PaperForge 🛠️📄

Topic in. Paper out.
One natural-language sentence in — a complete 8000+ word remote sensing manuscript out.
Literature mining → Google Earth Engine execution → figures, tables, formulas → DOCX, fully automated.

🚀 Quick Start · ✨ Features · 🏗️ Architecture · 📂 Output · 🎯 Examples · 🖥️ Platforms

📋 Table of Contents

What's New in v2.0
Overview
Features
Quick Start
Architecture
Installation
Workflow Detail
Output
Examples
CLI Reference
Platform Support
How PaperForge Compares
Roadmap
Requirements
FAQ
Community & Star
License

🆕 What's New in v2.0

Version 2.0 turns PaperForge from a prompt-only recipe into a runnable, testable, cross-platform Agent Skill:

v1 (before)	v2.0 (now)
Instructions only — the agent re-derives every code block	✅ Reusable Python modules: `gee_tasks.py`, `paper_writer.py`, `render_figures.py`, `literature.py`
No offline demo	✅ `python scripts/demo.py` → real figures + `.md` + `.html` + `.docx` in seconds, no GEE/internet
No tests	✅ `python scripts/test_pipeline.py` — 9 offline unit checks
Manual Codex setup	✅ `python scripts/package_codex.py --target ~/.codex/skills`
Non-standard frontmatter	✅ Spec-compliant `SKILL.md` (name/description/license/compatibility/metadata), <500 lines, progressive disclosure
No release channel	✅ GitHub Releases (`v2.0.0`) for version pinning

See the release page for the full changelog.

🔥 Overview

PaperForge is a production-grade Agent Skill that autonomously completes the entire remote sensing research workflow:

User says: "I want to study vegetation change in..."
                    ↓
    ┌────────────────────────────────────────────┐
    │  PaperForge Automatic Pipeline            │
    │                                            │
    │  Phase 1: ResearchX                       │
    │  ├── Ask: Chinese Core or SCI journal?    │
    │  ├── Multi-round literature search        │
    │  ├── Structured info extraction (15+ papers) │
    │  ├── Research gap analysis                │
    │  └── 3-5 GEE task design                  │
    │                                            │
    │  Phase 2: GEEPro                          │
    │  ├── Download GeoTIFFs (all years)        │
    │  ├── Compute statistics & accuracy        │
    │  └── Generate 300 DPI publication figures │
    │                                            │
    │  Phase 3: ResearchX                       │
    │  ├── 8000+ word manuscript                │
    │  ├── Introduction (1000+ words)           │
    │  ├── Methods with formulas & code         │
    │  ├── Embedded figures & tables            │
    │  ├── 25-30 formatted references           │
    │  └── Output: MD + HTML + DOCX             │
    └────────────────────────────────────────────┘
                    ↓
    Output/: {manuscript.md, .html, .docx, figures/, tables/, code/, tif/}

✨ Features

#	Feature	Description
1	Journal Type Selection	Asks Chinese Core or SCI before starting, adjusts length & citation style
2	Multi-Round Literature Search	3 rounds of targeted `web_search`, 15+ papers, 25-30 citations
3	Structured Extraction	Auto-extracts data source, algorithm, accuracy, innovation, limitation per paper
4	Gap Analysis	Cross-paper comparison → 3-5 research gaps → GEE-executable tasks
5	Automatic TIF Download	Downloads annual NDVI/FVC GeoTIFFs for all years to local disk
6	Publication Figures	6 auto-generated 300 DPI PNG figures (time series, maps, bar charts)
7	8000+ Word Manuscript	Complete SCI/Core manuscript with formulas, code snippets, tables
8	Multi-Format Export	`manuscript.md` + `manuscript.html` (browser-ready) + `manuscript.docx` (Word)
9	Offline Demo & Tests	`demo.py` + `test_pipeline.py` run without GEE or internet
10	Cross-Platform Skill	Claude Code, OpenAI Codex CLI, Cline, Cursor, Windsurf, Copilot
11	GEE Code Archive	All task scripts saved to `gee_code/`
12	GeoTIFF Archive	All NDVI rasters saved to `figures/ndvi_tif/`

🚀 Quick Start

30-second smoke demo (no GEE, no internet):

git clone https://github.com/xingguangYan/PaperForge.git
cd PaperForge
pip install -r requirements.txt          # or: pip install matplotlib numpy python-docx
python scripts/demo.py                   # → outputs/demo/manuscript.{md,html,docx}

Open outputs/demo/manuscript.html in your browser — that's the deliverable format produced by the full pipeline.

