# Clearmode

> Write and edit so anyone can read it: plain words, human voice, real structure, no filler. Enforces the CLEAR-100 standard on four axes (Plain, Human, Structured, Dense) with a zero-dependency checker. Use this skill whenever you draft, rewrite, review, or edit any prose a person will read: posts, docs, READMEs, emails, newsletters, landing pages, release notes, PR descriptions, or instructions for another agent. Also use it when asked to deslop, humanize, simplify, or de-jargon existing text.

- Skill: `eugeniughelbur/clearmode` (Agent Skill, multi-file: 14 files)
- Install (CLI): `npx skillmds@latest add eugeniughelbur/clearmode`
- Raw SKILL.md: https://api.skillmd.com/api/skills/eugeniughelbur/clearmode/raw
- Safety review: pending
- Works with: Claude Code, Claude.ai, OpenAI Codex
- Category: AI & ML
- License: MIT
- Author: eugeniughelbur (https://skillmd.com/u/eugeniughelbur)
- Updated: 2026-09-17
- Page: https://skillmd.com/skills/eugeniughelbur/clearmode

---


<!-- generated by scripts/compile_targets.py from rules/clear-100.json. do not hand-edit. -->

# clearmode

CLEAR-100 v1.0.0. Four axes, 43 rules, one checker.

Aerospace built [ASD-STE100](https://www.asd-ste100.org/) so a mechanic on a runway
could not misread a step. It works, and it reads like a machine wrote it, because it
was built for readers who cannot ask a follow-up question. CLEAR-100 keeps that
discipline and adds the two things STE never needed: a human voice, and a page shape
a person will actually scan.

| Axis | What it asks | Weight |
|---|---|---|
| Plain | A reader outside your field understands it on the first pass, with no dictionary and no follow-up question. | 30% |
| Human | It reads like a person who did the work, not like a model that read about the work. | 30% |
| Structured | The shape of the page matches the shape of the information. Lists look like lists, steps look like steps. | 25% |
| Dense | Every sentence carries information a reader could not have guessed. Delete it and the piece loses something. | 15% |

## Before you write a sentence

Ask three things:

1. Would someone outside my field understand this on the first pass?
2. Would a person who did this work say it this way out loud?
3. If I delete this sentence, does the reader lose anything?

A no to any of them means rewrite, not soften.

## Hard rules, no exceptions

- **H1 No slop vocabulary.** Say the specific thing the word was standing in for. If nothing specific is behind it, delete the sentence.
- **H2 No slop phrases.** Delete. If the sentence dies without the phrase, the sentence was the phrase.
- **H4 No em-dash, no en-dash used as one.** Use a period and a new sentence, a comma, a spaced hyphen, or parentheses.
- **H6 No negative parallelism.** State the claim once, directly.
- **H12 No assistant voice.** Delete. Published text has no counterparty to be polite to.
- **H14 No model self-reference.** Delete the sentence and answer the question, or say the fact is unverified and why.
- **D3 No throat-clearing opener.** Lead with the number, the name, or the claim.

## The rest of the rules

Load `references/rules.md` for all 43 rules with the reason, the fix, and a
before-and-after for each. Load `references/lexicons.md` for the word lists:
108 slop words, 68 slop phrases,
111 plain-word swaps, 79 jargon glosses.

## Profiles

Pick the profile that matches the reader, then hold its limits.

| Setting | general | technical | social | agent |
|---|---|---|---|---|
| Sentence cap | 25 | 28 | 20 | 20 |
| Sentences per paragraph | 4 | 5 | 2 | 6 |
| Reading grade target | 9 | 11 | 8 | 9 |
| Reading grade hard cap | 13 | 15 | 11 | 12 |

## The loop

1. Write or read the draft.
2. Run the checker: `python3 scripts/clearcheck.py FILE --profile <general|technical|social|agent>`.
3. Fix every `error`. Fix `warn` unless you can say why the rule is wrong here.
4. Re-run. Stop when the score clears the gate (80 by default) and no errors remain.
5. Never argue with a rule in the draft. Either fix the line or suppress it with `clear: ignore <RULE>` on that line and say why in the commit.

## What this skill will not do

- It will not make a false claim readable. Wrong and clear is still wrong.
- It will not strip a technical term that is the subject of the sentence. It glosses it once.
- It will not ban contractions. Formal reads synthetic.
- It will not touch code, code blocks, quoted material, or someone else's words.

