agentleFS
agiea-brainagiea-brain/agency-skills/skills/chris-dubois/punch-up/SKILL.md

punch-up

Skill1 starsChanged 17 days agoMITView on GitHub

Use this skill when a piece of short-form copy (headline, subhead, hero, CTA, subject line, tagline, ad line, bio, or short section) is wordy, soft, or forgettable. Produces a sharper rewrite, one or two alternates on a different angle, and a score against a six-dimension rubric, revised until it passes. Also use for "/punch-up", "make this punchier", "make this catchier", "tighten this", "this is too wordy", "cut this down", "sharpen this headline". Not for the body of a blog post or newsletter, and not for writing a page from scratch.

The file

We split this skill into the rules it gives an agent

11 rules. About 0 are common, 11 are distinctive, and 0 are very specific to this project.

  • ≈0

    Common · estimate

    Many files like this have a rule of this kind.

  • ≈11

    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:

    Check the length. If it's the body of a blog post or newsletter, say so: long-form copy should flow, and this would chop it.

    line 22
  2. Distinctive:

    If the client has a brand voice document, read it first.

    line 23
  3. Distinctive:

    Write one or two alternates on a different angle, each with a different toolbox pattern.

    line 28
  4. Distinctive:

    MUST say it straight before saying it great.

    line 55
  5. Distinctive:

    MUST score in a separate pass from drafting, and run the hard gates before the dimensions.

    line 56
  6. Distinctive:

    MUST pass at 48/60 with no dimension below 6 and Hook Strength at 7 or higher.

    line 57
  7. Distinctive:

    MUST stop at three revision cycles and name what is blocking.

    line 58
  8. Distinctive:

    NEVER fabricate a number, name, or claim, including claims about the reader.

    line 59
  9. Distinctive:

    NEVER present a failing rewrite as finished.

    line 60
  10. Distinctive:

    NEVER carry a number from the calibration pairs into real copy.

    line 61
  11. Distinctive:

    NEVER use this on long-form body copy.

    line 62

≈ “Common” is an estimate: at least 1 in 5 of the 372,429 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

Applies to short-form copy that has to land in one read. Produces a punched rewrite, alternates, and a passing score sheet, or a clear statement of the missing fact blocking the pass.

Verbose copy is a confidence problem: every extra word is the writer hedging against being wrong. Decide what the one claim is, state it in the fewest words that still land, and prove it fast. First say it straight, then say it great. Never skip straight.

The play

  1. Check the length. If it's the body of a blog post or newsletter, say so: long-form copy should flow, and this would chop it. Offer to sharpen its headline or opening line instead.
  2. If the client has a brand voice document, read it first. The rubric stays the same; the register follows the client.
  3. Name the job. What should this piece make the reader do or feel, and in what medium? If the user didn't say, infer it and state the assumption. Pull the word budget from references/rubric.md, using the closest row.
  4. Diagnose. Name the bloat patterns present and quote the offending words, in one or two lines. The patterns are in references/bloat-patterns.md.
  5. Say it straight. State the one claim in plain English. Never invent a claim the source doesn't support. If the copy has no real claim, that is the finding: say so at the top, ask for the missing fact, and still give a best attempt marked as failing.
  6. Say it great. Rewrite with one pattern from references/hook-toolbox.md, never three stacked. Cut 30%, then restore the one line you miss (on a single line, the one word you miss). Before the first rewrite, read references/calibration-pairs.md for the size of the cut.
  7. Write one or two alternates on a different angle, each with a different toolbox pattern. A synonym swap doesn't count.
  8. Score the primary in a separate pass from drafting, using references/rubric.md: hard gates first, then the six dimensions. Score pessimistically, take the lower number when torn, and quote the offending text for any dimension below full marks.
  9. Revise until it passes. Fix gates first, then the lowest dimensions, then re-score everything, because cutting for economy often breaks the hook. Maximum three cycles.
  10. Still failing after three cycles means the straight message is wrong or a real specific is missing. Say which, and ask for the number or the moment.
  11. Deliver in this format:
DIAGNOSIS: 1-2 lines, the bloat patterns found, quoted
PRIMARY: the punched rewrite
ALTERNATES:
A) different angle - toolbox pattern used
B) different angle - toolbox pattern used
PUNCH SCORE: __/60 (Hook __, Economy __, Concreteness __, Claim __, Rhythm __, Single __) PASS or FAIL
GATES: clean, or the gate failed and the words that failed it
BIGGEST CHANGE: one line
BLOCKING (only on a fail): the one line or missing fact stopping the pass

What good looks like

  • First thing to check that most people skip: whether there is a claim at all. Most bloated copy is padding wrapped around nothing a reasonable person could disagree with, and no amount of cutting fixes that.
  • The mediocre version: the same sentence with a third of the words removed and three hook tricks stacked on top. It's shorter and still says nothing, and the alternates are synonym swaps of the primary.
  • The output is right when it reads aloud in one clean pass, a stranger would stop on line one, every number traces to the source, and it scores 48/60 or better with Hook Strength at 7 or higher.
  • Labels and buttons can get longer and still be better. "Learn More" losing to "Get the pricing math" is a punch-up.

Rules

  • MUST say it straight before saying it great.
  • MUST score in a separate pass from drafting, and run the hard gates before the dimensions.
  • MUST pass at 48/60 with no dimension below 6 and Hook Strength at 7 or higher.
  • MUST stop at three revision cycles and name what is blocking.
  • NEVER fabricate a number, name, or claim, including claims about the reader. Leave a placeholder and surface the gap.
  • NEVER present a failing rewrite as finished. Deliver the best draft, its score sheet, and the blocking line.
  • NEVER carry a number from the calibration pairs into real copy.
  • NEVER use this on long-form body copy.

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.