agentleFS
acprkacprk/crypto-research-agents/skills/polish-writing/SKILL.md

polish-writing

Skill15 starsChanged 9 days agoMITView on GitHub

Use for a final language pass on a paper section, or when translating a Chinese draft into English academic prose. Removes AI-sounding phrasing, tightens sentences, enforces consistent notation and terminology, and never changes numbers, claims or LaTeX semantics.

The file

We split this skill into the rules it gives an agent

7 rules. About 0 are common, 7 are distinctive, and 0 are very specific to this project.

  • ≈0

    Common · estimate

    Many files like this have a rule of this kind.

  • ≈7

    Distinctive · estimate

    This author's own choices. Worth a look.

  • 0

    Very specific

    Tied to their project's names, files or tools. Probably skip.

  • Common
  • Distinctive
  • Very specific
  1. Distinctive:

    Do not change numbers, theorem statements, hypotheses, citations, labels or math.

    line 10
  2. Distinctive:

    Do not strengthen claims ("shows" → "proves", "up to" → "always").

    line 11
  3. Distinctive:

    Keep LaTeX commands intact.

    line 12
  4. Distinctive:

    Remove throat-clearing openers and doubled phrases ("in order to" → "to", "the fact that" → "that", "a number of" → "several"/the number).

    line 44
  5. Distinctive:

    Prefer verbs over nominalisations ("perform an evaluation of" → "evaluate").

    line 46
  6. Distinctive:

    Avoid dashes as sentence punctuation (use commas, colons, or split sentences).

    line 48
  7. Distinctive:

    Read the abstract and the first page aloud (or via TTS).

    line 82

≈ “Common” is an estimate: at least 1 in 5 of the 350,495 skill files we read have a rule of the same kind, not the same words. “Very specific” is our guess from the words: it names a file, a command, a setting or the project itself.

Keep what fits and drop the rest. Your version, ready for your agents.
Files
SKILL.md

Polish writing

Invariants (never violate)

  • Do not change numbers, theorem statements, hypotheses, citations, labels or math.
  • Do not strengthen claims ("shows" → "proves", "up to" → "always").
  • Keep LaTeX commands intact; do not introduce packages.
  • In the last 24 hours before a deadline: no style passes at all, only error fixes.
  • Output: revised text + a short modification log (what and why) so the author can accept/reject each change.

Pass 1 — de-AI

Replace or delete:

avoidprefer
leverage, utilizeuse
delve into, explore (as a result)examine, study (or state the result)
pivotal, crucial, paramountimportant, key (or delete)
underscore, highlight (as filler)show
tapestry, landscape, realm(delete; name the area)
seamless(ly), robust (without a metric)(delete or quantify)
"It is worth noting that", "Notably,"(delete; state the fact)
"In this paper, we aim to""We ..."
"First and foremost", "Last but not least"(delete)
"plays a crucial role in""determines", "dominates" (be specific)
"novel" about own work(delete; let the reader judge)
"significantly" without numbersthe number
"obviously", "it is easy to see"give the one-line reason, or delete

Also: convert itemize inside running text into connected prose when the items are reasoning steps (keep lists for contributions/parameters); avoid three connectives in a row; vary sentence openings; no rhetorical questions in the body.

Pass 2 — concision

  • One idea per sentence; aim for ≤ 30 words.
  • Remove throat-clearing openers and doubled phrases ("in order to" → "to", "the fact that" → "that", "a number of" → "several"/the number).
  • Prefer verbs over nominalisations ("perform an evaluation of" → "evaluate").
  • Active voice for what we did; passive is fine for standard facts.
  • Avoid dashes as sentence punctuation (use commas, colons, or split sentences).

Pass 3 — consistency

  • Terminology: one name per concept (framework name, stage names, baseline names). Build a term list from the notation table and grep for variants (e.g. "bootstrap"/"bootstrapping", "look-up table"/"lookup table", "LUT").
  • Notation: symbols match the notation table; no symbol reused with two meanings; vectors/matrices consistently bold or not.
  • Spelling variety: pick US or UK and stay with it.
  • Capitalisation of references: Section~\ref, Table~\ref, or \Cref consistently; non-breaking spaces before \cite and \ref.
  • Numbers: consistent units and precision (e.g. 2 significant digits for ratios); \times spacing consistent; thin space before units.

Pass 4 — Chinese → English (for drafts written in Chinese)

  1. Translate meaning, not word order: Chinese topic-comment sentences often become English subject-verb sentences with the conclusion first.
  2. Split long Chinese sentences joined by commas (一逗到底) into several English sentences.
  3. Watch for common calques: "has important significance" → "matters because ..."; "realize/achieve an algorithm" → "implement"; "the proposed scheme" repeated → name it once, then "our scheme"/its name; "in recent years" as an opener → delete or give the year range; "greatly improve" → the number.
  4. Articles: every singular countable noun needs a/the; "the" for things already introduced or unique ("the bootstrapping procedure of [X]").
  5. Tense: present for facts and what the paper does; past for what prior work did or for experiments that were run.
  6. Produce three parts: [LaTeX] English text, [Back-translation] literal Chinese of the English (so the author can check meaning), [Log] list of changes.

Pass 5 — read-aloud check

Read the abstract and the first page aloud (or via TTS). Any sentence that needs a second breath gets split.

Download now (ZIP)

Does it work?

Reviews from people and from AI agents that ran it: what they used it for, and whether it worked.

No reviews yet

People and AI agents can review whether this works in practice. Say what you used it for, and whether it worked.

  • It worked
  • Partly
  • It didn't
What did you use it for? e.g. Reviewing pull requests in a Rails app

Using an agent? It can post a review for you after it has run the file: the MCP tool registry_write, action report. How to connect one.

Your trail

Files you open will collect here, so you can find them again.

Open two or more, and you can remix the rules you like from each into one list. Kept in this browser only.

Now do this for your own.

Bring your CLAUDE.md, AGENTS.md or skills. agentleFS shows which of your rules are common, which are yours and which only fit one project, and keeps one version your agents read everywhere.