Wrong for the field
A materials paper and a cancer paper do not share phrasing, tense habits or notation. Generic tools apply one style to all of them.
Early access · v0.1
Refined before your advisor says “do it again.”
no more DASI polishes English research manuscripts to the conventions of your field, measured from published papers rather than guessed. Every edit leaves a receipt, and nothing is delivered until four deterministic checks pass.
Write in any language. Korean drafts get a dedicated anti-translationese layer.
Before
The bandgap of the film was measured through the UV–vis spectroscopy to be 2.31 eV by us, and band gap was increased compared to the bulk [12].
After
We measured a band gap of 2.31 eV for the film by UV–vis spectroscopy, which is larger than that of the bulk [12].
2.31 eV, UV–vis and [12] are verified byte-identical
- 71field overlays measured from real papers
- 1,646active CC BY 4.0 source records
- 4script gates before delivery
- 3files with every edited manuscript
The problem
Generic grammar tools fix commas. Reviewers judge register.
Silent meaning drift
A rewrite that moves a decimal, drops a citation or turns “band gap” into “bandgap” halfway through is worse than no rewrite.
Unexplainable edits
When a tool changes 40% of your text, you cannot tell your advisor why. You re-read everything to find out.
What it does
Field-aware editing with proof attached.
71 field overlays, corpus-measured
Each overlay carries style metrics, top terms, a phrase bank and a notation watch list mined from real papers across Nature’s 71 subject categories.
Meaning invariant, verified
Numbers, units, chemical formulas, citations, equations and DOIs must survive verbatim. A script blocks delivery otherwise. A change rate past 30% warns; past 50% it halts and asks.
No LLM self-grading
Journal coverage, integrity, terminology and abbreviations are checked by scripts that must exit 0. “I checked it myself” is not accepted as evidence.
Every edit leaves a receipt
A section-level journal records changed spans and deliberately kept spans, each tied to the rule that fired. An HTML integrity report ships with every manuscript.
Genre kept, not flattened
Academic register stays. No 20-word sentence caps, no forced active voice, because the target journals’ own measured statistics say otherwise.
Korean-author aware
A pitfalls layer blocks translationese such as “~를 통해” and “~에 의해” before it reaches the English draft. AI-generated drafts get their own detection layer.
Real deliverables
You get the manuscript, the journal and the report.
Every run produces three files. The journal explains each edit, and the report lets you diff-review them.
- manuscript.md
- manuscript.edits.json
- manuscript.integrity-report.html
You set the edit budget
The skill diagnoses light, standard or heavy. You choose low, mid or high. The effective workload is the smaller of the two, and several intensities arrive as separate files.
manuscript.edits.json · Illustrative sample
- changed“through … by us” → active voice
pitfalls.translationese - changedbandgap / band gap → one notation
terms.notation - kept“was measured” in Methods stays passive
core.section.methods
How it works
Six steps. One of them is not allowed to be an opinion.
Input typing
Korean manuscript (translate, then edit) or English / AI-draft manuscript (edit, with AI-tell layer when needed).
Field routing
Your instruction, then the folder name, then auto-detection over 71 overlays. At most one closed confirmation question if still unclear.
Manuscript state
Optional manuscript.json keeps defined abbreviations and fixed notations consistent across sessions.
Section-aware editing
IMRaD-aware: Methods stay past and passive, Results keep interpretation out, Discussion analyzes in present tense. “Intro only” requests are scope-locked.
Verification
Four deterministic gates run in order. Any failure blocks delivery.
Logging
Every edit pair and journal is logged to feed the benchmarks and the weekly overlay refresh.
Verification gates
Delivery is blocked until every gate exits 0.
| Gate | Script | Blocks delivery when |
|---|---|---|
| 0 · Journal coverage | check_journal.py | any diff hunk lacks a journaled changed/kept entry, a span exceeds 40 tokens, or a cited rule id is missing |
| 1 · Integrity | verify_integrity.py | numbers, units, formulas, citations, equations, DOIs or field terms drift, or the change rate exceeds the budget |
| 2 · Terminology | check_terms.py | notation variants disagree (bandgap vs. band gap) |
| 3 · Abbreviations | check_abbrev.py | an undefined abbreviation appears |
Field corpus coverage
Measured breadth, not a promise of perfection.
