agentleFS
AxectAxect/skills/overleaf-section-workflow/SKILL.md

overleaf-section-workflow

Skill4 starsChanged 20 days agoMITView on GitHub

Bilingual section-by-section workflow for Overleaf physics-paper drafts: Korean draft, user iteration, Opus-direct English LaTeX translation, out-of-tree build, and sync hygiene. Use when starting, iterating, translating, or finalizing an Overleaf paper section; merging Korean drafts into main.tex; verifying citations; enforcing no em/en dashes, no unsupported novelty claims, and no forward references in background sections; or building JCAP/PRD-style sections with figures and BibTeX.

The file

We split this skill into the rules it gives an agent

8 rules. About 0 are common, 5 are distinctive, and 3 are very specific to this project.

  • ≈0

    Common · estimate

    Many files like this have a rule of this kind.

  • ≈5

    Distinctive · estimate

    This author's own choices. Worth a look.

  • 3

    Very specific

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

  • Common
  • Distinctive
  • Very specific
  1. Distinctive:

    Use this skill whenever the user:

    line 19
  2. Distinctive:

    Do not use this skill for one-shot full-paper writes.

    line 26
  3. Distinctive:

    If any of these directories does not exist, surface the gap before proceeding.

    line 38
  4. Very specific:

    Use markdown headings (## §X.Y), markdown bold paragraph leads (X.).

    line 44
  5. Very specific:

    Use \cite{key} exactly as it will appear in LaTeX (preserves verbatim).

    line 47
  6. Very specific:

    Build hygiene: Never run pdflatex, bibtex, or any TeX command inside the Overleaf sync folder.

    line 66
  7. Distinctive:

    Translation is Opus-only: Do not delegate to Sonnet subagents in this skill, even when normal lab convention says to delegate translation.

    line 71
  8. Distinctive:

    Do not invoke these companions silently — surface what you are doing so the user can intervene.

    line 100

≈ “Common” is an estimate: at least 1 in 5 of the 364,980 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

Overleaf section workflow

A disciplined, section-by-section workflow for drafting a physics paper that lives on Overleaf, with a Korean intermediate draft and Opus-direct English translation. Designed to enforce paper-quality discipline learned from real JCAP-style drafting sessions.

When to use

Use this skill whenever the user:

  • Starts a new section of a paper drafted on Overleaf
  • Iterates a Korean draft toward an English LaTeX section
  • Translates a confirmed Korean draft directly into a .tex file
  • Sets up out-of-tree builds in a *_build directory
  • Asks for "section-by-section paper writing", "한글 초안 영어로 병합", "Overleaf 섹션 작성", or similar

Do not use this skill for one-shot full-paper writes; this workflow is per-section.

Three-directory contract

The workflow assumes the user maintains three sibling directories under ~/zbin/OverLeaf/ (or wherever the project lives):

DirectoryRole
~/zbin/OverLeaf/<PROJECT>/Sync folder mirrored to Overleaf via the overleap skill. .tex, ref.bib, figs/ live here. Never build here — build artefacts would sync back to Overleaf.
~/zbin/OverLeaf/<PROJECT>_draft/Korean drafts as section<N>_ko.md (one file per section). Plots embedded as markdown images. Iterated with the user before translation.
~/zbin/OverLeaf/<PROJECT>_build/Out-of-tree build. Contains build_<jobname>.sh and all build artefacts (.aux, .log, .pdf, .bbl). Style files (.bst, .sty) and ref.bib are copied here at build time.

If any of these directories does not exist, surface the gap before proceeding.

Workflow loop (per section)