Full pipeline (with Google Earth Engine):

python scripts/run_pipeline.py \
    --topic "Vegetation change monitoring in Guishan and Sheshan, Wuhan" \
    --project your-project-id \
    --region "Wuhan" \
    --time "2020-2025"

Literature + paper only (no GEE):

python scripts/run_pipeline.py --topic "Urban heat island effect" --dry-run

🏗️ Architecture

graph TB
    subgraph User["User Input"]
        A["Research Topic<br>(e.g., Guishan vegetation change)"]
        B["Journal Type<br>(Chinese Core / SCI)"]
    end

    subgraph P1["Phase 1: ResearchX"]
        direction LR
        P1S1["Parse input<br>area/time/data"]
        P1S2["Literature search<br>3 rounds"]
        P1S3["Extract info<br>15+ papers"]
        P1S4["Gap analysis<br>3-5 tasks"]
    end

    subgraph P2["Phase 2: GEEPro"]
        direction LR
        P2S1["Load S2 data<br>2020-2025"]
        P2S2["Compute NDVI<br>annual median"]
        P2S3["Download TIF<br>all years"]
        P2S4["Generate figures<br>6 x 300 DPI PNG"]
    end

    subgraph P3["Phase 3: ResearchX"]
        direction LR
        P3S1["Integrate results<br>tables+figures"]
        P3S2["Write manuscript<br>8000+ words"]
        P3S3["Formulas+code<br>in methods"]
        P3S4["Export<br>MD+HTML+DOCX"]
    end

    subgraph Output["Output"]
        O1["manuscript.md"]
        O2["manuscript.html"]
        O3["manuscript.docx"]
        O4["figures/ + tables/"]
        O5["gee_code/ + ndvi_tif/"]
    end

    A --> P1
    B --> P1
    P1 --> P2
    P2 --> P3
    P3 --> Output

Technology Stack

Layer        Components
────────────────────────────────────────────────────
AI Engine    ResearchX (literature + paper writing)
             GEEPro (Earth Engine execution)

Code         scripts/gee_tasks.py      GEE building blocks
             scripts/literature.py     corpus validation & citations
             scripts/paper_writer.py   md → html → docx export
             scripts/render_figures.py 300-DPI matplotlib figures

Data         Google Earth Engine: Sentinel-2, Landsat, MODIS, ERA5
             Local: GeoTIFF, PNG, CSV, JSON

Platform     Claude Code · OpenAI Codex CLI · Cline · Cursor · Windsurf · Copilot

📦 Installation

System Requirements

Component	Minimum	Recommended
Python	3.8+	3.10+
RAM	4 GB	8 GB+
Disk	2 GB	10 GB+
Network	GEE access (Phase 2)	Stable broadband
OS	Windows/macOS/Linux	Any

Step 1 — Clone & install

git clone https://github.com/xingguangYan/PaperForge.git
cd PaperForge
pip install -r requirements.txt

requirements.txt (see Requirements for the versioned list):

earthengine-api    # Google Earth Engine Python SDK
geemap             # GEE interactive mapping
geopandas          # Geospatial data handling
pandas             # Data analysis
numpy              # Numerical computation
matplotlib         # Visualization (300 DPI figures)
seaborn            # Statistical charts
requests           # HTTP for TIF downloads
python-docx        # DOCX manuscript generation
rasterio           # GeoTIFF I/O and processing
scipy              # Mann-Kendall and statistical tests

Step 2 — Authenticate GEE

earthengine authenticate
# or service account:
# ee.Initialize(project="your-project", credentials="service_account.json")

Get a GEE Project ID at code.earthengine.google.com (Avatar → Project Settings).

