITView on GitHub

Use when someone wants a piece of work to be better and hasn't said how, such as "make it better", "improve this", "elevate it", "polish this", "punch it up", "make it pop", "level it up", "it's meh", "boring", "generic", "flat", "reads like AI", "looks like a template", "not quite there", "I don't love it", "try again" on a draft, or a score like "6/10". Covers copy, posts, emails, headlines and names, websites and app screens, slides, images and image or video prompts, and documents, alongside any skill that builds the file. It finds what is actually wrong and raises the work instead of piling more on. Not for a change they've already named (shorter, more formal), a retry after an error, fiction, translation, prompts for AI agents, or code quality and bugs.

4 other projects use a version of this.

Adopted by

Most recently taken up by

  • Other versions4 projects use a version of this.

    Word for word · 0

    None found.

    Edited · 0

    None found.

    Wrote their own version · 4

    1. BenjaminG/ai-skills★ 4same name
    2. catcatcatstudio/cat-skills★ 3same name
    3. welshofer/elevate★ 2same name
    4. alanvaa06/award-craft★ 1same name

    How matches are found. Word for word: the exact same text. Edited: shares passages with this file (“99% same” means 99% of its five-word phrases also appear here; 10% or more counts). Their own version: similar meaning, or, weakest, the same name. We can’t tell which project had it first. Projects are counted among the 200 most-starred candidates, and mentions are links to the repository we found, up to 50 per site, so both counts are a floor.

  • In this repositoryOne other file it gives its agents.

The file

Show its rules and text

We split this skill into the rules it gives an agent

14 rules. 0 are common, 14 are distinctive, and 0 are very specific to this project.

  • 0

    Common

    Most of the 4 edited versions we read keep these.

  • 14

    Distinctive

    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:

    When someone says "make it better" and nothing more, they've felt the work is ordinary and can't yet say why.

    line 10
  2. Distinctive:

    Read the request against the work and decide which kind of better this is.

    line 28
  3. Distinctive:

    Every example in these files is one answer, not the answer.

    line 53
  4. Distinctive:

    Keep what they liked. Anything they praised stays.

    line 61
  5. Distinctive:

    Read the word they used. It's the best clue you have.

    line 63
  6. Distinctive:

    Check both ditches, below.

    line 103
  7. Distinctive:

    Test it and look at it. Run the tests below.

    line 105
  8. Distinctive:

    What-is-it: After the first screen, slide or couple of lines, can a stranger say what this is, who it's for and what to do next?

    line 121
  9. Distinctive:

    Cut: Delete each sentence that tells the reader what they already know or could work out.

    line 123
  10. Distinctive:

    Aloud: Read the words out. Has anyone ever said that sentence?

    line 125
  11. Distinctive:

    Keep the old version. Write the new one beside it, or make sure version control holds the old one, so the comparison stays possible.

    line 135
  12. Distinctive:

    Ask before spending money.

    line 137
  13. Distinctive:

    Never imply a check you didn't run.

    line 138
  14. Distinctive:

    When another skill builds the file (a deck, a document, a site), use it alongside this one.

    line 139

“Common” is measured: kept by at least 6 in 10 of the 4 edited versions of this file we could read. “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
<!-- Generated from skills/elevate/SKILL.md by scripts/sync_root_skill.py. Edit the source, then run the script. -->

Elevate

When someone says "make it better" and nothing more, they've felt the work is ordinary and can't yet say why. Your job is to see what they're feeling, name it in plain words, and fix that.

The reflex is to polish: more words, more adjectives, an extra section, a gradient where there wasn't one. That reflex is how ordinary work gets worse. Raising work rarely means adding more of what's already there. It means finding the default the work fell into and moving it up a level.

Describing, and the level above it

The most common reason work feels ordinary is that it describes. It tells the reader everything, so they have nothing left to bring. Watch the levels:

  1. "Our wood-fired pizzas combine fresh ingredients with hand-stretched dough for an authentic Italian experience." This describes the pizza.
  2. "Spend less time cooking and more time enjoying delicious food with the people you love." This describes the outcome, which is where most rewrites stop.
  3. "You've cooked all week. We've got tonight." Nothing is explained. The reader recognises the tiredness, then the relief, and finishes the thought themselves. Because they filled it in, it feels written about them.

Room is for the reason to care, never for what the thing is. A reader who finishes the thought still has to know what's on offer.

The same ladder runs through every medium. A web page climbs from the template, to the template restyled (most "better" passes stop there), to a page whose structure comes from how the subject actually works. A picture climbs from the stock metaphor, to the stock metaphor beautifully rendered, to a picture built on the idea itself. Each reference file shows its own ladder.

First, size the job

Read the request against the work and decide which kind of better this is. Everything after this depends on it.

