WantongC/journal-adapt-writing-skill: a corpus-grounded revision layer for one manuscript and one journal
Learn any journal's writing conventions from its published papers, then revise your manuscript to match — section by section.
At a glance
- What is it?
- The skill does not write your paper. It reads the target journal's published papers, builds an editable dynamic_writing_skill.md, and gives an agent a section-by-section revision framework with a priority order you can audit.
- Who is it for?
- Adopt it if you already have a readable corpus of 5 to 8 target-journal papers in Markdown and you want the revision rules to be visible and editable rather than buried in a prompt. Do not adopt it if you expect it to draft a paper, or if your corpus is still locked in scanned PDFs, because the README states that all corpus files should be fully readable Markdown or text before Phase 1.
- Can I use it commercially?
- Yes. MIT is a permissive licence: you can use, modify and sell software built on it, as long as you keep its copyright and licence notices.
- Is it still maintained?
- Yes. The repository last received commits 124 days ago.
- What is it written in?
- GitHub does not report a main language for this repository.
Answers come from the project's GitHub data, last synced on September 15, 2026, and from our analysis. They are not legal advice.
DEEP OPEN-SOURCE ANALYSIS
The gap journal-adapt-writing-skill targets: same manuscript, different journal
Discipline templates and anti-AI phrasing rules are reusable, and that is exactly their limitation. They apply the same instructions to every manuscript regardless of where it is going. The README frames the missing piece as journal adaptation: writing for Journal A and writing for Journal B may require different introduction logic, contribution framing, method exposition, result emphasis, and discussion scope, even when the underlying research is identical.
The tool is for researchers who already have a draft and a destination, and who want the revision rules to come from the destination's own published papers rather than from a general style guide. It is explicitly not an auto-writer. The README states that it does not auto-write a paper; the output is a visible, editable dynamic_writing_skill.md plus a revision framework the agent applies section by section. If you want a system that produces prose from scratch, this is the wrong layer.
Static base rules plus a generated dynamic layer
The architecture is two layers. Static skills are reusable base rules: discipline conventions, general academic constraints, anti-AI phrasing cleanup, citation and equation safety, or your own lab guide. The static layer is optional, and the README allows skipping it so the dynamic corpus drives the workflow alone.
The dynamic layer is generated per manuscript and per destination from a user-provided corpus. That corpus has three roles. Target-journal papers are the primary corpus and are required; the README recommends 5 to 8 papers and says they contribute the journal's local writing culture. Field-top or topic-similar papers are optional, 2 to 5 recommended, useful when the target-journal corpus is small. User or lab exemplars are optional, 1 to 3 documents, and preserve author, advisor or lab preferences when they do not conflict with the target journal.
The data flow runs from those three inputs into a corpus style profile, then into the dynamic writing skill, then through human review, then section revision, then a revision log. The README lists the static layer as an optional input feeding the same dynamic skill, which means you can bring an existing skill or none at all.
The P1 to P5 priority table is the part worth reading twice
Most writing tools bury their conflict resolution. This one publishes it. P1 is hard constraints: preserve facts, citations, equations, notation, numerical results, labels, and author-defined terminology. P2 is the target journal corpus, and the README says P2 usually beats P3 and P4. P3 is the secondary corpus and exemplars, used when target-journal evidence is absent or weak. P4 is the static base skill, applied when corpus signals do not decide. P5 is cleanup: removing AI-taste phrases, hollow transitions, generic contributions, and unsupported overclaims.
The README states that P1 always wins, and that any conflict changing revision behavior should be recorded in the revision log. That log is the audit trail. It is also the honest part of the design: the tool acknowledges that a journal's conventions can conflict with your lab's preferences or your static guide, and it makes you decide rather than silently picking one. The trade-off is that you have to read the log. A revision that follows P2 over P3 without explanation is a decision you agreed to by not overriding it.
Installing the skill and running a first revision
For Claude Code, the README gives a two-command install that copies the skill folder into your local skills directory:
mkdir -p ~/.claude/skills/journal-adapt
cp -R skill/* ~/.claude/skills/journal-adapt/For Codex, the README says to copy or symlink the skill/ folder into your Codex skills directory if your local setup supports custom skills, or to keep the repository open and ask Codex to use skill/SKILL.md directly. There is no package manager step and no version pinning in the documented install.
Next, prepare inputs. The minimum Markdown workflow does not require MinerU:
my_project/
├── corpus/
│ ├── target_journal_001.md
│ ├── target_journal_002.md
│ └── field_top_paper_001.md
└── manuscript.mdThe PDF route mirrors that layout with corpus_pdfs/ and manuscript.pdf, and requires a PDF-to-Markdown converter. MinerU is supported, but the README recommends Markdown input if MinerU is hard to install. The README's quick start section is truncated at the invocation step, so what you should see after running the skill is not documented there; consult docs/INSTALLATION.md, which the README links for installation and PDF conversion detail.
