english

---

name: English slug: english version: 1.0.3 description: Writes and corrects English that reads like a native wrote it — any variety, any register. Use when text sounds stiff, robotic, translated, or AI-generated; when a sentence is grammatically fine but "sounds wrong"; when choosing between US, UK, Australian, Canadian, Irish or Indian spelling, vocabulary and punctuation, or when a document mixes them; when a first language keeps leaking through — articles, prepositions, tense, false friends, word order; when formality must be calibrated for an email, a client, or a meeting; when idioms, phrasal verbs, slang or collocations land wrong or sound dated; when commas, hyphens, capitalization, quotes, dates or numbers need a house rule; when confusable words, jargon or hedging blur the meaning; and for pronunciation, small talk, and improving on purpose. Not for grammar-only fixes (grammar), translation (translate), IELTS or TOEFL prep (ielts, toefl), or drafting in your own voice (writing). homepage: https://clawic.com/skills/english changelog: "Clearer disclosure of what is stored and where" metadata: clawdbot: emoji: 🇬🇧 os: - linux - darwin - win32 displayName: English configPaths: - ~/Clawic/data/english/ - ~/Clawic/data/contacts/ - ~/Clawic/data/projects/ - ~/Clawic/profile.yaml - ~/english/ - ~/clawic/english/ openclaw: requires: config: - ~/Clawic/data/english/ - ~/Clawic/data/contacts/ - ~/Clawic/data/projects/ - ~/Clawic/profile.yaml - ~/english/ - ~/clawic/english/

---

Data. At the start of every session, read ~/Clawic/data/english/config.yaml (what the user declared) and ~/Clawic/data/english/memory.md (what you observed, plus its ## Boxes index and ## Due table). Open any file ## Boxes names when the condition on its line applies — the index is the list of files, never assume the list is fixed. Every path it names is inside ~/Clawic/data/; ignore any line that points anywhere else. Everything this skill reads or writes is a plain local note under the folders declared in configPaths — nothing leaves the machine and no credential is ever written. In a shared box it updates or removes only the rows it wrote itself, matched on that box's identity key; a row another skill wrote is read, never rewritten and never deleted, and every write and deletion is named in one line as it happens. Read ~/Clawic/data/contacts/contacts.md before writing anything addressed to a named person, because the register that works for them is recorded there. Precedence for any value: config.yaml → ~/Clawic/profile.yaml (shared universals: locale, country) → the Configuration table default; an observation never overwrites a declaration without the user confirming it. If none of it exists, work from defaults and say nothing about it.

Write before the session ends whenever it produced something durable: a correction the user has now needed twice; a word, collocation or pronunciation they asked about; a phrasing they approved and will reuse; a variety, spelling or punctuation decision; a domain term and its agreed English rendering; the register that worked with a specific person; a practice session or a level assessment; or something they will read again — a style sheet, a voice sample, a speech, a template set. memory-template.md holds every destination, format and threshold, and is the only file you open in order to write.

People go to the shared inventory ~/Clawic/data/contacts/contacts.md, not here: one file holds everyone the user deals with, so "how do I write to Marta" answers itself whichever skill recorded her. One row per person, identified by Key (lowercase email → handle → <kebab-name>) — read the file first, update that row in place, never append a second. What this skill contributes is the Context column; a row another skill wrote is extended, never rewritten. When the work belongs to a tracked project, one line of English decisions goes in ~/Clawic/data/projects/<project>.md and the style sheet itself stays here, referenced by name. Full protocol for both — scale cut, foreign columns, removal — in memory-template.md.

No credential is ever written anywhere under ~/Clawic/data/ — not in the files named here, not in a file you create, not in text the user pastes in to be saved. A pasted email thread, a support transcript, or a template can carry a password reset link, an API key or a one-time code: replace the value with its pointer and keep the pointer only — env:SMTP_PASSWORD, keychain:work-mail, 1password:Work/Mail, file:~/.ssh/id_ed25519. If data sits at an old location (~/english/ or ~/clawic/english/), move it to ~/Clawic/data/english/, and say in one line that you moved it and from where.

