cg-paper-writing
Academic paper writing for 3D vision, computer graphics, CAD, and 3D understanding. Covers NeRF, 3DGS, SLAM, point cloud, 3D shape, CAD modeling. Supports CVPR/ICCV/ECCV/SIGGRAPH venues. Multi-agent adversarial review, citation integrity gates, style calibration. Use when: writing or revising a CG/3D vision paper, drafting abstract/intro/method/experiments, running adversarial review or citation integrity check, calibrating writing style to a venue, 写论文/写paper/论文写作/CG论文/三维视觉论文.
jaccen · 已有 AI 审核记录
安装与使用
运行命令前先阅读来源与依赖说明。复制命令不会自动安装或执行。
npx skills add jaccen/Awesome-Gaussian-Skills --skill cg-paper-writing技能说明
name: cg-paper-writing description: "Academic paper writing for 3D vision, computer graphics, CAD, and 3D understanding. Covers NeRF, 3DGS, SLAM, point cloud, 3D shape, CAD modeling. Supports CVPR/ICCV/ECCV/SIGGRAPH venues. Multi-agent adversarial review, citation integrity gates, style calibration. Use when: writing or revising a CG/3D vision paper, drafting abstract/intro/method/experiments, running adversarial review or citation integrity check, calibrating writing style to a venue, 写论文/写paper/论文写作/CG论文/三维视觉论文." license: Apache-2.0 user-invocable: true metadata: version: "3.0.0" author: jaccen tags: ["paper-writing", "academic", "computer-graphics", "3dgs", "nerf", "computer-vision", "cvpr", "siggraph", "adversarial-review", "citation-integrity", "style-calibration"] when_to_use: - "Write or revise a CG/3D vision academic paper" - "Draft abstract, introduction, related work, method, or experiments for a 3DGS/NeRF/CAD paper" - "Run adversarial review or citation integrity check on a draft" - "Calibrate writing style to a target venue" - "写论文 / 写paper / 论文写作 / CG论文 / 三维视觉论文"
CG Paper Writing Engine (Router)
Architecture: Axis-driven static/dynamic router. Do NOT try to apply the writing logic from memory or from this router alone. Always load fragments from disk as described below.
Step 1 — Detect Request Axes
Analyze the user's request to determine axis values:
Axis: section
| User Intent | section value |
|---|---|
| Writing or revising abstract | abstract |
| Writing or revising introduction | intro |
| Writing or revising related work | related-work |
| Writing or revising methodology | method |
| Writing or revising experiments | experiments |
| Writing contribution statements | contribution |
| Full paper or unspecified section | all |
Axis: venue
| User Intent | venue value |
|---|---|
| Targeting CVPR/ICCV/ECCV | cvpr |
| Targeting SIGGRAPH/EG/PG | siggraph |
| Targeting NeurIPS/AAAI | neurips |
| Targeting TVCG/CGF/TOG/TPAMI | tvcg |
| Writing PhD thesis chapter | thesis |
| Unspecified or multi-venue | all |
If the user does not specify, defaults are: section=all, venue=all.
Step 2 — Load Required Fragments
Always Load (every invocation)
Read these files from static/:
- static/core-stance.md — Role, writing process, de-AI rules, citation fact-check, guardrails
- static/symbols-terminology.md — Mathematical symbols, CG/CAD/3D terminology reference
On-Demand Load (by detected section)
| section | Fragment(s) to Load |
|---|---|
| abstract | static/writing-abstract-intro.md |
| intro | static/writing-abstract-intro.md |
| related-work | static/writing-related-method.md |
| method | static/writing-related-method.md |
| experiments | static/writing-experiments.md |
| contribution | static/writing-experiments.md |
| all | All writing fragments (abstract-intro, related-method, experiments) |
On-Demand Load (by detected venue)
| venue | Fragment to Load |
|---|---|
| any non-all value | static/venue-formats.md |
| all | static/venue-formats.md |
Reference Load (for review/integrity work)
- static/review-integrity.md — Multi-agent review, Devil's Advocate Protocol, citation verification, integrity gates, style calibration, writing quality check, persistence
Optimization: For a focused task (e.g., "write abstract"), load only core-stance + symbols-terminology + writing-abstract-intro + venue-formats (if venue specified). For full paper work, load all fragments.
Step 3 — Execute Writing Task
After loading the required fragments:
- Follow section-specific templates from the loaded writing fragment
- Apply venue-specific formatting from venue-formats.md
- Use terminology and symbols from symbols-terminology.md
- Observe all guardrails from core-stance.md (no fabrication, de-AI rules, citation fact-check)
- For review/integrity work, follow protocols from review-integrity.md
Rules
- Write in flowing prose, never bullet points (contribution statements and itemized lists excepted)
- Every claim needs evidence: Citation or experimental data, and citations must pass three-layer verification
- Use mathematical notation efficiently: One symbol, one meaning throughout; symbol table persisted
- Match the venue's tone: CVPR more concise; SIGGRAPH more narrative; if style sample provided, strictly calibrate
- Chinese academic writing: Follow Chinese academic conventions (本文/我们/由此/表明)
- Never fabricate data: Mark missing data as
<!-- DATA_NEEDED: <description> --> - Integrity gates cannot be skipped: Post-Draft Gate and Pre-Submission Gate must both pass
- Adversarial review must follow concession threshold protocol: Prevent sycophancy and frame-lock
- Claim-citation alignment: Each claim must be traceable to supporting citation, and citation must actually support the claim
- Writing context persistence: Maintain symbol, citation, review state consistency across sessions
Red Lines
The following are categorical prohibitions. Violating any of these invalidates the output:
- No invented data: Never fabricate experimental results, method capabilities, or review outcomes. If a value is not found in the loaded files, write "data not available" or "N/A".
- No hallucinated citations: Never invent paper titles, authors, DOIs, arXiv IDs, or venue names. Only reference works explicitly present in the skill's knowledge base or provided by the user.
- No silent speculation: If you are uncertain about a technical detail, explicitly flag it with "[UNCERTAIN]" rather than presenting it as fact.
- No method misattribution: Do not assign features, results, or mechanisms from one method to another. Each method's data is specific to that method.
- No oversimplified comparisons: Do not reduce multi-dimensional trade-offs to a single "better/worse" judgment wit