Where the corpus preparation breaks down
The README is direct about the failure mode: all corpus files should be fully readable Markdown or text before Phase 1, and if a PDF conversion is incomplete you should retry conversion, use another converter, provide clean Markdown or text, or replace the paper. That is a real constraint, not a formality. A two-column journal PDF converted with dropped equations or merged captions gives the style profile bad evidence, and the profile then drives P2 decisions across every section.
The second limitation is scale. Five to eight target-journal papers is a small sample. If the journal's conventions vary by article type (empirical versus theory, short format versus full length), a corpus that mixes them produces a blended profile with no way to tell which pattern came from which. The README does not document how the skill separates article types within a corpus, so if your target journal publishes heterogeneous formats, curate the corpus yourself rather than relying on the tool to sort it.
The third is scope. The README states the skill does not auto-write a paper. Teams looking for a drafting pipeline will find a revision framework instead, and the human review step in the flowchart is not optional decoration. This is the wrong tool if you have no draft yet.
How it differs from a general academic writing skill
The README names several static skills it can sit on top of, including hanlulong/econ-writing-skill for economics and referee-style guidance, Master-cai/Research-Paper-Writing-Skills for ML, CV and NLP papers, SNL-UCSB/paper-writing-skill for CS systems and networking, lishix520/academic-paper-skills for philosophy and interdisciplinary work, and blader/humanizer for generic AI-writing cleanup.
The difference in approach is where the rules come from. A static skill encodes a field's conventions once and applies them everywhere. journal-adapt generates rules per destination from that destination's published papers, and layers them over the static rules through the P1 to P5 order. If your target is a general field convention, a static skill alone is cheaper and simpler. If your problem is that two journals in the same field want different introductions and different contribution framing, the static skill has nothing to say about that distinction, and the corpus layer is the only part of the stack that does. The README also permits a custom static input (your own SKILL.md, a lab guide, a journal checklist) or no static skill at all.
Licence, maintenance and what an upgrade costs you
The repository is MIT licensed, which permits reuse and modification with the usual attribution and warranty terms; that is a statement about the licence text, not legal advice, and you should read LICENSE for the operative wording. The repository is not archived. The last push was on 2026-05-15, and the most recent release listed is v1.0 from 2026-05-13, while the README badge shows version 1.1, so the version badge and the release list do not agree.
Upgrade cost is low in the mechanical sense: the install is a folder copy, and the artefact you care about, dynamic_writing_skill.md, is generated per manuscript rather than shipped with the tool. That also means an upgrade does not silently change your existing skills. What it can change is the generation behaviour, so if you have tuned a dynamic skill by hand, keep it outside the copied skill/ directory. The README does not document a rollback procedure or a version compatibility matrix, and the docs/INSTALLATION.md link is the only upgrade-adjacent reference given.
Editorial conclusion
Adopt it if you already have a readable corpus of 5 to 8 target-journal papers in Markdown and you want the revision rules to be visible and editable rather than buried in a prompt. Do not adopt it if you expect it to draft a paper, or if your corpus is still locked in scanned PDFs, because the README states that all corpus files should be fully readable Markdown or text before Phase 1. Before you start, verify two things: that the PDF conversion actually produced complete files, and that your P1 hard constraints (citations, equations, notation, numerical results, terminology) are written down, since the priority table puts them above every corpus pattern.
Frequently asked questions
What is journal-adapt-writing-skill?
It is a static plus dynamic academic writing skill framework. The dynamic layer is generated for one manuscript and one writing destination from a user-provided corpus of target-journal papers, optional field-top papers, and optional user or lab exemplars, producing an editable dynamic_writing_skill.md.
What are the five priority levels in journal-adapt-writing-skill?
P1 hard constraints (facts, citations, equations, notation, numerical results, labels, terminology), P2 the target journal corpus, P3 secondary corpus and exemplars, P4 the static base skill, and P5 cleanup rules. The README states that P1 always wins and that P2 usually beats P3 and P4.
Can journal-adapt-writing-skill write the paper for me?
No. The README states that it does not auto-write a paper; it gives the agent an auditable revision framework for section-by-section academic rewriting, with a human review step before section revision.
How many papers do I need for the corpus in journal-adapt-writing-skill?
The README recommends 5 to 8 target-journal papers as the primary corpus, 2 to 5 field-top or topic-similar papers as the optional secondary corpus, and 1 to 3 user or lab exemplars. All corpus files should be fully readable Markdown or text before Phase 1.
Community notes