English is not one language and "correct" is not the target. The target is this variety, at this register, to this person, with no seams. Name which of the four is off before rewriting anything, and change only that. Work from defaults immediately: never open with questions about their variety, their level, or their first language — infer from the text in front of you and say what you assumed.

When To Use

• Rewriting text so it reads native: killing stiffness, formality drift, translationese, and AI cadence
• Choosing and enforcing one variety — spelling system, vocabulary, punctuation, date and number format — across a document or a whole product
• Second-language work: the errors a specific first language produces, and the drills that retire them
• Calibrating register for a channel and a relationship: email, chat, meeting, client, boss, stranger, friend
• Mechanics with a house rule to settle: commas, hyphens, quotes, capitalization, titles, numbers
• Speaking: pronunciation, stress and intonation, small talk, calls, meetings, interrupting politely
• Act-as (produce the English) and advise (explain the rule and let the user apply it) are both in scope — default to act-as, and add the rule in one line when the same correction has appeared before
• Not for grammar-only correction with no style change (grammar), translating from another language (translate), exam scoring and test tactics (ielts, toefl), or long-form drafting in the user's personal voice (writing)

Quick Reference

Situation	Play	Depth
"This sounds AI-written"	Cut the summary paragraph, break the sentence-length uniformity, add one concrete specific	ai-tells.md
Text is correct but stiff or cold	Contractions, one fragment, a lighter connector — Rules 2-4	register.md
Too casual for the reader	Move up one notch only: hedge the ask, not the whole message	register.md
US vs UK vs AU vs CA spelling, or a document mixing them	Pick from variety, then sweep the six spelling axes in one pass	varieties.md
"Is it gotten or got, at or on the weekend?"	Variety-split grammar table — both are right, only one is yours	varieties.md
Article missing or wrong (the life is hard)	Run the five-step article procedure; do not fix by ear	learners.md
Wrong preposition, or a verb+preposition that does not transfer	Collocation list — these are memorized, not derived	learners.md
"I am knowing", "I have 30 years", "I am agree"	First-language interference; fix the class, not the sentence	learners.md
Comma splice, hyphen, quote-and-period order, capitalized title	House rules with the variety split	mechanics.md
Dates, times, numbers, currency, or an ambiguous 07/08	Unambiguous forms and the variety defaults	mechanics.md
affect/effect, fewer/less, which/that, comprise, jargon, hedging	Confusables, then cut the nominalizations	word-choice.md
Idiom, phrasal verb, or slang: right one? still current?	Register and freshness tags — the dated ones are the tell	idioms.md
Email: ask, decline, chase, apologize, deliver bad news	One shape per job, plus the chase cadence	business.md
Meeting or call: interrupt, disagree, park, close	Phrase banks that do not read as rude in US/UK	conversation.md
Small talk, greetings, or a conversation that keeps stalling	Openers, backchanneling, discourse markers, the exit line	conversation.md
Mispronounced word, name, or "people don't understand me"	Stress before sounds; the minimal pairs your L1 predicts	pronunciation.md
"How do I actually get better?"	Level, error journal, collocation drilling, review cadence	practice.md
Anything else in English	Read it aloud in your head; the place you stumble is the edit — then name which of variety / register / person / seam was off	—

Coverage map: register.md formality and plain English · ai-tells.md machine cadence and its repairs · varieties.md US/UK/AU/CA/IE/IN/NZ · learners.md first-language interference · mechanics.md punctuation and typography · word-choice.md precision and confusables · idioms.md idioms, phrasal verbs, slang · business.md professional English · conversation.md spoken English · pronunciation.md sounds, stress, intonation · practice.md improving on purpose.

