── SKILL
cut-the-fluff✓ APPROVED

Cut a bloated doc to its verdict-first core.

Aggressive editing, not summarizing. Every sentence earns its place, rewritten in a voice you define.

Inside: The compression rules, the verdict-first restructure, and a voice pass.

01 / HOW TO USE

Install it in one line, or paste it in.

1
In Claude Code (one command)
Copy the install line, paste it into your terminal, and restart Claude Code. The skill installs itself to ~/.claude/skills/ and runs automatically when it is relevant.
2
In Claude, ChatGPT, or a Project (no terminal)
Open the file, then upload it to your chat or paste its contents in. A skill is just a markdown file of instructions, so any capable AI can follow it.
New to skills? A skill is a plain-text file that teaches your AI a workflow. Point any capable assistant at it and it follows the steps, on your data.
02 / MAKE IT YOURS

Connect your context. Set it to your motion.

SET IT TO YOUR MOTION

Set these once and the output sounds like you, not a generic editor.

What you setWhat it controlsExample
Your voiceThe tone of the rewrite"Short sentencesno hedgingfirst personadmit the mistake before the lesson."
Your audienceWho reads the output, which decides what counts as fluff"Time-poor execs" vs "frontline reps who need the how"
What counts as a receiptThe evidence a claim needs to make the top block"A specific numbera named sourceor a before/after pair."
Your kill listWords you never want in your writingEdit the default list below.
Your action verbsThe lead verbs your action items useSwap the verb table for words that sound like you giving direction.
03 / THE FULL SKILL

Everything the skill does, in full.

── WHAT THIS DOES

Cuts documents to their essential core, restructures them verdict-first, and rewrites them in a voice you define. Aggressive editing, not summarizing. Every sentence earns its place. Three jobs: compress a bloated document without losing a single number, humanize the numbers so they land as behavior, and restructure so a reader can act in five minutes. There is also an Explainer mode for documents that explain how something works.

── THE METHOD
  1. 1
    Humanizing numbers, the core move

    A number alone is data; a number attached to a behavior is a story. Apply five rules to every figure in the narrative layer:

    1. Round to human scale. About 70%, not 68.3%. The exact figure lives one click away in the evidence block.
    2. No naked numbers. Every figure rides with the behavior that produces it or the consequence it causes.
    3. Rates become frequencies. 47% becomes "about 1 in 2." 64% becomes "2 of every 3."
    4. Weekly abstractions become daily rituals. 8.8 per week becomes "fewer than 2 a day."
    5. One number per sentence. A before/after pair counts as one.

    Worked example. Before: "Inbound holds 68% of closed volume and 78% of won dollars." After: "About 7 of every 10 closed deals, and nearly 8 of every 10 won dollars, ride on one channel."

  2. 2
    What fluffy looks like
    1. Context bombing. Paragraph 1 explains why we are doing this at all. Cut it.
    2. Methodology exposition. Nobody asked how. Lead with what you found.
    3. Buried verdict. The conclusion is in paragraph 5. Move it to sentence 1.
    4. Symmetric coverage. Three items get equal treatment when one is far more important. Weight to importance.
    5. Over-qualified claims. Pick a lane.
    6. Show-your-work tables. Trim to the load-bearing columns.
    7. Same point three ways. State the finding once, with the number.
    8. Boilerplate transitions. Delete "it is worth noting," "in conclusion," "building on the above."
    9. Prose-buried lists. A sentence enumerating 3+ parallel actions. Break it into a list.
    10. Administrative metadata. "Data sources," "build sequence," appendices. Cut or move to a literal appendix.
  3. 3
    The compression protocol

    Work in sequence.

    1. Find the verdict. One plain sentence. If you cannot, the document does not know either.
    2. Find the evidence. The 2 to 4 facts that prove the verdict. Keepers.
    3. Rebuild the structure. Verdict first. Evidence second. Context last.
    4. Sentence-level cuts. Delete any sentence that does not change what the reader knows or does.
    5. Voice pass. Apply your voice rules.
    6. Formatting pass. Pull prose-buried lists into numbered lists.
    7. Thrash filter (for strategic plans). If the plan responds to more than 2 problems, it is reactive. Pick the 1 to 2 whose math moves the number.
  4. 4
    Your voice rules
    • Self-implicating first. "We built the wrong thing," not "the data showed underperformance."
    • Failure-derived insights beat generic best practices.
    • Plain declaratives over bumper-sticker aphorisms.
    • Specific numbers beat vague magnitude.
    • No bow-tied endings.

