proofread-document
/proofread-document
Wyszukuje literówki i błędy językowe w dokumentach LaTeX i markdown, podaje lokalizację i poprawkę, a zmiany wprowadza na żądanie.
- Autor
- elias-ramzi ↗
- Repozytorium
- weblatexmcp ↗
- Licencja
- MIT
- Ostatnia zmiana repozytorium
- 05.10.2026
Ocena i bezpieczeństwo
- Pliki w paczce
- 1
- Skrypty
- 0
- Komentarze HTML
- 0
- Adresy URL w SKILL.md
- 0
Brak sygnałów. Repozytorium zmienione 05.10.2026, bez odwołań do wycofanych modeli i API.
MIT, otwarta: można używać i zmieniać, z zachowaniem informacji o autorze.
Co robi i kiedy
Wyszukuje literówki i błędy językowe w dokumentach LaTeX i markdown, podaje lokalizację i poprawkę, a zmiany wprowadza na żądanie.
Wyzwalacze z opisu
Oryginalny opis (EN)
Hunt typos and language errors in a LaTeX/markdown document — spelling, doubled or missing words, subject/verb agreement, punctuation, quotes, unescaped LaTeX characters, and inconsistent hyphenation or capitalization of repeated terms. Reports each finding with its line and an exact minimal replacement, and applies fixes only when asked. Use when the user asks to "proofread", "check for typos", "find spelling/grammar mistakes", or "correct the text" of a paper or draft. Never rewrites prose for style and never touches a .bib. Operates on projects served by the web-latex-mcp MCP server.
Jak działa
- The one hard rule: you hunt typos, not writing
- What counts as a typo — report these
- What is not yours — do not report
- How to write a finding
- Workflow
- After you finish
Podgląd SKILL.md
---
name: proofread-document
description: Hunt typos and language errors in a LaTeX/markdown document — spelling, doubled or missing words, subject/verb agreement, punctuation, quotes, unescaped LaTeX characters, and inconsistent hyphenation or capitalization of repeated terms. Reports each finding with its line and an exact minimal replacement, and applies fixes only when asked. Use when the user asks to "proofread", "check for typos", "find spelling/grammar mistakes", or "correct the text" of a paper or draft. Never rewrites prose for style and never touches a .bib. Operates on projects served by the web-latex-mcp MCP server.
project: optional
---
# Proofread a document
Find typos in a paper served by the `web-latex-mcp` MCP server and report them as exact,
minimal substitutions. **Report only by default** — nothing is edited until the user says
to.
## The one hard rule: you hunt typos, not writing
A typo is text that is _wrong_ — a reader would call it an error, not a choice. If a
sentence is merely clumsy, wordy, passive, repetitive, over-hedged, or something you would
have phrased better, it is not yours. Leave it exactly as it is.
When you catch yourself justifying a finding with "clearer", "smoother", "stronger", "more
concise", or "reads better", delete it: that is the tell that you crossed from correcting
into rewriting. The paper's voice is the author's.
## What counts as a typo — report these
- Misspellings, including in `\label`/`\ref` keys when the mismatch is provable from the
document itself.
- Doubled words (`the the`), missing articles or prepositions, dropped words.
- Subject/verb and singular/plural disagreement; wrong verb tense against the paper's
present-tense convention.
- Punctuation: missing or doubled periods and commas, space before punctuation, a missing
`~` before `\cite`/`\ref`, `e.g.`/`i.e.` without the following comma, `...` where
`\dots` belongs.
- Quotes: `"` in LaTeX where ` `` ` and `''` belong.
- Unescaped `%`, `&`, `_`, `#` in prose.
- Inconsistent capitalization or hyphenation of a term used more than once (`state of the
art` vs `state-of-the-art`, `Transformer` vs `transformer`).
- Broken math/text spacing that is unambiguously wrong (`$x$is`).
