Skip to content
Agentmart
Submit a tool

plain-english

Extracts the signal from bloated or evasive text: the bottom line, every claim with its strength marked, what it does not say, and what it implies without…

Skill by ptrandev

IndexedMIT★ 1
Write a review

Install

Works with Claude Code, Claude apps, Codex, Cursor, Gemini CLI, GitHub Copilot, VS Code, OpenCode, Goose, Windsurf. Read the repository before you give any tool access to your data.

npx agentmarket install skill/ptrandev-plain-english

Asks which of your tools to add it to (Claude Code, Codex, Cursor, VS Code, Windsurf, Gemini CLI, Claude Desktop) and shows the change first. See https://agentmart.dev/cli.

Get the weekly diff

The weekly diff: one email a week with new agent tools, notable updates and grade changes, plus security alerts for tools you follow. Double opt-in, one-click unsubscribe.

We use your email only for this newsletter. Privacy

About

Extracts the signal from bloated or evasive text: the bottom line, every claim with its strength marked, what it does not say, and what it implies without claiming. Adds nothing from outside the source. Use for "what is this actually saying", "cut through this", "TL;DR this", or on pasted marketing, legal, policy, or AI-generated writing.

Use cases: writing, marketing, legal

No reviews yet. Installed it? Reviews come from verified installers only. How reviews work

Permissions

What this skill can do and what it asks for, from its SKILL.md. Checked Oct 25, 2026.

Tools
No tools found
Secrets
None asked for
Install scripts
None found

Network hosts it talks to 1

  • codeload.github.com · found in install command

Files it touches

  • ~/.claude/skills/plain-english

What's inside

Skill name
plain-english
Description
Extracts the signal from bloated or evasive text: the bottom line, every claim with its strength marked, what it does not say, and what it implies without claiming. Adds nothing from outside the source. Use for "what is this actually saying", "cut through this", "TL;DR this", or on pasted marketing, legal, policy, or AI-generated writing.
License
MIT
Instructions
150 lines in SKILL.md · view on GitHub

Files (2)

  • SKILL.md
  • examples.md · references/

SKILL.md (first lines)

# Plain English

Pull the signal out of the noise. **Do not** rewrite the text. Select from it.

The reader is someone who has to act on this text. The bottom line is the
consequence for the reader, not the topic of the text.

## The guarantee

> Add nothing. Drop nothing that changes the reader's take.

**Never** carry a hedge, bound, condition, scope limit, or attribution forward as
its own text. It rides along as part of the claim it modifies.

## Output contract

**Never** write anything before the bottom line. No preamble, no "Here is the
breakdown".

```
**Bottom line:** <the one thing that matters to this reader, or "Nothing actionable.">

**What it claims**
- **<strength>:** <the claim, with its hedge, number, and scope attached>
  <optional: one sentence naming what the hedge costs, only when the label alone does not show it>

**Not stated**
- <what a reader needs and does not get>

**What it implies but never claims**   <- only when this gap exists
- <the reading the text induces, and the device that induces it>
```

Rules for the shape:

- Length tracks real content, not input length. Three sentences of substance
  produce three bullets. A 2,000-word post with one claim gets one. **Never** pad
  a section to make the output look proportionate to the input.
- Drop **Not stated** only when nothing material is missing. That is rare.
- Drop **What it implies but never claims** whenever there is no gap. That is
  common, and an empty section is worse than none.

## Strength labels

Every claim gets exactly one label.

| Label | Means | Trigger words |
| --- | --- | --- |
| `fact` | Asserted flatly, checkable now | none needed |
| `hedged` | Asserted with an escape hatch | the **Hedges** row under [What must ride along](#what-must-ride-along-with-a-claim) |
| `attributed` | Someone else's assertion, not the author's | the **Attribution** row under [What must ride along](#what-must-ride-along-with-a-claim) |
| `promise` | A future act the author commits to | will, plans to, by <date>, in the coming weeks |
| `opinion` | A value judgment, not falsifiable | best, critical, exciting, industry-leading |

Judgment calls:

- Label by what the sentence does, not by its verb. "We will consider it" is
  `hedged`, not `promise`, because the committed act is only consideration.
- A hedged promise is `hedged`. The escape hatch wins.
- Attribution outranks the rest. "The vendor says it cuts costs 40%" is
  `attributed`, even though the inner claim is a fact.
- If a sentence carries two separately falsifiable claims, split it into two
  bullets. **Never** merge two claims under one label.

## What must ride along with a claim

Attach each of these to its bullet, in plain words.

| Carry | Words and test |
| --- | --- |
| **Hedges** | may, might, could, should, expects, aims to, is designed to, we believe, we expect, estimated, projected, planned, targeted | <!-- lint-style: ignore -->
| **Attribution** | according to, X says, reportedly, per, sources indicate, said |
| **Bounds** | about, roughly, up to, at least, at most, more than, fewer than. "up to 40%" is not "40%". "at least 7 days" is not "7 days". |
| **Quantifiers** | all, most, many, some, few, none, only |
| **Conditions and exceptions** | if, unless, until, provided that, subject to, except, excluding, other than |
| **Negation** | not, no, never, cannot |
| **Time and scope limits** | as of `<date>`, in `<year>`, so far, in `<region>`, for `<group>`, on `<platform>` |
| **Numbers, units, dates, currency, versions, names, percentages** | copied verbatim. Convert nothing, round nothing. |

Judge these by meaning. `essentially` and `effectively` are hedges when they mean
"close but not exactly". `only` is a bound in "only 3 users" and filler in "just
click here". `significant` is a claim when the source gives the number and vague
when it does not.

## Add nothing

- **Never** add context, examples, definitions, or corrections from your own
  knowledge, however right. If the source is wrong, say the source claims it and
  put the correction nowhere.
- Keep an undefined domain term as the source wrote it. **Do not** gloss it. If it
  blocks meaning, that belongs under **Not stated**.
- Vagueness is data. "significant improvement" with no number is extracted as a
  claim with no number, so the reader learns the number is missing.
- **Do not** resolve an ambiguous pronoun by guessing.
- **Do not** add a cause the source only implies by adjacency. That belongs under
  **What it implies but never claims**.

## Finding what is not stated

List only absences that block the reader's decision.

- Who acts, decides, or is accountable. Passive voice and "it was determined"
  are the tell.
- When. "Soon", "in due course", "in the coming weeks" are absences, not dates.
- How much. A percentage with no baseline, a gain with no denominator.
- Who is affected, and how they find out.
- What happens if it fails, or what the tradeoff cost.
- The comparison. "Faster" and "better" need a "than what".

List the absence, not a complaint about it. Write "Who decided", not "They
conveniently avoid saying who decided".

## What it implies but never claims

Include this section only when the text induces a belief it never asserts. Every
bullet names the device, so the reader can check you.

- **Adjacency posing as cause.** "Sales fell. We changed the pricing." implies a
  link the text never claims.

From ptrandev/agent-skills, MIT.

More from ptrandev/agent-skills

Signals

Stars
1
Forks
0
Open issues
0
Last commit (this folder)
Sep 4, 2026 (1 month ago)

From GitHub (fetched Oct 24, 2026).

Details

License
MIT
First seen
Oct 24, 2026
Last checked
Oct 24, 2026

More from ptrandev

Cite this page

plain-english skill. Agent Mart. https://agentmart.dev/skills/ptrandev-plain-english (last updated 2026-10-24)