    Kill list (edit for yourself): "delve," "leverage" as a verb, "impactful," "seamlessly," "robust," "actionable insights," "it is worth noting," "in conclusion," "the data shows," rule-of-three lists where only two items matter, sycophantic openers, hyphenated filler ("data-driven," "best-in-class"), corporate-ops verbs as headlines ("activate," "operationalize," "roll out").

  5. 5
    Action verbs, intent not operations

    An action item lives or dies on its lead verb. Lead with verbs that sound like a person giving direction. The number or structural detail lives in the why line, never in the headline.

    Drop (corporate-ops) Use (direction)
    Activate Go after
    Stand up Decide on
    Operationalize Run
    Implement Build / Ship
    Roll out Launch
    Leverage Use
  6. 6
    Output contract

    Produce four things: the "If you read nothing else, do this" block (story arc in 3 lines + numbered actions + deeper reads), the compressed document, the cut list (one line per major cut), and the word-count delta.

  7. 7
    The 5-minute-read structure

    Lead with a block a reader finishes in 5 minutes and can act on: the story as an arc (3 lines), the numbered do-this list, deeper reads as links with a reason to click. Everything below is evidence, not argument.

  8. 8
    Explainer mode

    For documents that explain how something works, restructure around the reader's four questions: What is it? Where does the data come from? How does it add up? What do I do with it? Two closers are mandatory: one worked example end to end, and an execution bridge (numbered steps or copy-paste prompts).

  9. 9
    The trust contract

    For anything the reader acts on without re-deriving it, trust is the product. State what was done for them, keep caveats visible, hand over the conclusion and the next move, never the homework.

── OUTPUT (EXAMPLE)
EXAMPLE OUTPUT
Dashboard-first is no longer our headline. It is confirmed. Nearly 3 of every
4 interactions happen there. The editor sees about 1 in 3 users weekly. The
sidebar, barely 1 in 8.

The sidebar is underperforming enough to warrant a discoverability fix, not
just more onboarding. The editor needs more digging before we draw a conclusion.
One principle follows: dashboard-first, everything else secondary until the
sidebar numbers move.
── EXAMPLE CHAT
EXAMPLE CHAT
You
Cut this down: [pastes a 3-paragraph exec summary with three buried stats and a hedged close]
CLAUDE
Found the verdict in paragraph 3 and moved it to sentence 1, then stripped the naked percentages into frequencies. Before (130 words): "...usage patterns indicate that the dashboard experience sees significant engagement at 73%, while the editor interface shows moderate adoption at 34%, and the sidebar navigation..." After (80 words): "Dashboard-first is confirmed. Nearly 3 of every 4 interactions happen there. The editor sees about 1 in 3 users weekly. The sidebar, barely 1 in 8, needs a discoverability fix, not more onboarding." Cut list: deleted the methodology paragraph, killed "significant" and "moderate" (vague magnitude), turned three raw percentages into human frequencies, moved the verdict to sentence one.
You
Now do the same for the strategic plan attached. We've got six initiatives listed.
CLAUDE
Ran the thrash filter: six initiatives responding to six different problems means the plan is reactive. Cut to the two whose math actually moves the pipeline number this quarter, and moved the other four to an appendix as "considered, not funded." The insight this catches every time: a document that gives six things equal weight is hiding the fact that nobody picked. One sharpener: tell me your kill list once, the words you never want to see, and I will apply it automatically on every future cut instead of you flagging them each time.
04 / GO FURTHER

Where an operator takes this next.

The read is step one. Here is where an operator takes it once the manual version proves out.

Set your voice and your kill list once, then let the pre-send check do the rest.

01
Run it before anything ships

Wire cut-the-fluff as a pre-send check in Slack, so no doc over 300 words posts to a channel without a verdict-first pass.

02
Catch drift automatically

Point a scheduled Claude task at your Notion or Google Docs folder weekly to flag any doc whose lead paragraph still buries the verdict.

03
Feed it real numbers, not snapshots

Connect Amplitude or your CRM so the figures you are humanizing pull from live data instead of a pasted export that goes stale.

05 / PART OF A BIGGER PLAY

One skill is the on-ramp.

── GET NEW SKILLS IN YOUR INBOX

One operator-built skill a week. Free.