agentleFS
fortunto2fortunto2/solo-factory/skills/paper/SKILL.md

solo-paper

Skill18 starsChanged 4 months agoMITView on GitHub

Turn a research result (math, algorithms, video/audio, systems, datasets) into a solid preprint or conference paper — LaTeX (amsart/article) built locally with tectonic, written from the project's research journal, references verified against databases, reviewed for errors and overclaims, code and data with a DOI, then submitted to arXiv (endorsement, category, license, AI-use disclosure). Use when the user says "напиши статью", "препринт", "на архив", "arXiv", "оформи результат", "статья на конференцию", "куда подать статью", "paper draft", "write this up", or a result is ready to be published. Do NOT use for publishing datasets alone (use /open-dataset), for printable PDFs from markdown (use /pdf-print), or for blog posts (use /content-gen).

  • Installs packages

The file

We split this skill into the rules it gives an agent

6 rules. About 0 are common, 4 are distinctive, and 2 are very specific to this project.

  • ≈0

    Common · estimate

    Many files like this have a rule of this kind.

  • ≈4

    Distinctive · estimate

    This author's own choices. Worth a look.

  • 2

    Very specific

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

  • Common
  • Distinctive
  • Very specific
  1. Distinctive:

    Check the venue's AI-use policy and anonymity rules before uploading the preprint.

    line 38
  2. Very specific:

    Literature check first: Search for newer work on the exact question: OpenAlex API …

    line 48
  3. Distinctive:

    Put what you found in the related-work paragraph, including results that make yours smaller.

    line 52
  4. Very specific:

    References from databases only: Every \bibitem/BibTeX entry is copied from OpenAlex/Crossref/arXiv/zbMATH metadata (DOI or arXiv id), never …

    line 62
  5. Distinctive:

    Every reference resolves (DOI/arXiv id opens the right work).

    line 93
  6. Distinctive:

    Every number in the PDF appears in the journal or artifacts.

    line 94

≈ “Common” is an estimate: at least 1 in 5 of the 372,429 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

paper — from a verified result to an arXiv preprint

A preprint is judged in the first page: a precise claim, its scope, and how a reader can check it. Write it from what was recorded while the work happened (journals, logs, code), not from memory, and publish nothing a reviewer has not tried to break.

Before writing: is there a paper?

Yes if there is a new, checkable statement (theorem/lemma, classification, bound, construction, reproducible measurement) that survived a literature check. No for a leaderboard position, a polish of someone else's construction, or a negative result too thin to stand alone (an appendix at most). Say which, in one line, before drafting.

Choose the venue (any field, not only mathematics)

ResultFirst stopPeer-reviewed target
Mathematics (lemma, classification, construction)arXiv math.*combinatorics journals (e.g. Electron. J. Combin., Discrete Math.)
Algorithms, systems, video/audio/visionarXiv cs.* / eess.* (cs.CV, cs.MM, eess.IV, eess.AS)conference workshops first (CVPR/ICCV/ECCV, ACM MM, ICASSP, ICIP, NeurIPS): shorter, faster, higher acceptance; then main tracks or journals
Astronomy, sky/sensor measurementsarXiv astro-ph.IM / astro-ph.EP; Research Notes of the AAS (short, fast)PASP, MNRAS instrumentation papers, citizen-science journals
Biology, bioinformaticsbioRxiv / arXiv q-biofield journals; method papers in Bioinformatics, PLOS Comput Biol
Tools, datasets, benchmarksarXiv + DOI'd repodataset/benchmark tracks (NeurIPS D&B), JOSS for research software
  • A preprint establishes priority within days. Peer review is what counts as a publication in most evaluations, including talent/extraordinary-ability visas, grant panels and hiring. Plan both: the arXiv version and one reviewed venue.
  • Check the venue's AI-use policy and anonymity rules before uploading the preprint; most allow arXiv, some restrict publicity during review.
  • Reviewing for a conference or journal is a separate, citable professional activity (program-committee member, workshop reviewer). Volunteer for workshops in your area once you have one paper there; keep the invitation and confirmation emails.
  • Conference papers usually need a registered author to present, which means travel or visa logistics and a fee. Check this before submitting.

Workflow

  1. Literature check first. Search for newer work on the exact question: OpenAlex API (api.openalex.org/works?search=…&sort=publication_date:desc, no key), arXiv listing, zbMATH Open, the community site of the problem (e.g. erdosproblems.com), a web search. Feed each source a query with a known answer first: a silent zero is not "nothing new". Put what you found in the related-work paragraph, including results that make yours smaller.
  2. Skeleton from the journal. Map research-journal entries to sections: goal → abstract/introduction; what lines in order → method (keep the dead ends that justify the final method); result lines → results, numbers verbatim; controls → their own subsection; artifacts → reproducibility.
  3. LaTeX. docs/paper-<slug>/main.tex; amsart for mathematics, article otherwise. Theorem environments, one table per numeric claim, \url for code. Build: brew install tectonic once, then tectonic main.tex (fetches packages itself). Fix every Overfull/undefined warning.
  4. References from databases only. Every \bibitem/BibTeX entry is copied from OpenAlex/Crossref/arXiv/zbMATH metadata (DOI or arXiv id), never typed from memory.
  5. Code and data with a DOI (/open-dataset): public repo, CITATION.cff, release, Zenodo DOI; cite it in the paper. arXiv also accepts small ancillary files (anc/).
  6. Adversarial review. A separate reviewer (strongest model available, briefed to refute) checks: is every claim implied by the evidence, scope stated (computer-assisted, what it depends on), numbers match the artifacts, overclaims. Fix, then a light /humanize pass on prose only, and diff every number before and after: identical.
  7. Submit (outward-facing: the author decides). See below.

arXiv submission

  • Author: the human(s) who direct the work and take responsibility. AI systems are not authors; disclose their use in the text (method or acknowledgements), e.g. "The search was designed and run with AI agents (…) under the author's direction".
  • Endorsement: a first submission to a category usually needs an endorser, someone who has published in that category recently. Start the submission, get the endorsement code from arXiv, and email it to a suitable person: the authors of the paper you build on are the natural choice. Send a short note with the result and the PDF, and attach no code unless they ask. Draft the email; the author sends it.
  • Category: the narrowest fitting primary (e.g. math.CO), at most one or two cross-lists.
  • License: CC BY 4.0 matches open data and code.
  • Metadata: title and abstract as plain text with TeX math; comments field gives the page count and "code: <DOI>".
  • Identity: an ORCID linked to the arXiv account, and the same name form in the paper, CITATION.cff and Zenodo.

Verification

  • tectonic main.tex builds with zero warnings.
  • Every reference resolves (DOI/arXiv id opens the right work).
  • Every number in the PDF appears in the journal or artifacts.
  • The reviewer's blocking list is empty and their changes are applied.

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.