Step 3 — Verify

python scripts/check_environment.py --project YOUR_PROJECT_ID

China Network Configuration

# Windows PowerShell
$env:HTTP_PROXY = "http://127.0.0.1:7890"
$env:HTTPS_PROXY = "http://127.0.0.1:7890"

# Linux / macOS
export HTTP_PROXY=http://127.0.0.1:7890
export HTTPS_PROXY=http://127.0.0.1:7890

🔬 Workflow Detail

Phase 1 — Literature Mining & Task Design

Parse input — extract core object, target variable, time range, spatial range, data preference; ask clarifying questions when anything is missing.
3-round search — broad → method-focused → gap mining.
Structured extraction — data source, algorithm, accuracy, innovation, limitation per paper (JSON schema in templates/task_template.json).
Gap analysis — cross-compare → 3-5 GEE-executable tasks.

Example extracted record:

{
  "title": "Monitoring vegetation dynamics using Sentinel-2 time series",
  "year": 2023,
  "journal": "Remote Sensing of Environment",
  "algorithm": "Random Forest",
  "accuracy": {"OA": 0.92, "Kappa": 0.89},
  "innovation": "Multi-temporal texture fusion",
  "limitation": "Single year only"
}

Phase 2 — GEE Execution & TIF Download

s2 = ee.ImageCollection("COPERNICUS/S2_SR_HARMONIZED")
s2_ndvi = s2.map(lambda img: img.addBands(
    img.normalizedDifference(["B8", "B4"]).rename("NDVI")))

For every year and site: compute the annual median, download the GeoTIFF to figures/ndvi_tif/, and compute statistics (mean NDVI, linear trend, Mann-Kendall, FVC via the dimidiate pixel model, vegetation grade, change detection).

Phase 3 — 8000+ Word Manuscript

Section	Word Count
Abstract	300-500
1. Introduction	1000-1500
2. Study Area & Data	800-1000
3. Methods	1500-2000
4. Results	2000-2500
5. Discussion	800-1000
6. Conclusion	300-400
References	25-30 entries

Key formulas (rendered in the manuscript):

FVC   = (NDVI - NDVI_soil) / (NDVI_veg - NDVI_soil)    # dimidiate pixel model
y     = β0 + β1 t + ε                                   # linear trend
S     = Σ_{i<j} sgn(x_j - x_i)                          # Mann-Kendall
OA    = (TP + TN) / (TP + TN + FP + FN)                 # overall accuracy
Kappa = (p_o - p_e) / (1 - p_e)                         # Cohen's kappa

Full step-by-step details: references/01_workflow.md.

📂 Output

outputs/YYYYMMDD_HHMMSS_topic_paper/
│
├── manuscript.md            # ★ Complete manuscript (8000+ words)
├── manuscript.html          # ★ Styled, browser-ready HTML
├── manuscript.docx          # ★ Word document (figures embedded)
├── manuscript_stats.json    # ★ Word-count & size report
│
├── figures/                 # 300 DPI PNG figures
│   ├── fig1_study_area.png
│   ├── fig2_ndvi_timeseries.png
│   ├── fig3_ndvi_comparison.png
│   ├── fig4_vegetation_grade.png
│   ├── fig5_change_detection.png
│   ├── fig6_multiyear_ndvi.png
│   └── ndvi_tif/            # ★ All GeoTIFFs, e.g. site1_ndvi_2020.tif
│
├── tables/                  # CSV + Markdown tables
├── gee_code/                # Complete GEE scripts
└── README.md                # Reproduction guide

See a live example by running python scripts/demo.py and opening outputs/demo/manuscript.html.

🎯 Examples

Example 1 — Guishan & Sheshan Vegetation Change (real GEE run)

python scripts/run_pipeline.py --topic "Vegetation change in Guishan and Sheshan, Wuhan" \
    --project ee-myproject --region "Wuhan" --time "2020-2025"

Metric	Guishan	Sheshan
NDVI 2020	0.144	0.214
NDVI 2025	0.115	0.153
FVC change	0.118→0.081 (-31.1%)	0.204→0.129 (-37.0%)
Trend slope	-0.0073/yr (p=0.056)	-0.0152/yr (p=0.056)
Degraded area	30.8%	24.6%

Deliverables: 8000+ word manuscript (md/html/docx), 6 figures, 12 GeoTIFFs, 4 tables, GEE scripts.