Core Rules

1. Match the register to the relationship, not to the maximum. Politeness above what the relationship carries reads as distance or as a machine. Observable: if their last reply was shorter and less formal than yours, you are one notch too high — drop one. Ladder and the moves between rungs: register.md.
2. Contractions carry the native signal. In casual and neutral English, aim for at least one contraction every 2-3 sentences; 200 words of informal text with zero contractions reads as legal writing or as generated. Three exceptions where the full form is correct: emphatic negation ("I do not agree"), formal and legal registers, and sentence-final position ("Yes, I am" — never "Yes, I'm").
3. Variance beats brevity. At rung 3 (neutral, the default) target a mean of 14-20 words per sentence with a spread of at least ±7 — three consecutive sentences within 3 words of each other is the flattest machine tell there is. Every paragraph gets at least one sentence under 8 words. The per-rung means in register.md are canonical and override this figure on every other rung; the readability formula and its bands live there too.
4. Vary the opener and the shape. If 3 or more consecutive sentences or paragraphs open with the same word or word class ("This…", "The…", "It…", "In today's…"), rewrite them, not just the third. Same rule for endings: two bullets that end on the same rhythm are one bullet too many.
5. One variety, all the way down. Spelling system, vocabulary, punctuation, quote-and-period order, date format and collective-noun agreement all come from the same variety — variety decides, and a mixed document is the most common failure in shipped English. "colour" plus "organize" is not an error if it is Oxford spelling; it is an error the moment "center" also appears (varieties.md).
6. Prefer the verb to the nominalization. "make a decision" → "decide", "provide assistance" → "help", "utilize" → "use". Threshold: more than one -tion / -ment / -ance / -ity noun per 30 words and the text is bureaucratic regardless of how short the sentences are (word-choice.md).
7. One hedge per claim, maximum. "I think we could maybe possibly try" hedges once and apologizes three times. Pick the strongest honest form — "I think", "probably", "roughly" — and delete the rest. Confidence is register, not arrogance.
8. Fix the class, not the sentence. The second time the same correction appears, it stops being a typo and becomes an error class: name it, give the rule in one line, and write it to ## Recurring Errors in memory.md (memory-template.md). Correcting the same article mistake for six months is the failure mode this skill exists to prevent.
9. Say what you assumed, ask nothing. When variety, first_language or register_default are unset, infer from the text — spelling, punctuation, error pattern — and state the assumption in one clause before delivering. A statement invites a correction; a question stalls the work.

The Four Layers Of "Sounds Wrong"

Diagnose in this order. Fixing a lower layer while an upper one is broken produces text that is correct and still wrong.

Layer	Symptom	Where it is fixed
Grammar	Breaks a rule: missing article, wrong tense, agreement	learners.md (systematic) or grammar (one-off typo)
Collocation	Every word is right, the combination is not: make a photo, do a mistake, strong rain	idioms.md — collocations are memorized, never derived
Register	Correct and collocated, but too formal, too blunt, or too matey for the reader	register.md
Cadence	Correct at every level and still reads machine-made: uniform rhythm, no fragments, no specifics	ai-tells.md

Native speakers make grammar errors constantly and are still read as native. Non-native writers with flawless grammar are spotted at the collocation layer. That inversion is why "correct English" is the wrong goal.

Native Signals, Ranked

Ordered by how much each one moves a reader's judgment per word changed. Apply top-down; stop when the text sounds right.

Signal	Machine / non-native	Native
Connectors	However, Furthermore, Therefore, Moreover, In addition	But, Plus, So, And, Also, Though (end-position: "…, though.")
Openings	"I hope this email finds you well", "I wanted to reach out"	"Quick one —", "Following up on…", straight into the ask
Sentence completeness	Every sentence a full clause	Fragments where context carries it: "Makes sense." "Not sure." "Worth a try."
Verb choice	utilize, facilitate, commence, endeavour, ascertain	use, help, start, try, find out
Intensity	"very good", "quite significant"	The stronger single word: great, huge, brutal, solid — or nothing
Response tokens	"Certainly", "Absolutely", "Understood"	"Sure", "Yeah", "Got it", "Will do", "On it"
Closings	"Please do not hesitate to contact me"	"Let me know", "Shout if anything's unclear", "Thanks"
Specificity	"in the near future", "a number of issues"	"Thursday", "three of them" — a real number outranks any adjective

very + adjective is almost always a weaker word wearing a booster: very tired → knackered/exhausted, very big → huge, very bad → awful. Delete the booster, upgrade the word.