A score of 100 is a saturation cap on corpus breadth and stability, not a claim of perfect proofreading. Correction ability is measured separately by a deterministic benchmark. Fields under 10 papers ship overlays flagged as immature, and the skill tells you so.
| Field | Papers | Coverage / 100 |
|---|---|---|
| Optics and photonics | 121 | 100.0 |
| Engineering | 105 | 100.0 |
| Chemistry | 103 | 100.0 |
| Nanoscience and technology | 103 | 100.0 |
| Biochemistry | 51 | 100.0 |
| Mathematics and computing | 105 | 99.3 |
| Biotechnology | 53 | 98.7 |
| Cancer | 53 | 98.0 |
| Neuroscience | 54 | 96.2 |
| Physics | 102 | 95.0 |
| Materials science | 51 | 93.8 |
| Energy science and technology | 53 | 81.2 |
Showing 12 of 71 fields. docs/readiness-fields.html ↗
Built on Claude
Where Claude sits in the product, exactly.
no more DASI is an agent skill. Claude does the language work, and deterministic scripts do the checking.
| Feature | How Claude is used | Value to the author |
|---|---|---|
| Translate-then-edit | Claude translates a non-English draft into English first, with the Korean-author pitfalls layer loaded so literal patterns are not carried over. | Meaning preserved, no translationese. |
| Field-aware editing | Claude edits each manuscript section with the matched field overlay (phrase bank, tense rules, notation list) in context, inside the chosen edit budget. | Wording that matches the target field’s measured conventions. |
| Rule-cited journal | Claude writes edits.json: every changed and every deliberately kept span, tied to the rule that fired. | Every edit is explainable to an advisor. |
| Gate-driven repair | When a script gate fails, its concrete output goes back to Claude to fix. Claude never grades its own work. | Numbers, citations and terms never silently drift. |
Runs in Claude Code and any agent harness that loads Claude Skills.
Get it
Install the skill in under a minute.
The distributable skill lives in skills/nomoredasi/. Link it into your agent’s skills directory, start a new session, and ask for a proofread.
- Clone the repository
- Link the skill into your agent
- Ask: “proofread this paper for Nature”
Just ask your agent. Korean works too.
$ git clone https://github.com/yelixir-dev/no-more-dasi-eng.git
$ ln -s "$(pwd)/no-more-dasi-eng/skills/nomoredasi" \
~/.claude/skills/nomoredasi
> proofread this paper for Nature
nomoredasi v0.1 — type B / field: Optics and photonicsFAQ
Questions authors ask first.
Is this ready for production use?
It is v0.1, early access. The corpus is uneven by field, and the skill says so when an overlay is immature.
Does it change my numbers, units or citations?
No. Numbers, units, formulas, citations, equations and DOIs must survive verbatim, and the integrity gate blocks delivery otherwise.
Which journals and styles does it cover?
The 71 overlays follow Nature’s subject categories. Venues with different house styles (IEEE, ACM) currently ride the field-independent core rules.
Where does the style data come from?
The licensed corpus holds 1,646 active CC BY 4.0 source records. Each paper is listed in the public attribution registry, and the original authors keep their copyright.
Which languages can I write in?
Any. Korean drafts get a dedicated pitfalls layer. Other languages are translated and edited under the core rules without a language-specific pass.
Does it make my prose plain?
No. It keeps academic register on purpose and refuses to oversimplify into plain-English rule sets.
Company
Who is behind it.
no more DASI is independent work: corpus mining, rule distillation and the verification gates are built and maintained by one developer in Songdo, Incheon.
- Business name
- 김잔디
- Business registration no.
- 397-01-04135
- Address
- #317-Ga2, 3F, Bldg. D, 158 Harmony-ro, Yeonsu-gu, Incheon, Republic of Korea (Songdo TimeSpace)
- Contact
- [email protected]