English | 简体中文

RE-paper-writing

An authoritative skills repository for academic paper planning, literature grounding, section drafting, revision, and LaTeX-based submission workflows.

This repository is a curated home for skills that directly support writing and revising research papers. The canonical loadable skill set lives under skills/. Other root-level directories are source snapshots and upstream references, not the normalized skill catalog.

Most projects should use a subset of these skills. The repository is intended as a staged workflow with clear entry points, not as a checklist that every paper must invoke from start to finish.

Positioning

• Keep only skills that directly support paper planning, literature review for writing, section drafting, citation work, tables, LaTeX preparation, self-review, revision, and rebuttal writing
• Normalize upstream content into a portable repository layout with repository-relative paths instead of one author's local machine paths
• Treat skills/ as the authoritative directory for future paper-writing-related skill work

Included Skills

The curated collection currently contains 35 skills:

Planning and Literature Grounding

• research-planning: turn a topic or draft idea into a paper plan and section architecture
• literature-search: multi-source academic search with structured JSONL output
• systematic-review: artifact-first systematic review pipeline for building a reusable paper corpus
• literature-review: synthesize a collected corpus through multi-perspective review
• source-material-ingestion: convert PDFs, DOCX, PPTX, XLSX, and related source files into Markdown writing inputs
• citation-management: validate, harvest, repair, and export BibTeX references
• paper-memory-ledger: persist stable project facts, venue choices, experiment conclusions, and writing conventions across sessions

Drafting and Manuscript Construction

• experiment-report-bridge: turn local experiment artifacts into a paper-ready Markdown report before full manuscript drafting
• ml-paper-writing: orchestrate a complete paper draft from a research repo or mature result set
• latex-writeup-loop: write directly inside template.tex with section, citation, and compile loops
• long-form-manuscript-polish: convert bullet-heavy or slide-like long drafts into professional prose-first manuscripts
• reverse-outline-flow-check: diagnose section flow using reverse outlines, paragraph roles, and logical transitions
• claim-evidence-map: map headline claims to explicit support anchors and downgrade unsupported wording
• experiment-section-audit: redesign Experiments around contribution-linked validation, strong baselines, and reviewer-readable evidence
• related-work-writing: write thematic, comparative Related Work sections
• survey-generation: generate survey manuscripts from a curated literature corpus
• paper-writing-section: draft or revise one paper section at a time
• table-generation: convert result files into publication-ready LaTeX tables
• paper-schematics: design graphical abstracts, method flowcharts, and other conceptual manuscript figures
• academic-plotting: create numeric plots and other data-backed publication figures

Submission, Review, and Revision

• paper-session-compaction: compact long drafting or revision sessions into a reliable continuation brief
• venue-submission-strategy: adapt framing, page budget, and reviewer-facing packaging to a target venue
• reporting-guidelines-check: audit compliance with study-type standards such as CONSORT, STROBE, and PRISMA
• claim-rigor-audit: stress-test claims for bias, confounding, evidence strength, and statistical overreach
• manuscript-scorecard: score manuscript quality dimension by dimension and track revision progress
• latex-formatting: venue-aware LaTeX cleanup and format checks
• submission-qa-gate: run deterministic pre-submission checks for labels, TODO markers, citation coverage, and compile readiness
• paper-compilation: compile papers and debug LaTeX build failures
• paper-evidence-verifier: verify that manuscript numbers and conditions match real experiment artifacts
• citation-verification-gate: verify references against real APIs and prune hallucinated citations
• self-review: run structured pre-submission paper review
• review-ensemble: run multi-reviewer, meta-review-style manuscript assessment
• paper-revision: map reviewer feedback to concrete manuscript edits
• paper-redline-diff: generate change-tracked manuscript diffs for revision review and rebuttal support
• rebuttal-writing: produce point-by-point reviewer responses

See skills/README.md for the skill catalog and routing guidance.

Canonical Workflow

1. Frame the paper and build the evidence base: research-planning -> literature-search -> systematic-review or literature-review
2. Normalize source material and keep durable context: source-material-ingestion -> paper-memory-ledger
3. Convert finished experiments into a writing handoff: experiment-report-bridge
4. Choose one primary drafting engine: ml-paper-writing for whole-manuscript drafting, latex-writeup-loop for in-template LaTeX drafting, paper-writing-section / related-work-writing / survey-generation for section-scoped work
5. Apply section-level specialists only where needed: reverse-outline-flow-check, long-form-manuscript-polish, claim-evidence-map, experiment-section-audit, table-generation, paper-schematics, academic-plotting
6. Run trust and compliance checks before export: citation-management, paper-evidence-verifier, claim-rigor-audit, citation-verification-gate, reporting-guidelines-check, venue-submission-strategy
7. Run the submission stack: latex-formatting -> submission-qa-gate -> paper-compilation
8. Run the review loop: self-review or review-ensemble -> paper-revision -> paper-redline-diff -> rebuttal-writing
9. Use continuity tools only when needed: paper-session-compaction for long-thread handoff, paper-memory-ledger for durable project facts