  • They named a change you could measure (shorter, more formal, cut the intro, fewer slides). Do exactly that, inside the current angle, and stop.
  • An earlier note never got applied. Apply it, say which, and stop.
  • It's close: "nearly", "better, but", a 7 or 8 out of 10. Repair it. Keep the angle and everything that works, and fix only the places that drop the level. It should still be recognisably the same piece.
  • It isn't working: "flat", "generic", "cheap", 6 out of 10 or lower, a second rejection. Find a new angle with the full method below.
  • It's already strong. Say so, make the one or two changes you'd defend, and say why the rest stays.
  • They asked what's wrong rather than for a new version. Give the diagnosis and offer the fix, without rewriting unasked.
  • There's nothing to work on, or you can't see it. Ask for it, and nothing else.

Match the size of the reply to the size of the work. A five-word line gets a better line and a sentence of why, not an essay.

Open the reference now

Load the reference for the medium in front of you before you diagnose, because the defaults you're hunting for are listed there. Mixed work loads several: a landing page needs websites, headlines and writing.

Working onRead
Copy, posts, articles, newsletters, emails, scriptsskills/elevate/references/writing.md
Headlines, taglines, names, subject lines, buttonsskills/elevate/references/headlines-and-names.md
Websites, landing pages, app screens, interactive toolsskills/elevate/references/websites.md
Slides and decksskills/elevate/references/slides.md
Images, illustrations, logos, thumbnails, image and video promptsskills/elevate/references/images.md
Documents, reports, one-pagers, proposals, plans, guides, chartsskills/elevate/references/documents.md

Every example in these files is one answer, not the answer. Don't reuse an example's look, palette, structure or wording. If two pieces in a row come out the same shape, the second has fallen into this skill's own default.

Before you touch it

Re-read the whole conversation. A bare "make it better" after earlier rounds often means one of their notes never landed. List every note they've given, check each against the current version, and fix any that slipped. That alone settles many of these requests.

Work out whose work it is. If a model made it, hunt for a model's defaults. If they wrote it themselves, their voice is the asset: keep their wording wherever it isn't the problem, describe the failure in the reader's terms ("the point arrives in paragraph four"), and treat a habit as voice unless it's the thing that's failing. If the work has to stay theirs, such as an application, coursework or a speech, show the two or three changes that matter and why, rather than handing back a rewrite.

Keep what they liked. Anything they praised stays. If a fix would change something they approved, keep it and say what that costs.

Read the word they used. It's the best clue you have.

They saidWhat it usually means
boring, flat, mehNothing at its centre, or it explains where it should land.
generic, could be anyone'sIt took the first answer anyone would reach for.
reads like AI, looks AIThe tells are showing, and underneath them it's describing.
cringe, trying too hardPerformance: hype, forced jokes, invented scenes, cleverness that needs decoding.
busy, clutteredToo many things competing, so nothing is the hero.
make it popNothing is dominant. Make one thing win on size, contrast or position. Glow, gradients and motion are the polish reflex.
more premiumEverything at one volume, cramped spacing, default type, too many colours. Use fewer elements and more space, with one detail done perfectly.
cheap, template, amateurNo brand system and no craft finish.
dated, make it modernName what dates it and fix those parts. Chasing this year's trend dates it again next year.
more professionalUsually inconsistency and errors. Fix those and keep the personality, because sanding it off makes it generic.
punchierLong run-ups and buried verbs. Lead with the claim. Fragments are the overcorrection.
doesn't sound like meTheir words got paraphrased or smoothed. Put their lines back and match their own writing.
wordy, longIt explains what the reader already knows, or says one thing several ways.
confusingThe reader has to decode something: a story before the point, jargon, a clever line, a diagram of terms they haven't met.
it's just not landingBefore blaming the style, check the audience, whether the point arrives late, whether the claim has any proof, and whether there's a clear next step.

Rejections carry information. A second outright rejection means the angle is wrong, so a third polish of the same construction wastes their time; come at it from somewhere else. "Better, but" isn't a rejection, so keep the angle and push it further. After a third miss, stop guessing. Say what you think you keep missing, then either ask one pointed question or show two short, sharply different directions (a headline and an opening, not finished pieces) and say what each one bets on.

When they dislike a part, take it out if the piece stands without it, and say so. If the piece needs it, as with a headline, a name or a call to action, replace it from a different angle rather than offering a variation of the one they disliked. "I hate it" about the whole piece is a rejection, not a deletion.

Don't hand the job back. "What would you like me to improve?" makes them do the diagnosis they came to you for. Ask one question only when you can't tell who the work is for or what it has to do, and a wrong guess would waste the round.

The method

A repair uses steps 1, 2, 5 and 7. A new angle uses all of them.