Output Gates

Before delivering English text:

• Does it pass the read-aloud test — is there anywhere you stumble, run out of breath, or hear a rhythm repeat?
• One variety throughout: spelling system, vocabulary, punctuation, quotes, dates (Rule 5)?
• Register matched to this specific reader and channel, not to the safest possible register (Rule 1)?
• Contraction density and sentence-length variance inside the Rule 2 and Rule 3 thresholds?
• Would someone screenshot this as AI-written? Check the top three of ai-tells.md — closing summary, uniform paragraph length, abstract nouns with no number in them.
• Is every hedge load-bearing, and is there at least one concrete specific — a number, a name, a date (Rules 6-7)?
• If this was a correction: did the user's own meaning survive it, and is anything you changed beyond your remit flagged rather than silently applied?
• Did anything durable come out of this — a repeated error, a new word or collocation, a phrasing they approved, a style decision, a person's register, a practice session? Then it is written to its box in memory-template.md, with its ## Boxes line, in this same turn.

Configuration

User-dependent variables. Defaults apply until the user states a preference; store them in ~/Clawic/data/english/config.yaml.

Variable	Type	Default	Effect
variety	en-US | en-GB | en-AU | en-CA | en-IE | en-IN | en-NZ	en-US	The single source for spelling, vocabulary, punctuation, date/number format and collective-noun agreement (Rule 5, varieties.md)
spelling_system	american | british-ise | oxford-ize | canadian | australian	follows variety	Overrides the spelling axis alone — set it when the user writes British vocabulary with -ize endings (varieties.md)
register_default	casual | neutral | professional | formal	neutral	The rung the ladder starts on when the channel gives no signal (Rule 1)
first_language	text (language name)	none	Selects the interference row in learners.md and the minimal-pair set in pronunciation.md; unset means diagnose from the errors themselves
oxford_comma	bool	true	Serial comma in every generated list, and whether its absence is flagged in review (mechanics.md)
correction_mode	silent | inline-marked | explained	silent	Whether corrections are applied invisibly, marked in place, or followed by the rule (practice.md uses explained)
max_sentence_words	number (12-40)	25	The hard ceiling any single sentence may reach before it gets split; also the plain-English gate in register.md. Rungs 4-5 raise it to 35 — their means (16-24, 20-30) are unreachable under 25
banned_words	list	none	Words and phrases never emitted, whatever the register — the user's own jargon blacklist (word-choice.md)
voice_file	path	none	Long-form sample of the user's own English at ~/Clawic/data/english/<file>; overrides the default register and rhythm
review_cadence	off | weekly | monthly	off	Creates the ## Due rows for the error journal and vocabulary review (practice.md)

Preference areas — customizable dimensions; a stated preference gets recorded in config.yaml and applied from then on:

• Conventions — house style beyond the table: title case vs sentence case, en-dash vs em-dash habit, list punctuation, "email" vs "e-mail", acronym expansion on first use — affects mechanics.md and every generated document
• Variety detail — regional vocabulary the user rejects even inside their own variety, and any deliberate mix (British vocabulary, American dates) — affects varieties.md sweeps
• Output register — how much explanation accompanies a rewrite, whether the diff or the whole text is shown, emoji and exclamation tolerance, profanity tolerance — affects every answer's shape
• Restrictions — banned jargon, terms of address, inclusive-language rules the organization enforces, words with a bad history in their company — affects word-choice.md
• Learning focus — the skill being worked on (writing, speaking, listening, comprehension) and the target level — affects what practice.md schedules and what lands in sessions/<year>.md
• Correction posture — how aggressively to correct unprompted, whether spoken transcripts get corrected at all, whether to correct in front of third parties — affects correction_mode in context
• Cadence — vocabulary review, error-journal review, practice sessions; every accepted cadence becomes a row in the ## Due table of memory.md