Boundary Rules

• Choose one primary drafting engine at a time. Do not default to mixing ml-paper-writing, latex-writeup-loop, and paper-writing-section in the same step.
• Use paper-schematics for conceptual, non-numeric figures. Use academic-plotting for numeric or data-backed visuals.
• Use claim-evidence-map for paper-visible support mapping, paper-evidence-verifier for artifact-grounded numeric checks, and claim-rigor-audit for inference validity.
• Use submission-qa-gate for overall readiness. Use paper-compilation only when LaTeX build/debug is the bottleneck.
• Use self-review for a quick single pass. Use review-ensemble only when you need a stricter committee-like signal.
• Use paper-memory-ledger for stable cross-session facts. Use paper-session-compaction for one long thread that needs a compact handoff.

Shared artifact conventions:

• outputs/<topic-slug>/search_results/*.jsonl for raw retrieval results
• outputs/<topic-slug>/paper_db.jsonl for the shared literature corpus
• outputs/<topic-slug>/reading_notes.md, synthesis.md, and gaps.md for intermediate review artifacts
• references.bib for the active bibliography

Repository Layout

skills/
  citation-management/
  latex-formatting/
  literature-review/
  literature-search/
  source-material-ingestion/
  paper-memory-ledger/
  experiment-report-bridge/
  ml-paper-writing/
  latex-writeup-loop/
  long-form-manuscript-polish/
  reverse-outline-flow-check/
  claim-evidence-map/
  experiment-section-audit/
  paper-session-compaction/
  claim-rigor-audit/
  manuscript-scorecard/
  paper-schematics/
  reporting-guidelines-check/
  venue-submission-strategy/
  paper-evidence-verifier/
  submission-qa-gate/
  paper-compilation/
  paper-revision/
  paper-redline-diff/
  paper-writing-section/
  rebuttal-writing/
  related-work-writing/
  research-planning/
  review-ensemble/
  self-review/
  survey-generation/
  systematic-review/
  table-generation/
  academic-plotting/
  citation-verification-gate/

Usage

Commands below assume you run them from the repository root.

Example: search and build a portable literature corpus

python skills/systematic-review/scripts/search_semantic_scholar.py \
  --query "long-context reasoning agents" \
  --max-results 20 \
  --api-key "$S2_API_KEY" \
  -o outputs/long-context-reasoning-agents/search_results/s2.jsonl

python skills/systematic-review/scripts/bibtex_manager.py \
  --jsonl outputs/long-context-reasoning-agents/paper_db.jsonl \
  --output outputs/long-context-reasoning-agents/references.bib

Recommended environment:

• Python 3.10+
• Optional environment variable: S2_API_KEY
• Optional dependency: PyMuPDF for PDF extraction in self-review and systematic-review

Curation Rules

• A skill must directly improve paper planning, writing, citation integrity, review quality, or LaTeX submission readiness
• Skills for experiment execution, generic code generation, GitHub repo mining, or presentation-only workflows do not belong in this curated set
• Prefer scriptable and auditable skills over opaque prompt bundles when the task is repetitive or fragile

Provenance

The current curated skill set was extracted from the agent-research-skills/ source snapshot already present in this repository and normalized into the authoritative skills/ layout.

Additional paper-writing skills were also extracted from the AI-research-SKILLs/20-ml-paper-writing/ source snapshot and normalized into the same catalog.

Additional workflow-specific skills were extracted from ai-scientist/, especially its LaTeX writeup and ensemble review loops.

Additional verification-oriented skills were extracted from AutoResearchClaw/, especially its evidence-grounding and citation-audit gates.

Additional submission-oriented skills were extracted from claude-scientific-skills/, especially its venue strategy, reporting-compliance, and manuscript-schematic guidance.

Additional writing-tooling skills were extracted from claude-scientific-writer/, especially document ingestion, quantitative manuscript scoring, and claim-rigor audit workflows.

Additional session-oriented writing skills were extracted from EvoScientist/, especially its artifact-grounded writing agent, context-compaction flow, and persistent project memory conventions.

Additional LaTeX-adjacent paper-writing skills were extracted from latex-document-skill/, especially its revision redline workflow, long-form manuscript anti-pattern guidance, and pre-submission QA checks.

Additional reviewer-facing writing skills were extracted from Research-Paper-Writing-Skills/, especially its reverse-outlining workflow, claim-support alignment checks, and experiments-writing guidance.

Notable curation decisions:

• deep-research was renamed and simplified as systematic-review
• Hard-coded user-local paths were replaced with repository-relative references
• Only paper-writing-related skills were retained; experiment, coding, and general research-automation skills were excluded