  1. Name the job in one sentence: who it's for and what it has to make them feel or do. This decides what up a level means. For work meant to move someone (marketing, posts, pitches, the front page of a site), up means leaving room. For work someone has to follow or decide from (guides, documentation, support replies, reports, read-alone decks, work email), up means they reach the answer sooner and never have to guess: answer first, steps literal and complete, every term introduced.
  2. Find the default. Name the failure in one sentence, in the work's own terms: "It describes the product instead of the moment it changes", "Every slide carries the same weight", "The ask is hidden in the last paragraph". If you can't say it in a sentence, you haven't found it yet.
  3. Move it up. For persuasion, that means getting out of the describing layer: hand the reader something they already know and let them finish the thought. What it is and what it does still arrive, just not first. Swapping the theme while still describing is the same failure in new clothes.
  4. Commit. Pick the one thing the piece will be remembered for and give it room. If the work has no central idea, bring one and say why it works, because work that only removes faults comes back clean and forgettable. If it already has one, push that further rather than adding a second. Then go all the way with it, since the safe, small version of a good idea is as dull as having none.
  5. Cut what the reader already knows, what the line before already showed, and anything that's only there because pieces like this usually have it. Never cut what the thing is, why to believe it, the reasoning that connects one point to the next, or what to do next.
  6. Check both ditches, below.
  7. Test it and look at it. Run the tests below. If the work can be rendered, render it and look: a page on desktop and on a phone, a deck as a sheet of thumbnails, an image at full size.
  8. Hand it over, as described at the end.

The two ditches

Work fails in two directions, and fixing one tends to drive you into the other.

The first ditch is the default: describing, explaining, the template layout, the stock picture, the same section stacked six times, every benefit spelled out.

The second is the overcorrection: cleverness that needs decoding, riddles, fragments, a page stripped so bare it says nothing, invented scenes with days and times in them, edge for its own sake. Most anti-AI rewrites land here. They cut the explaining and leave a fact sheet, swap describing for a riddle, or ban the obvious defaults and land on the next ones.

The road between is plain and specific, and trusts the reader to finish the thought. Each reference names both ditches for its medium.

The tests

  • What-is-it. After the first screen, slide or couple of lines, can a stranger say what this is, who it's for and what to do next? This is the floor, and nothing else counts if it fails.
  • Second guess. Name the first "better" ideas anyone would reach for with this brief, such as a till receipt on a bookkeeper's page or a quiet still life with a coffee cup. If yours is one of them, it's the next default. Push it further or find another.
  • Cut. Delete each sentence that tells the reader what they already know or could work out. If the piece reads better, leave them out. If it collapses, the idea was weak and the explaining was propping it up.
  • Stranger. A cold reader, once, at speed. Do they get it without help? For work someone has to follow, ask instead whether a first-timer could act on it alone.
  • Aloud. Read the words out. Has anyone ever said that sentence? "Stress-free bookkeeping for busy business owners" is fine word by word, and nobody has ever said it out loud. For formal work, ask whether a careful professional would write it.
  • Side-by-side. On a new angle, put the old and the new next to each other. If one could be mistaken for the other, you restyled it. On a repair, the differences should sit exactly where you diagnosed and nowhere else.

Guardrails

  • Change the angle, never the facts. Never invent a statistic, testimonial, client, quote, promise or result.
  • Judge a claim by where it came from, never by how it looks. Anything the person wrote, supplied or confirmed is theirs, so it stays, absolutes included. Proof a model or template produced (round-number stats, a logo strip, testimonials that could be about anyone) isn't a fact yet, so it comes out. If you can't tell where a claim came from, leave it in.
  • Whatever they need to check goes at the top of your note: proof you took out, claims you couldn't place, absolutes that need backing, and any new line in their voice about what they've seen or felt.
  • Work in the piece's own language, register, spelling and typography. The tells in these files are English defaults, and formal registers stay formal.
  • Stay inside their constraints: brand rules, length limits, legal lines and the platform's format.
  • Keep the old version. Write the new one beside it, or make sure version control holds the old one, so the comparison stays possible.
  • In a codebase, raise the work inside the codebase. Use its framework, components, tokens and conventions, keep the change to what the diagnosis needs, then run the build and look at the result. Don't switch stacks or libraries to get a new look.
  • Ask before spending money. Generate several images or videos only when it's free or they've asked; if a step costs credits, write the prompt and check first.
  • Never imply a check you didn't run. If you couldn't search, render or test, say so.
  • When another skill builds the file (a deck, a document, a site), use it alongside this one. This skill decides what better is, and that one makes the file.

Handing it over

Lead with the work. If anything needs checking, that list opens your note. Then add one or two sentences in the work's own terms about what was failing for the reader and what changed, such as "Your three asks were buried in the last paragraph. They're in the second line now, each with its reason." or "Every slide had a topic for a title. The titles now make the argument on their own, and the detail moved into your speaker notes."

Keep this skill's own vocabulary out of the note (default, level, room, ditch, describing) unless they use those words themselves. Don't list the tests or the notes you checked, don't open with "Here's an elevated version", and don't reuse last round's sentence shape. Keep elevate, elevated, unlock and unleash out of the work itself, since they're the filler this skill exists to remove.

This method is for the round where they asked for better. Once they approve the work or move on to something else, drop it, along with the closing note.

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.