[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"project-83831":3},{"id":4,"name":5,"fullName":6,"owner":7,"repo":5,"description":8,"homepage":9,"htmlUrl":10,"language":11,"languages":10,"totalLinesOfCode":10,"stars":12,"forks":13,"watchers":14,"openIssues":15,"contributorsCount":15,"subscribersCount":15,"size":15,"stars1d":16,"stars7d":17,"stars30d":17,"stars90d":15,"forks30d":15,"starsTrendScore":18,"compositeScore":19,"rankGlobal":10,"rankLanguage":10,"license":20,"archived":21,"fork":21,"defaultBranch":22,"hasWiki":23,"hasPages":23,"topics":24,"createdAt":10,"pushedAt":10,"updatedAt":25,"readmeContent":26,"aiSummary":10,"trendingCount":15,"starSnapshotCount":15,"syncStatus":27,"lastSyncTime":28,"discoverSource":29},83831,"NaturePanelForge","littlepeachs\u002FNaturePanelForge","littlepeachs","NaturePanelForge is a code-first workflow for turning scientific figure images and open-access Nature-family papers into panel-level, executable plotting-code reconstruction tasks.","https:\u002F\u002Fuu543493-83c1-74a94416.nma1.seetacloud.com:8448\u002F",null,"Python",142,5,54,0,6,78,72,82.13,"MIT License",false,"main",true,[],"2026-06-12 04:01:42","\u003Ch1 align=\"center\">\n  \u003Cimg src=\"docs\u002Fassets\u002Ficon.png\" alt=\"NaturePanelForge icon\" width=\"180\">\u003Cbr\u002F>\n  NaturePanelForge\n\u003C\u002Fh1>\n\n\u003Cp align=\"center\">\n  \u003Cb>Forge Nature-level scientific panels into executable plotting code.\u003C\u002Fb>\u003Cbr\u002F>\n  Retrieve open-access papers, split full figures into reviewed panels, classify them with Qwen, and reproduce statistical panels with Codex agent loops.\n\u003C\u002Fp>\n\n\u003Cp align=\"center\">\n  \u003Ca href=\"https:\u002F\u002Flittlepeachs.github.io\u002FNaturePanelForge\u002F\">\u003Cimg alt=\"Project page\" src=\"https:\u002F\u002Fimg.shields.io\u002Fbadge\u002FProject%20Page-GitHub%20Pages-111827?style=flat&logo=githubpages&logoColor=white\">\u003C\u002Fa>\n  \u003Ca href=\"https:\u002F\u002Fuu543493-83c1-74a94416.nma1.seetacloud.com:8448\u002F\">\u003Cimg alt=\"Gallery live demo\" src=\"https:\u002F\u002Fimg.shields.io\u002Fbadge\u002FGallery-live%20demo-2563eb?style=flat&logo=googlechrome&logoColor=white\">\u003C\u002Fa>\n  \u003Ca href=\"#usage\">\u003Cimg alt=\"Usage four modes\" src=\"https:\u002F\u002Fimg.shields.io\u002Fbadge\u002FUsage-4%20modes-0f766e?style=flat&logo=python&logoColor=white\">\u003C\u002Fa>\n  \u003Ca href=\"#agent-workflow\">\u003Cimg alt=\"Agent loop refine\" src=\"https:\u002F\u002Fimg.shields.io\u002Fbadge\u002FAgent%20Loop-reproduce%20%2B%20refine-ea580c?style=flat&logo=openai&logoColor=white\">\u003C\u002Fa>\n  \u003Ca href=\"docs\u002Fintro_and_methods.md\">\u003Cimg alt=\"Method and stats\" src=\"https:\u002F\u002Fimg.shields.io\u002Fbadge\u002FMethod-stats%20%2B%20dataset-7c3aed?style=flat&logo=readme&logoColor=white\">\u003C\u002Fa>\n  \u003Ca href=\"#install-the-codex-skill\">\u003Cimg alt=\"Codex skill install\" src=\"https:\u002F\u002Fimg.shields.io\u002Fbadge\u002FCodex%20Skill-installable-0891b2?style=flat&logo=gnubash&logoColor=white\">\u003C\u002Fa>\n\u003C\u002Fp>\n\n\u003Cp align=\"center\">\n  \u003Cimg src=\"docs\u002Fassets\u002Fdiagram.png\" alt=\"NaturePanelForge input-output schematic: target scientific panel and optional metadata are converted into runnable Python plotting code.\" width=\"100%\">\n\u003C\u002Fp>\n\n\u003Cp align=\"center\">\n  \u003Ca href=\"#usage\">Usage\u003C\u002Fa> |\n  \u003Ca href=\"#quick-start-demo\">Quick Start Demo\u003C\u002Fa> |\n  \u003Ca href=\"#agent-workflow\">Agent Workflow\u003C\u002Fa> |\n  \u003Ca href=\"#codex-reproduction-examples\">Examples\u003C\u002Fa> |\n  \u003Ca href=\"#setup\">Setup\u003C\u002Fa> |\n  \u003Ca href=\"docs\u002Fintro_and_methods.md\">Method\u003C\u002Fa> |\n  \u003Ca href=\"TUTORIAL.md\">Tutorial\u003C\u002Fa>\n\u003C\u002Fp>\n\n\u003Cp align=\"center\">\n  \u003Cb>English\u003C\u002Fb> | \u003Ca href=\"README.zh-CN.md\">中文\u003C\u002Fa>\n\u003C\u002Fp>\n\nNaturePanelForge is a code-first workflow for turning scientific figure images and open-access Nature-family papers into panel-level, executable plotting-code reconstruction tasks. The repository does not depend on the hosted gallery demo.