  1. Sync the Overleaf project via overleap if not already mirrored.
  2. Draft the section in Korean at <PROJECT>_draft/section<N>_ko.md:
    • Use markdown headings (## §X.Y), markdown bold paragraph leads (**X.**).
    • Math inline as $...$, display as $$...$$ with explicit \tag{X.Y} for cross-reference.
    • Embed plots as markdown images: ![caption](figs_<topic>/foo.png).
    • Use \cite{key} exactly as it will appear in LaTeX (preserves verbatim).
    • Build up narrative as: motivation → introduce families → unify → identify limitation → contribution (see references/06_section_narrative.md).
  3. Iterate with the user. Specifically watch for:
    • Em-dashes (—, U+2014) or en-dashes (–, U+2013) — both are AI-style markers; replace with periods, commas, parens.
    • Forward references (§\ref{...}, "본 연구의 §5 에서 다룬다") in background sections.
    • Absolute language ("표준이다", "확실하다") that should be softened.
    • Novelty claims without hedging.
    • Citations whose attribution doesn't match the cited paper's actual content.
    • Plot layout issues (legend on top of curves, xlim/ylim wasting empty space).
  4. User confirms the Korean draft is paper-grade.
  5. Translate Opus-direct into English LaTeX at <PROJECT>/<jobname>.tex. NEVER use a Sonnet subagent for translation in this skill — the user has explicitly required Opus-only. Conversion rules in references/05_translation_rules.md.
  6. Copy figure PDFs from ~/Documents/Project/Research/.../outputs/<YYYYMMDD>_<topic>/plots/ to <PROJECT>/figs/.
  7. Verify the build via bash <PROJECT>_build/build_<jobname>.sh. Build must produce zero undefined warnings.
  8. Commit-triage the working tree at session boundaries (use the commit-triage skill).

Core rules (non-negotiable)

These come from real session pain points. Do not skip.

  1. Build hygiene: Never run pdflatex, bibtex, or any TeX command inside the Overleaf sync folder. Always build in <PROJECT>_build/ via the supplied build_<jobname>.sh. If the user sees *.aux, *.log, *.pdf inside the sync folder, treat that as an emergency and clean it up.
  2. No em-dashes or en-dashes in any draft or final LaTeX. They are AI-style markers; verify after every edit with grep -cP "[\x{2013}\x{2014}]" <file> (must return 0).
  3. No forward references in background sections (§Background / §Setup). Subsequent sections may forward-reference earlier ones; earlier sections must stand alone.
  4. No "본 연구" preview outside design statements. "본 연구는 GBP 를 채택한다" is fine. "본 연구의 §5 에서 다루는 DeCLA" is not.
  5. Citation must support the actual claim, not just be metadata-correct. After bibtex generation, verify that the cited paper says what the body claims it says.
  6. Translation is Opus-only. Do not delegate to Sonnet subagents in this skill, even when normal lab convention says to delegate translation. The user explicitly forbids this because translation goes directly into the paper.
  7. Plots use scienceplots ["science", "nature"] with dpi=300, no title, no font override, no no-latex flag. Save both PNG and PDF: PNG into the Korean draft, PDF into <PROJECT>/figs/. Legend placement uses bbox_to_anchor=(1.02, 0.5) with loc="center left" whenever data would overlap the legend.

Resources

  • references/00_workflow_overview.md — sequential workflow steps with checkpoints.
  • references/01_writing_style.md — Korean / English style rules (em-dash ban, forward-ref ban, hedged claims, paper-grade tone).
  • references/02_citation_verification.md — three-source verification (InspireHEP / CrossRef / Scholar) and content-vs-metadata distinction.
  • references/03_plotting_conventions.md — scienceplots conventions, legend placement, xlim/ylim tuning, plot-script directory layout.
  • references/04_build_hygiene.md — out-of-tree build pattern, build.sh template, what to do when sync folder is contaminated.
  • references/05_translation_rules.md — Korean markdown → English LaTeX conversion rules (with worked examples).
  • references/06_section_narrative.md — narrative structure (motivation → introduce → unify → limitation → contribution) and how to slot a contribution claim safely.
  • references/07_lessons_log.md — incidents and lessons from prior sessions, grouped by category.
  • templates/build_template.sh — copy-and-rename build script (set JOBNAME).

Companion skills

The workflow orchestrates several existing skills. Invoke them in turn:

StageSkill
Syncoverleap
Plot generationscienceplot-py (or write inline if scope is small)
Plot regeneration / legend fixdirect Bash + uv run
Reference searchreference-search (broad lit)
Novelty verificationdeep-research in lit-review / fact-check mode
BibTeX generationbibtex-gen
Commitcommit-triage

Do not invoke these companions silently — surface what you are doing so the user can intervene.

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.