# Speak Fking English

> Run before every final response: make it concise and clear, cut AI tells, re-pitch confusion, and use visuals only when they help.

- Skill: `jesse-merhi/speak-fking-english` (Agent Skill)
- Install (CLI): `npx skillmds@latest add jesse-merhi/speak-fking-english`
- Raw SKILL.md: https://api.skillmd.com/api/skills/jesse-merhi/speak-fking-english/raw
- Safety review: pending
- Works with: Claude Code, Claude.ai, OpenAI Codex
- Category: Coding & Dev Tools
- Author: jesse-merhi (https://skillmd.com/u/jesse-merhi)
- Updated: 2026-09-22
- Page: https://skillmd.com/skills/jesse-merhi/speak-fking-english

---


# Speak fking English

Run this over the complete draft as the last editing pass before returning or
saving it.

Before selecting a route, freeze exact names, technical terms, quoted text,
code, logs, commands, and evidence. Every selected pass preserves them character
for character.

## Route the pass

- When the user asks you to use `speak-fking-english`, invokes it with the
  harness's skill syntax, or asks for the full unslop, de-slop, or AI-tells
  pass, run the shared pass, add the deep catalogue pass, apply the final
  brevity pass, and then return the result.
  Merely discussing the skill does not select this explicit route.
- For a "wait, what?" or re-pitch request that does not explicitly invoke this
  skill, run the reader reset, apply the final brevity pass, and return.
- For a "show me" or visual-support request that does not explicitly invoke
  this skill, run the visual filter, apply the final brevity pass, and return.
- When another skill calls this one, return the revised text to that skill
  instead of addressing the user directly. Preserve the audience chosen by the
  calling workflow. Treat the call as implicit unless the user explicitly
  invoked this skill for that artifact.
- For every other model-selected call, including a task that matches the skill
  description or the final-response checkpoint, run the shared pass below,
  skip the deep catalogue pass, apply the final brevity pass, and return. This
  is the implicit route.

## Shared pass

1. Apply the [reader reset](references/reader-reset.md) to the complete draft
   without changing its facts, scope, or requested action.

   Done when the reader can understand or act without reconstructing missing
   context.

2. Apply the [visual filter](references/visual-filter.md) after the reader reset.
   Give it the complete draft and any real evidence supplied by the owning
   workflow. Add support only when it materially reduces comprehension effort,
   and never replace or weaken real evidence.

   Done when prose is the explicit choice or the smallest useful support answers
   one clear teaching question, with every evidence claim still pointing to the
   behavior that produced it.

3. Apply the [natural-writing pass](references/natural-writing.md) to the
   complete draft, including any support added by the visual filter.

   Done when the result is direct, concrete, easy to follow, and recognisably
   written by a person, with every fact and evidence claim unchanged.

## Deep catalogue pass

Run this only on the explicit route. Apply the
[full AI-tells catalogue](references/ai-tells.md) to the result of the shared
pass. Keep the facts, scope, requested action, intended tone, exact names, and
evidence unchanged.

Done when no catalogued pattern survives and the draft still says exactly what
it needs to say.

## Final brevity pass

Apply the [brevity pass](references/brevity.md) last, after every other selected
pass. On the implicit route, it follows the shared pass. On the explicit route,
it follows the deep catalogue pass. On the focused reader-reset and visual-filter
routes, it follows the named pass. Keep the context, evidence, qualifications,
and next action the reader needs. Cut everything else. Do not run another
rewriting pass after it.

Done when every remaining sentence earns its place and removing another one
would make the response less useful or less accurate.

## Return

For chat, return the revised draft as the whole final response. For a calling
skill, return the revised text for that skill to use with its intended audience.

Done when the output stands alone, contains no duplicated explanation, and uses
the return path expected by the caller.

This skill incorporates MIT-licensed guidance adapted from Matt Pocock's
`wait-what`, HumanLayer's `show-me`, and pstack's `unslop`. See
[references/upstream-licenses.md](references/upstream-licenses.md).