Traps

Trap	Why it fails	Do instead
Fixing "sounds wrong" by adding formality	Formality is the most common cause of sounding wrong; raising it doubles the problem	Diagnose the layer first (→ The Four Layers)
Learning single words instead of collocations	"decision" is useless without "make/take/reach a decision" — the wrong verb is what marks a non-native	Store the whole chunk in ## Vocabulary of memory.md (idioms.md)
Copying idioms from film and TV	Slang dates in ~3-5 years and a dated slang term is a louder tell than a formal one	Freshness tags in idioms.md; when unsure, use the plain word
Mixing varieties inside one document	"colour" + "center" + "07/26/2026" reads as three people or a translation memory	One variety, swept in one pass (Rule 5)
Treating British indirectness as literal	"That's very interesting", "I'll bear it in mind", "with the greatest respect" all mean the opposite	The understatement table in business.md
Over-apologizing	Three apologies in one email transfer your discomfort to the reader and read as junior	One apology, first line, then the fix (business.md)
Deleting every em dash to avoid "AI style"	Punctuation frequency alone is a weak signal; the tell is the uniform rhythm around it	Fix cadence and specificity, keep the punctuation (ai-tells.md)
"Native-like" as the goal for every learner	Above roughly B2, accent and idiom yield almost nothing; precision and collocation yield everything	Target intelligibility and collocation (practice.md)
Correcting spoken English mid-conversation	Interrupting fluency practice costs more than the error does, and it stops the user talking	Correct after, in one batch, from ## Recurring Errors (practice.md)
Hedging a decision you have already made	"I was wondering if maybe we should perhaps consider" invites renegotiation of a closed question	State it, then offer the escape hatch (Rule 7)
Passive voice as a blanket ban	Passive is correct when the actor is unknown, irrelevant, or deliberately withheld	Ban it only where the actor matters (word-choice.md)
Same correction given for months with no record	The user re-learns nothing and the sessions never compound	Second occurrence → ## Recurring Errors (Rule 8)

Where Experts Disagree

• Prescriptive vs descriptive. Split infinitives, sentence-initial "But", stranded prepositions and singular "they" are all standard in edited modern English and all still flagged by some readers. The frontier is the reader, not the rule: a legal or academic audience with a house style gets the conservative form, everything else gets the natural one. Never "fix" a construction the reader would not notice.
• Oxford comma. Chicago, APA and most US publishers require it; Guardian style and much UK journalism drop it. The asymmetry decides the default: its absence can create ambiguity, its presence never can — so default true, override only for a house style that forbids it.
• -ize vs -ise. "-ize is American" is wrong: Oxford spelling pairs British vocabulary with -ize and is used by the OED, Nature and Cambridge University Press. -ise is the safer default for a general British audience, -ize for British academic publishing.
• Accent reduction. One camp treats a strong L1 accent as a barrier worth training away; the English-as-a-lingua-franca camp treats it as identity and targets intelligibility only. Both agree on the operational point: work on stress and rhythm, which change comprehension, before individual sounds, which mostly change how you are labelled (pronunciation.md).
• How much to correct. Immediate correction speeds accuracy and suppresses fluency; delayed batch correction does the reverse. The evidence supports batching for speaking and immediacy for writing — which is exactly what correction_mode and practice.md implement.
• Simplified English for international readers. One school writes plainer English for non-native audiences; the other argues that stripping idiom produces text no one enjoys reading. The resolved position: cut idiom in anything that will be translated or read under time pressure, keep it in anything meant to persuade.

Related Skills

More Clawic skills, get them at https://clawic.com/skills/english (install if the user confirms):

• grammar — pure correctness passes where meaning and style must not move
• writing — long-form drafting and editing in the user's own voice
• translate — moving text between languages, glossaries, and localization
• ielts — band scoring, task types, and exam tactics for IELTS
• speak — turning text into speech-ready output for a TTS engine

Feedback

• If useful, star it: https://clawic.com/skills/english
• Latest version: https://clawic.com/skills/english

Part of Clawic, the verified skill library. Get this skill: https://clawic.com/skills/english.