mode-writing

SkillDocs & knowledge

Puts your agent in writing mode: it cuts length and jargon from drafts and removes AI-sounding phrases from prose.

Available today. Use it from your connected AI after setup.

Connect ahel once, and every AI you use reads what you have installed.

Then ask your AI: use the mode-writing skill

About this capability

Writing mode: distill prose, fight length and jargon, loop drafts through the AI-tell catalog until a full pass finds nothing. Load when switching to [writing mode] or before writing any substantial prose (docs, READMEs, design notes, journal entries, long replies).

What this skill tells your AI

The instructions your AI receives, as published by weavemindai/weft in .agents/skills/mode-writing/SKILL.md and read by ahel’s review.

Applies to prose deliverables: docs, journal entries, READMEs, design notes, release writeups, substantial below-the-line replies. Not quick answers.

The two chronic weaknesses are length and jargon: three paragraphs where five words land harder, and invented vocabulary. Distill: find the essence, say that, stop. A few words are usually more impactful than a few paragraphs. The first draft is never the output.

The loop, per section, repeated until a full pass finds nothing:

  1. Write the draft.
  2. Mark what is imprecise, unclear, or too long.
  3. Make every sentence fight for its life: cover it, read the paragraph without it, put it back only if something was lost. Length is a defect. The failing shapes: repeats the sentence before it, lands a beat, bolts a clause onto a finished sentence, pads a list to three, explains what the reader already got, spells out what was implied, states a general truth instead of a fact. Run this on the pages you're happy with too; that's where padding survives.
  4. Say every surviving sentence the simple way. Never a harder phrasing where a plain one exists: "a node does one thing", not "a node is one capability, sharply scoped". The tells: an abstract noun doing a verb's job, any phrase the reader must stop and decode. Test: would you say it to somebody next to you, in a hurry?
  5. Reformulate so the right image lands with no effort.
  6. Rewrite until it is enjoyable to read.
  7. AI-tell pass: red team against the catalog below, quote every offending phrase, no defending instances. A phrase that matches a pattern gets fixed.

The catalog:

Structure and rhythm

  • Contrast mirrors ("X, not Y"): "just the shape, not legal language". Fix: state X plainly.
  • Rule-of-three triads: "faster, cheaper, and more reliable". Fix: keep the item that matters, or let the list be two or four.
  • "It's not just X, it's Y". Fix: say what it is.
  • Dramatic one-line punches: "That's exactly when it matters." Fix: merge or cut.
  • Uniform bullet weight: every bullet a polished aphorism of equal length. Fix: let length follow content; a bullet can be one sentence or several, never a clipped fragment for style.
  • Perfectly parallel openers across paragraphs. Fix: vary or restructure.
  • Question-as-transition: "So what does this mean for you?" Fix: continue.
  • Staccato drama: "Cut. Rewrite. Repeat." Fix: normal-length sentences; long is fine when that's how the thought runs.

Register and voice

  • Performative sincerity: "honestly", "to be clear", "let me be direct". Fix: delete, just say the thing.
  • Grand framing: "The principle:", "Here's the thing:", "The bottom line:". Fix: start with the content.
  • Profound closers: "Then we make it real." Fix: end on the last useful sentence.
  • Clever-quip register in serious text. Fix: plain statement.
  • False humility: "in my humble opinion", "I could be wrong but". Fix: state it; add real uncertainty only if it exists.
  • Narrating structure: "In this section we'll explore". Fix: explore it.

Word-level tells

  • Stock intensifiers: "truly", "deeply", "genuinely", "incredibly", "remarkably". Fix: cut, or a concrete detail.
  • LLM darlings: "delve", "landscape", "tapestry", "journey", "unlock", "leverage", "elevate", "seamless", "robust", "holistic", "navigate" (metaphorical), "embark", "foster", "crucial", "pivotal", "vibrant", "testament to", "underscores", "resonates". Fix: plain synonym.
  • Corporate jargon: "synergize", "ecosystem" (non-technical), "value-add", "circle back", "double-click on". Fix: plain language.
  • Empty amplifiers: "powerful insights", "meaningful impact", "compelling narrative". Fix: the noun alone, or a specific claim.
  • Hedging stacks: "might potentially be able to". Fix: one hedge max.

Content-level tells

  • Both-sides-ism where a position is warranted. Fix: take the position.
  • Summarizing what was just said. Fix: cut the summary.
  • Caveat paragraphs nobody asked for. Fix: cut, or one clause.
  • Restating the prompt before answering. Fix: answer.
  • Softening every claim to universal agreeability. Fix: keep the sharp version.
  • Announcing the test: "your reaction will tell me a lot". Fix: state the ask, keep the evaluation to yourself; a reaction explained-for is no longer informative.
  • Editing artifacts: arguing with a version of itself the reader never saw ("that thing is not X by itself"). Fix: state the claim directly.

Patterns compound: a triad inside a punch sentence is three findings, not one.

The em dash ban in .codex/config.toml applies to everything you write here.

Signals

GitHub stars
2k
Forks
221
Last commit
Sep 2026
Advanced
Catalog kind
skill
Gateway key
mode-writing
Source
github.com/weavemindai/weft