Edit LaTeX manuscripts without breaking markup

--- name: edit-latex-manuscripts-without-breaking-markup description: "Use when polishing or reformatting a paper written in LaTeX: improves academic prose and checks submission formatting while keeping every command, equation, label and citation intact." --- Apply this whenever you are asked to improve, shorten, proofread or reformat text that contains LaTeX markup. Protect the markup first. Before editing, note every command, environment, equation, label, reference, citation key and macro in the passage. These must come back character for character unless the person explicitly asks to change them. Never convert inline math to words, rename labels, merge citations or restructure environments. When polishing prose: 1. Work paragraph by paragraph and edit only the words around the markup. 2. Tighten wordy constructions, fix grammar and article use, keep tense consistent within each section (present for claims, past for what was done), and match the field's normal level of hedging. 3. Keep the author's technical terms; do not swap them for synonyms. 4. Do not add claims, results or citations. If a sentence makes a claim without support, flag it with a comment line starting with % instead of fixing it. When checking formatting for a submission: 1. Ask which venue and year the paper targets if it is not stated, and remind the person to confirm requirements against the venue's current author instructions, since templates change from year to year. 2. Check for packages that alter margins or spacing, anonymization problems for double-blind review (author names, acknowledgements, self-revealing citations), missing required sections, figure readability and consistent reference style. Output shape: return the edited LaTeX in a code block, then a short table of changes (location, change, reason). List anything that needs the author's decision separately. Keep the result compilable; if you are unsure whether a change compiles, do not make it.

```latex We propose a sparse estimator that converges in $O(n \log n)$ time (Theorem~\ref{thm:rate}). Unlike prior work~\cite{lee2021,okafor2022}, it requires no tuning of $\lambda$. % CHECK: 'outperforms all baselines' below has no table reference. ``` | Location | Change | Reason | |---|---|---| | Para 1 | 'In this paper, we are proposing' -> 'We propose' | Concise, standard tense | | Para 1 | Removed 'very' before 'efficient' | Unsupported intensifier | Needs your decision: the acknowledgements name your lab, which breaks double-blind review.