Home

Claude skills for journal article writing

Baca dalam bahasa Indonesia

Six skills for writing, illustrating, polishing, submitting, and revising journal articles at Q1 (Scopus/WoS) standard — including one that runs a systematic literature review from nothing to a submission-ready manuscript.

They work in your language — ask in English, get English; ask in Indonesian, get Indonesian. The manuscript's own language follows the target journal, independently of the conversation.

Just want to get started? Download a skill and upload it to Claude Desktop — Installation has the direct download links and four numbered steps. No git, no terminal.

Which one do I need?

Start from where you are, not from the skill names.

Your situation	Skill
You have data and results but no idea how to assemble an article	nulis
Your figures all end up as bar charts, or a reviewer called them uninformative	visualisasi-data
You need journal-spec figures for .tex or .docx — column width, fonts, 300 dpi	visualisasi-data
The draft is done but reads stiff, long-winded, or "AI-ish"	polish-manuscript
The manuscript is finished and about to go to a journal	submit
It was rejected without review and you do not know why	submit
You received a major/minor revision decision and reviewer comments	revisi
You want to write a systematic review	slr-cowork
A reviewer said your SLR method is indefensible	slr-cowork

Not sure? See Workflow — how the six connect.

What each one does

nulis — structure. Coaching built on genre analysis: it guides you move by move, maps gap → RQ → design → results → contribution as one line per RQ running through five sections, and calibrates how boldly claims may be stated for your field. Four modes: outline, draft a section, audit an existing draft, refine a passage.

visualisasi-data — figures. Picks the visual form from what a row of data actually is — scalp positions become a topography, regions become a map, studies in a synthesis become a forest plot — instead of defaulting to bars. Then holds the figure to data fidelity, visible spread, colour-blind safety, and journal print specification. Fourteen domain routes, and a structure table so an unlisted field is never a dead end.

polish-manuscript — prose. A ten-dimension audit of a draft whose structure is already sound. Its distinguishing feature is a fidelity gate: every number and citation present before editing must still be present after, or the passage is reverted and reported.

submit — the gate. Sweeps for desk rejection risk, which affects 40–70% of submissions and mostly has nothing to do with scientific quality. It runs cheapest-and-most-lethal first and stops at the first fatal finding.

revisi — after the decision. Breaks reviewer comments into trackable items, decides each on its merits rather than by vote count, and writes a response letter that only claims changes that actually exist — and tells the editor where to find them.

slr-cowork — systematic review. Nine stages from an agreement form to a PRISMA 2020 manuscript, with gates that catch unanswerable questions early and arithmetic that has to reconcile. Handles doctrinal-normative corpora where clinical instruments have no object.

Scope

Any field, any research design. What differs between fields is not the workflow but the convention — and these skills pick the right convention instead of flattening everything.

The first step in every session is to ask your field and research type, then load the matching conventions. The calibration is real, not decorative:

• Claim strength. Natural and life sciences use boosters (show, demonstrate) most freely; mathematics and physical sciences most sparingly; humanities hedge most. Flattening this makes a manuscript read as though it came from the wrong room.
• Structure. IMRaD is not a law of nature. Pure mathematics uses theorem–proof convention; computer science often IDBRC; humanities frequently essay-style. Nothing is forced.
• Reporting standards. Follow the research type, not the field: CONSORT, PRISMA, COREQ, SRQR, GRAMMS, APA JARS — selected to match your design.
• Language. The conversation follows yours; the manuscript follows the target journal. The two are separate — you can discuss an English manuscript in Indonesian.

For non-English corpora, the search and appraisal paths are adapted rather than forced — see slr-cowork and the FAQ.

What sets them apart

Gates that stop, not lists that sprawl. submit runs in order from cheapest and most lethal. One fatal finding makes the rest irrelevant: it stops, reports, and does not carry on sweeping. No point polishing hyphens in a manuscript aimed at the wrong journal.

What can be counted is counted. Acronyms, spelling, p-values, units, word limits, two-way citation cross-checks, PRISMA cascade arithmetic — all handled by deterministic scripts you can re-run, not judged by impression.

Citations are never invented. A tiered ladder: MCP tools if present, then DOI resolution, then an explicit "unverified" flag. The rule that holds at all three tiers: a citation is never accepted because it looks plausible — a plausible-looking author–year–journal combination is the signature pattern of a fabricated reference.

Editing must not destroy evidence. polish-manuscript has a fidelity gate: every number and every citation present before editing must still be present after. If one goes missing, that passage is reverted and reported — not quietly patched.

A step that cannot run is not counted as passed. A script that fails to execute is reported as a skipped step, never as "a problem with the manuscript". The two are never mixed in one findings table — and the skill is required to say what it skipped.

Each skill stands alone. Fully functional on its own. If a neighbour is installed, some steps go deeper — but nothing stalls because another skill is absent.

---

Installation · Workflow · Requirements · FAQ · Licence