--- name: plain-text-normalization source: https://app.decimal.ai/s/plain-text-normalization@1/SKILL.md source_sha256: c8343b88a639 --- # Plain-text normalization ## Contract Enforces three fixed plain-text conventions when cleaning up prose without rewording it: (1) every sentence starts with a capital letter, (2) exactly one space separates one sentence from the next, and (3) no line carries trailing whitespace. Apply whenever asked to tidy, clean up, or make presentable a block of raw text — a pasted note, chat transcript, dictation, or scraped paragraph — where the job is to fix spacing and capitalization, not to rewrite the wording. (Trailing spaces below are shown as `␣` so they are visible.) ## Rules 1. **Sentence-initial capitalization.** The first alphabetic character of every sentence is uppercase. A sentence begins at the very start of the text and immediately after every sentence-ending `.`, `?`, or `!`. Uppercase that first letter even if the input typed it lowercase. Do not change the case of any other word (leave existing proper nouns and mid-sentence words as-is). 2. **Exactly one space after a sentence-ending period.** Between a period that ends a sentence and the first character of the next sentence there is exactly ONE ordinary space: - Never ZERO — a period glued to the next word (`done.Next`, `done.next`) gets a single space inserted (`done. Next`). - Never TWO OR MORE — a run of spaces (`done. Next`, `done. Next`, or a tab) collapses to one (`done. Next`). This is the modern single-space-between-sentences convention, not the typewriter two-space habit. The same one-space rule applies after a sentence that ends in `?` or `!`. 3. **No trailing whitespace.** Strip every space and tab that sits at the end of a line (before the newline) and at the very end of the text. A blank line becomes truly empty — no spaces on it. This never removes meaningful content, only the invisible run-off at line ends. 4. **Preserve the wording.** Change ONLY capitalization and whitespace. Do not add, drop, reorder, or reword sentences; do not touch internal punctuation, numbers, or line breaks other than trailing-space removal. The output is the same text, correctly cased and spaced. ## Worked examples **Rule 1 — a lowercase sentence start is capitalized (both the first word and the one after a period).** ``` BEFORE: the kettle boiled. we brewed a fresh pot. AFTER: The kettle boiled. We brewed a fresh pot. ``` **Rule 2 (missing space) — a period glued to the next sentence gets one space.** ``` BEFORE: Add the flour.Stir gently until it thickens. AFTER: Add the flour. Stir gently until it thickens. ``` **Rule 2 (extra spaces) — two spaces after the period collapse to one.** ``` BEFORE: The trail was steep. The summit view repaid every step. AFTER: The trail was steep. The summit view repaid every step. ``` **Rule 1 + 2 combined — glued AND lowercase.** ``` BEFORE: the garden needs water.the tomatoes are wilting. AFTER: The garden needs water. The tomatoes are wilting. ``` **Rule 3 — trailing spaces at line ends are stripped (`␣` = a trailing space).** ``` BEFORE: Pack the tent.␣␣ Fill the water bottles.␣ AFTER: Pack the tent. Fill the water bottles. ``` **Rule 4 — spacing/case only; wording is untouched (a `?` sentence too).** ``` BEFORE: did the ferry leave?we missed it by minutes. AFTER: Did the ferry leave? We missed it by minutes. ``` ## Edge cases & exceptions - **Abbreviations are not sentence ends.** A period inside `e.g.`, `i.e.`, `etc.`, `Dr.`, `Mr.`, `U.S.`, `a.m.` does NOT start a new sentence: keep the following word's case unchanged and do not force a capital. `we ship many formats, e.g. json and yaml.` → `We ship many formats, e.g. json and yaml.` (leading `we` capitalized; `json` stays lowercase). - **Decimals and version numbers.** The dot in `3.14`, `v2.0`, or `1.5 cups` is not a sentence terminator — never insert a space and never capitalize after it. - **Dotted tokens.** Dots inside `config.yaml`, `example.com`, or `a.b.c` are part of the token; do not treat them as sentence breaks. - **Ellipsis.** Leave `...` (or `…`) intact — do not split it into separated periods and do not insert spaces between the dots. - **Already-correct text.** If a sentence is already capitalized and single-spaced, leave it exactly as-is; do not "re-fix" what already conforms. - **First line of the text.** Its opening letter is capitalized like any sentence start, even though no period precedes it. ## Do / Don't - **Never** leave a sentence starting with a lowercase letter. **Always** uppercase the first letter of every sentence. - **Never** keep two-or-more spaces after a sentence-ending period. **Always** collapse them to exactly one. - **Never** leave a period glued to the next sentence. **Always** insert one space. - **Never** leave trailing spaces or tabs at the end of a line. **Always** strip them. - **Never** reword, reorder, or delete the author's sentences. **Always** change only capitalization and whitespace. - **Never** treat an abbreviation dot or a decimal point as a sentence boundary. **Always** limit the sentence-boundary rules to real sentence ends. ## Common mistakes - Mirroring the input's spacing: passing through the two spaces after a period that the source used instead of normalizing to one. - Failing to split a glued `word.Word`, leaving the two sentences run together. - Capitalizing the first sentence but missing the capital on later sentences that followed a period on the same line. - Leaving invisible trailing spaces at line ends because they are not visible in the rendered text. - Over-correcting: capitalizing the word after `e.g.` / `Dr.` / a decimal, or inserting a space inside `3.14` or `config.json`. - Silently rewording or "improving" the sentences instead of only fixing spacing and case. ## Quick checklist - [ ] Every sentence starts with a capital letter (first sentence and each one after a `.`, `?`, or `!`). - [ ] Exactly one space after each sentence-ending period — no glued periods, no double spaces. - [ ] No trailing spaces or tabs at any line end. - [ ] Wording, order, numbers, and internal punctuation unchanged. - [ ] Abbreviation dots, decimals, and dotted tokens not treated as sentence ends.