Example 2 — Yellow River Basin FVC (dry run)

python scripts/run_pipeline.py --topic "FVC change in Yellow River Basin" --dry-run

Example 3 — Taihu Lake Eutrophication (dry run)

python scripts/run_pipeline.py --topic "Taihu Lake eutrophication monitoring" --dry-run

💻 CLI Reference

python scripts/run_pipeline.py --topic "TEXT" [OPTIONS]

Option	Type	Required	Default	Description
`--topic`	str	✅	-	Research topic in natural language
`--project`	str	⚠️	-	GEE Project ID (needed for Phase 2)
`--region`	str	❌	auto	Study area description
`--time`	str	❌	last 5 yrs	e.g. `"2020-2025"`
`--tasks`	int	❌	4	Number of tasks (3-5)
`--format`	str	❌	gb	`gb` / `apa` / `mla`
`--journal`	str	❌	prompt	`chinese` / `sci`
`--no-deep`	flag	❌	false	Disable deep learning
`--dry-run`	flag	❌	false	Scheme only, no GEE
`--phase`	str	❌	all	`all` / `literature` / `gee` / `paper`

python scripts/run_pipeline.py --topic "..." --project xxx                       # full pipeline
python scripts/run_pipeline.py --topic "..." --dry-run                           # no GEE
python scripts/run_pipeline.py --topic "..." --project xxx --format apa --journal sci
python scripts/run_pipeline.py --topic "..." --phase literature
python scripts/demo.py                                                           # offline demo
python scripts/test_pipeline.py                                                  # offline tests
python scripts/package_codex.py --target ~/.codex/skills                         # Codex install

🖥️ Platform Support

Platform	Install
OpenAI Codex CLI	`python scripts/package_codex.py --target ~/.codex/skills`, then `$paperforge`
Claude Code	copy folder → `~/.claude/skills/PaperForge` or `.claude/skills/PaperForge`
Cline	add block to `.clinerules` (see `references/01_workflow.md` §Platform)
Cursor	add block to `.cursorrules`
Windsurf	add block to `.windsurfrules`
GitHub Copilot	add block to `.github/copilot-instructions.md`

⚖️ How PaperForge Compares

Capability	Manual prompting	Other paper-AI tools	PaperForge
Literature mining → gap analysis	⚠️ partial	✅	✅
GEE code generation	⚠️ manual	⚠️ partial	✅ automated + fallback
GeoTIFF download	❌	❌	✅ all years/sites
300-DPI publication figures	❌	⚠️	✅ 6 figures
DOCX + HTML + MD export	❌	⚠️	✅
Works with/without GEE (dry-run)	—	❌	✅
Offline demo + test suite	❌	❌	✅
Cross-platform Agent Skill	❌	⚠️	✅

🗺️ Roadmap

More indices — EVI, LAI, LST, chlorophyll-a, water clarity
LandTrendr / CCDC time-series breakpoint detection
Deep-learning baselines (U-Net, SegFormer) via --deep
Parallel multi-site GEE export with task batching
arXiv / Scopus / Web of Science live-query adapter for Phase 1
LaTeX + PDF export for journal submission
Web UI (Streamlit) front-end

Contributions welcome — open an issue or PR!

📋 Requirements

earthengine-api>=1.0.0      # Google Earth Engine
geemap>=0.30.0               # Interactive GEE mapping
geopandas>=0.14.0            # Geospatial operations
pandas>=2.0.0                # Data manipulation
numpy>=1.24.0                # Numerical computing
matplotlib>=3.7.0            # Figure generation (300 DPI)
seaborn>=0.12.0              # Statistical visualization
requests>=2.28.0             # HTTP downloads
python-docx>=1.0.0           # DOCX generation
rasterio>=1.3.0              # GeoTIFF I/O
scipy>=1.10.0                # Statistical tests

GEE datasets referenced: Sentinel-2 (COPERNICUS/S2_SR_HARMONIZED), Landsat 8/9 (LANDSAT/LC08/C02/T1_L2), MODIS (MODIS/061/MOD13Q1), ERA5 (ECMWF/ERA5_LAND/MONTHLY_AGGR), JRC surface water, Hansen forest change.