## What is not yours — do not report
- Style, tone, flow, word choice, sentence restructuring, splitting or joining sentences,
reordering clauses, trimming redundancy, replacing a word with a better one. A sentence
you would have written differently is not a typo.
- Anything the LaTeX writing guide covers as a _convention_ rather than an error —
"we"/"our" overuse, caption phrasing, section header sentences, float placement. Those
are the author's to apply, not yours to enforce.
- Anything inside `\begin{verbatim}`, `lstlisting`, `minted`, or a comment line starting
with `%`, unless the user puts comments in scope.
- Citation keys, numbers, results, or claims — you cannot verify those by reading.
- Anything in a `.bib` file. Skip `.bib` files entirely; normalizing a bibliography is
`format-bibliography`, and checking one against a bibliography service is `verify-citations`.
## How to write a finding
- **Never guess.** If you are not confident the text is wrong, drop it. A false positive
costs the user more than a missed typo, and a file with no findings is a valid result.
- Each replacement is a **minimal, in-place substitution**: change only the wrong
characters, preserve surrounding whitespace, and never reflow a line. The project may
use one-sentence-per-line; keep line breaks exactly as they are.
- `oldText` must occur **exactly once** in the file — extend it with surrounding words
until it is unique. If you cannot make it unique, give the line number and do not apply
anything.
## Workflow
1. **Pick the project.** If the user didn't name one, `list_projects` and ask which — do
not guess.
2. **Build the work list.** `list_files`, then take every `.tex`, `.md`, and `.txt` file,
**excluding** every `.bib` and anything under a build/output directory. If the user
named a file or directory, narrow to that. Say how many files you found before you
start; if there are more than ~25, give the count and ask whether to do all or a subset.
3. **Read each file in full** with `read_file` and collect findings per the rules above.
4. **Report.** One table grouped by file: line, category, old → new, and a one-clause
reason. Merge findings that contradict each other across files (a term hyphenated one
way here and another there) into a single consistency finding naming every site. Then
stop and ask whether to apply.
5. **Apply, only when the user says so.** `edit_file` per finding, one at a time, re-reading
on any failure rather than retrying blind. Never pass `confirmBibEdit`; never pass
`overrideExternalChanges`.
6. **Prove nothing broke.** After applying to a LaTeX project, `compile` and paste the
error count and any remaining error verbatim. If the compile regresses, revert the
offending edit rather than papering over it.
## After you finish
Report concisely: files scanned, findings by category, fixes applied vs reported, compile
status, and anything you left for the user to decide. Do not `commit` or `push` unless
asked.
Wersje
W bazie nie ma innych skilli o nazwie proofread-document.
Migawka bazy z 13.11.2026 · wersja paczki f0aa96856114 (skrót SHA-256 całej paczki)
Pobrania w ostatnich 30 dniach: 0 · ranking
Zgłoś problem z tym skillem: szkodliwa treść, błędny opis albo prośba autora o usunięcie.
Podobne skille
- 100proofread
Sprawdza gotowy tekst pod kątem literówek, gramatyki, interpunkcji i spójności, poprawiając tylko błędy i wypisując każdą zmianę. - 92ase-docs-proofread
Sprawdza dokumenty pod kątem błędów ortograficznych, wielkich liter, interpunkcji, podziału wyrazów i gramatyki. - 100wps-proofread
Sprawdza chińskie dokumenty pod kątem literówek, interpunkcji i błędów językowych, grupując uwagi na trzy poziomy. - 100grammar-check
Sprawdza tekst pod kątem gramatyki, pisowni, interpunkcji i logiki i wylicza poprawki z miejscem, błędem i powodem, bez przepisywania. - 100latex-check-chat
Sprawdza wzory LaTeX w odpowiedziach i dokumentach na claude.ai oraz w Cowork przed wysłaniem. - 90expert-language
Sprawdza pisownię, interpunkcję i kontekst tekstów po turecku i po angielsku oraz usuwa ślady pisania przez AI z tekstu.