| name | paper-write |
| description | Full paper writing orchestration — from blank LaTeX project to camera-ready PDF. Delegates to paper-intro-writing (Introduction), paper-related (Related Work), and reads research-refine/experiment-plan outputs. Use when user says "写论文", "write paper", "开始写 paper", "paper writing", "draft paper", or wants to produce a submission-ready paper from experiment results. |
Paper Write: Full Paper Writing Orchestration
Orchestrate blank LaTeX → camera-ready PDF. Delegates to atomics; does NOT reimplement. Writing philosophy and technique references live in references/writing-guide.md.
Pipeline
Experiment results → refine-logs/
│
├─ P0: LaTeX init (template, Git, CI)
├─ P1: Nugget freeze (single-ping test)
├─ P2: Empty tables & figures (Eisner SOP)
├─ P3: Abstract (Black Mad-Libs)
├─ P4: Introduction → delegate to paper-intro-writing
├─ P5: Method → from refine-logs/FINAL_PROPOSAL.md
├─ P6: Experiments → from refine-logs/EXPERIMENT_PLAN.md
├─ P7: Related Work → delegate to paper-related
├─ P8: Conclusion
├─ P9: Ernst 4-round revision
└─ P10: Submission checklist
The writing philosophy, sentence-level techniques, and ESL pitfalls are in references/writing-guide.md. Read it before starting.
Parameters
Parse from $ARGUMENTS:
--venue <name> — ICLR, NeurIPS, ICML, CVPR, CoRL, ICRA, RSS--overleaf — use Overleaf instead of local LaTeX--skip-intro — skip P4 (paper-intro-writing already run)--light — skip P9-P10
Phases
P0: LaTeX Init
mkdir -p paper/{figures,data,tables} && cd paper && git init
P1: Nugget Freeze
Write one sentence each: 中文 nugget + English nugget. If you can't, the paper isn't ready.
P2: Empty Tables & Figures
From EXPERIMENT_PLAN.md: create empty tables with headers + captions, placeholder figures. \input{} them into main.tex.
P3: Abstract
Fill in Black Mad-Libs template. Iterate 2-3 rounds. Structure words: Unfortunately → In contrast → While promising → Therefore.
P4: Introduction → paper-intro-writing
Delegate. Read its output, verify Widom 5-paragraph structure, verify topic-sentence audit passes.
P5: Method
Read refine-logs/FINAL_PROPOSAL.md. Transform proposal → paper Method section. Start with running example. Every equation: text + figure + equation.
P6: Experiments
Read refine-logs/EXPERIMENT_PLAN.md. Transform plan → paper Experiments section. Headline table first. Each subsection = one claim. Every number answers "compared to what?"
P7: Related Work → paper-related
Run paper-related on the paper's own arXiv ID + key competitors. Organize thematically (not laundry list). Each paragraph ends with gap → "In contrast, we…"
P8: Conclusion
Four paragraphs: restate nugget → summarize results → limitations (2-3, honest) → future work. Tokekar Q5 must be non-empty.
P9: Ernst 4-Round Revision
Round 1: section-level (serves thesis?). Round 2: paragraph-level (single message? topic sentence?). Round 3: sentence-level (Williams subject-verb, Zinsser clutter removal). Round 4: word-level (LARG micro-rules, linter, TTS read-aloud).
P10: Submission Checklist
See references/writing-guide.md Appendix A for T-4w / T-2w / T-1w / T-1d checklists.
Output
paper/
├── main.tex, references.bib, Makefile
├── figures/ (vector PDF), tables/ (auto-generated), data/
└── README.md
Called by
manual invocation, survey-to-deck (indirect, via research-refine chain)
Calls
- paper-intro-writing (P4)
- paper-related (P7)
- Reads: refine-logs/FINAL_PROPOSAL.md, refine-logs/EXPERIMENT_PLAN.md