❓ FAQ

Q1: Can I use PaperForge without GEE? Yes — --dry-run completes literature + task design + paper scheme (results marked simulated). python scripts/demo.py needs no GEE and no internet.

Q2: Does it download GeoTIFFs? Yes — every annual NDVI/FVC raster goes to figures/ndvi_tif/.

Q3: How many words? Target 8000+, minimum 6000 (intro alone 1000+).

Q4: Output formats? Markdown + styled HTML + Word DOCX (figures embedded).

Q5: Reference formats? --format gb (GB/T 7714-2015, Chinese core) · --format apa · --format mla.

Q6: GEE auth fails? Run earthengine authenticate, or use a service account.

Q7: China network? Set HTTP_PROXY / HTTPS_PROXY (see Installation).

Q8: Memory limit? PaperForge auto-scales (bigger scale, smaller ROI).

Q9: How do I install it into Codex? python scripts/package_codex.py --target ~/.codex/skills, then $paperforge.

⭐ Community & Star

If PaperForge saves you time on literature review, GEE execution, or manuscript writing, please:

⭐ Star this repository — it signals demand and feeds the roadmap.
👀 Watch for releases (v2.0 is published on the Releases page).
🐛 Issues for bugs, 🌱 PRs for features — both welcome.

📄 License

PaperForge — Topic in. Paper out.
Professional Automated Remote Sensing Research Pipeline

Navigation Menu

Repository files navigation

PaperForge 🛠️📄

📋 Table of Contents

🆕 What's New in v2.0

🔥 Overview

✨ Features

🚀 Quick Start

🏗️ Architecture

Technology Stack

📦 Installation

System Requirements

Step 1 — Clone & install

Step 2 — Authenticate GEE

Step 3 — Verify

China Network Configuration

🔬 Workflow Detail

Phase 1 — Literature Mining & Task Design

Phase 2 — GEE Execution & TIF Download

Phase 3 — 8000+ Word Manuscript

📂 Output

🎯 Examples

Example 1 — Guishan & Sheshan Vegetation Change (real GEE run)

Example 2 — Yellow River Basin FVC (dry run)

Example 3 — Taihu Lake Eutrophication (dry run)

💻 CLI Reference

🖥️ Platform Support

⚖️ How PaperForge Compares

🗺️ Roadmap

📋 Requirements

❓ FAQ

⭐ Community & Star

📄 License

About

Releases

Packages

Contributors

Languages

Name		Name	Last commit message	Last commit date
assets		assets
references		references
scripts		scripts
templates		templates
.gitignore		.gitignore
LICENSE		LICENSE
README.md		README.md
SKILL.md		SKILL.md
requirements.txt		requirements.txt

Latest commit

History

Folders and files

Repository files navigation

PaperForge 🛠️📄

📋 Table of Contents

🆕 What's New in v2.0

🔥 Overview

✨ Features

🚀 Quick Start

🏗️ Architecture

Technology Stack

📦 Installation

System Requirements

Step 1 — Clone & install

Step 2 — Authenticate GEE

Step 3 — Verify

China Network Configuration

🔬 Workflow Detail

Phase 1 — Literature Mining & Task Design

Phase 2 — GEE Execution & TIF Download

Phase 3 — 8000+ Word Manuscript

📂 Output

🎯 Examples

Example 1 — Guishan & Sheshan Vegetation Change (real GEE run)

Example 2 — Yellow River Basin FVC (dry run)

Example 3 — Taihu Lake Eutrophication (dry run)

💻 CLI Reference

🖥️ Platform Support

⚖️ How PaperForge Compares

🗺️ Roadmap

📋 Requirements

❓ FAQ

⭐ Community & Star

📄 License

About

Resources

Stars

Watchers

Forks

Releases

Packages

Contributors

Languages