agentleFS
Evan-Kim2028Evan-Kim2028/agent-skills/skills/writing/writing-prose/SKILL.md

writing-prose

Skill1 starsChanged 3 months agoMITView on GitHub

House writing floor — human naturalness, anti-slop, coherent opinionated structure without persona cosplay or LLM word-soup. Use when drafting or editing posts, docs, essays, threads, READMEs, or technical explainers that should sound like a sharp human, not generic AI, and no named voice pack is required. Prefer writer-style (default pack: evan) for a named voice. Prefer writing-technical for research/build-log form; writing-docs for pure procedures; style-clarity-grace for turgid syntax; elements-of-style for usage/concision mechanics. Prefer marketing for offer/story/ad frameworks. Prefer writing hub when the writing path is unclear.

4 other projects use a version of this.

Adopted by

Most recently taken up by

The file

We split this skill into the rules it gives an agent

5 rules. 0 are common, 5 are distinctive, and 0 are very specific to this project.

  • 0

    Common

    Most of the 4 edited versions we read keep these.

  • 5

    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:

    Before drafting: one sentence thesis or job (“argue X”, “teach how Y works”, “recommend Z because …”).

    line 39
  2. Distinctive:

    Every section must earn its keep against that spine.

    line 41
  3. Distinctive:

    If unverified: say so. Confident wrong is worse than plain uncertain.

    line 50
  4. Distinctive:

    No treadmill: every block must add information

    line 77
  5. Distinctive:

    Default shorter than the model wants.

    line 83

“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

Writing prose — human, opinionated, coherent

Job: Kill default LLM prose. Keep the take sharp and the logic tight.

This is not a celebrity voice pack. For “sound like Evan / Kaue / person X” use writer-style (this install defaults to the evan pack). This skill is the house floor when no persona is loaded.

Lineage: naturalness + deslop + facts-first principles from solanabr/writer-style-skill (MIT, Superteam Brazil / Kaue), adapted for non-persona writing. Full engine: writer-style. Pack credits: ATTRIBUTION.md.

When to load

  • “Make this less AI” / “tighter” / “more human”
  • Blog, docs, internal essay, changelog, thread without a named voice
  • Technical post that must stay correct and readable
  • Edit pass after marketing draft still sounds corporate-LLM

Non-negotiables

1. One spine

Before drafting: one sentence thesis or job (“argue X”, “teach how Y works”, “recommend Z because …”).

Every section must earn its keep against that spine. Cut “on the other hand” blocks that never change the recommendation.

Bad LLM opinion: long balanced paragraphs that end in “it depends” with no costed pick.
Good opinion: pick, name the tradeoff, say who should ignore you.

2. Facts first (technical)

Neutral fact-sheet for every number, API, flag, version. Style after freeze. Never invent certainty.

If unverified: say so. Confident wrong is worse than plain uncertain.

3. Naturalness floor (always)

From the writer-style naturalness rules (always-on spirit):

  • Uneven rhythm — long calm runs + short punches; punches clump, not metronome
  • One human seam per passage — real number from use, named credit, small confession, loose run-on
  • Calm ≠ clean — don’t over-tidy; over-smooth is the tell
  • One mood per piece — confessional / dry / mentor / builder — don’t sample every register

Seams ≠ catchphrases. Humanity is structure and rhythm, not forced slang.

4. Deslop (subtractive)

Cut clusters of tells (not single innocent words):

  • Em-dash decoration piles, “delve / landscape / robust / nuanced / pivotal”
  • Fake profundity (“X is the language of trust”) → concrete fact instead
  • Restatement loops (“In other words… Essentially… At its core…”)
  • Modular paragraphs you could reshuffle with no damage

Preserve: specific hard-to-fake detail, mixed feelings, real asides. Don’t sand those off.

5. Logical coherence

  • Each paragraph depends on the last (setup → payoff)
  • No treadmill: every block must add information
  • Claims have warrants; metaphors don’t replace mechanism
  • End with a verdict or next action, not a summary of summaries

6. Length discipline

Default shorter than the model wants. If a section doesn’t change the reader’s action or understanding, delete it.

Workflow

1. Spine sentence + audience + form (post / docs / thread)
2. Fact-sheet if technical (voice OFF)
3. Draft in house prose (spine + seams + unevenness)
4. Reshuffle test + treadmill cut
5. Deslop pass
6. Read aloud; kill press-release lines

Quick checklist

  • Thesis in one sentence
  • Technical facts frozen / sourced
  • Uneven sentences; not all 18 words
  • ≥1 real seam (specifics, not filler slang)
  • No paragraph reshuffle-safe
  • Opinion has a costed pick if the piece argues
  • No AI-tell cluster
  • Shorter than first draft

Anti-patterns (LLM opinionated style)

FailureFix
Wordy both-sides theaterPick + tradeoff in two sentences
Even, polished “blog voice”Flatten body; spike only at seams/verdicts
Aphorism without factGround or cut
Section headings as decorationEach section must advance the spine
Confidence without sourceVerify or hedge honestly
Applying Hotz/Vitalik costumeUse writer-style craft layers properly, or stay house prose

Hand off

NeedSkill
Named author voice (default: Evan)writer-style
Research / empirical / build-log formwriting-technical
Runbook / API reference / how-towriting-docs
Turgid / buried actorsstyle-clarity-grace
Usage / punctuation / omit needless wordselements-of-style
Offers / StoryBrand / adsmarketing
Product UI microcopyproduct-design
Validate voice markers / toolswriter-style tools (validate_voice.py)

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.