\n\n![NaturePanelForge gallery walkthrough](docs\u002Fassets\u002Fshow_preview.gif)\n\n[![Watch walkthrough video](docs\u002Fassets\u002Fwatch_walkthrough_button.svg)](docs\u002Fassets\u002Fshow_compressed.mp4)\n\n![NaturePanelForge gallery home](docs\u002Fassets\u002Fnature_panel_forge_web1.png)\n\n![NaturePanelForge gallery catalog](docs\u002Fassets\u002Fnature_panel_forge_web2.png)\n\nFor the project introduction, methods, current dataset counts, Qwen score distributions, and Codex refine complexity distribution, see [Introduction and Methods](docs\u002Fintro_and_methods.md).\n\n## Quick Start Demo\n\nThe quick start has two simple paths:\n\n### **Mode 1. Direct Code Mode**\n\n**Run the checked-in plotting code directly.** This is the fastest path: no Codex loop and no skill installation.\n\n\u003Ctable>\n\u003Ctr>\n\u003Ctd width=\"50%\" align=\"center\">\u003Cb>Direct Code Mode Target\u003C\u002Fb>\u003C\u002Ftd>\n\u003Ctd width=\"50%\" align=\"center\">\u003Cb>Direct Code Mode Reproduction\u003C\u002Fb>\u003C\u002Ftd>\n\u003C\u002Ftr>\n\u003Ctr>\n\u003Ctd>\u003Cimg src=\"docs\u002Fdemo\u002Fquick_start_bubble_plot\u002Ftarget.png\" alt=\"Direct code mode target GO enrichment bubble plot panel\" width=\"100%\"\u002F>\u003C\u002Ftd>\n\u003Ctd>\u003Cimg src=\"docs\u002Fdemo\u002Fquick_start_bubble_plot\u002Freproduce_panel.png\" alt=\"Direct code mode reproduced GO enrichment bubble plot panel\" width=\"100%\"\u002F>\u003C\u002Ftd>\n\u003C\u002Ftr>\n\u003C\u002Ftable>\n\nOne command:\n\n```bash\nbash scripts\u002Frun_quick_start_demo.sh\n```\n\nIt rerenders the checked-in script and images:\n\n```text\ndocs\u002Fdemo\u002Fquick_start_bubble_plot\u002Freproduce_panel.py\ndocs\u002Fdemo\u002Fquick_start_bubble_plot\u002Freproduce_panel.png\ndocs\u002Fdemo\u002Fquick_start_bubble_plot\u002Freproduce_panel.pdf\n```\n\n### **Mode 2. Install Codex Skill Mode**\n\n**Install the bundled Code Skill, then paste the full System Prompt inside Codex.** Codex uses `codex-panel-reproduce` to read the target image, write editable Python\u002Fmatplotlib code, render the reproduction image\u002FPDF, and save review files.\n\n\u003Ctable>\n\u003Ctr>\n\u003Ctd width=\"50%\" align=\"center\">\u003Cb>Codex Skill Mode Target\u003C\u002Fb>\u003C\u002Ftd>\n\u003Ctd width=\"50%\" align=\"center\">\u003Cb>Codex Skill Mode Reproduction\u003C\u002Fb>\u003C\u002Ftd>\n\u003C\u002Ftr>\n\u003Ctr>\n\u003Ctd>\u003Cimg src=\"docs\u002Fdemo\u002Fquick_start_skill_bubble_plot\u002Ftarget.png\" alt=\"Codex Skill mode target GO enrichment bubble plot panel\" width=\"100%\"\u002F>\u003C\u002Ftd>\n\u003Ctd>\u003Cimg src=\"docs\u002Fdemo\u002Fquick_start_skill_bubble_plot\u002Freproduce_panel.png\" alt=\"Codex Skill mode reproduced GO enrichment bubble plot panel\" width=\"100%\"\u002F>\u003C\u002Ftd>\n\u003C\u002Ftr>\n\u003C\u002Ftable>\n\n**Install Skill Prompt:**\n\n```text\nInstall the NaturePanelForge Codex skill for me: https:\u002F\u002Fgithub.com\u002Flittlepeachs\u002FNaturePanelForge\n```\n\nCodex will read the repository, install the bundled `codex-panel-reproduce` skill, and verify that it is available locally.\n\n**English System Prompt:**\n\n```text\nPlease use the installed codex-panel-reproduce skill to reproduce this scientific paper panel as editable Python\u002Fmatplotlib code.\n\nTarget image: docs\u002Fdemo\u002Fquick_start_skill_bubble_plot\u002Ftarget.png\nOptional PDF: docs\u002Fdemo\u002Fquick_start_skill_bubble_plot\u002Ftarget.pdf\nOutput root: UserRuns\u002Fmy_skill_test\nPanel id: my_skill_test\nChart type: bubble_plot\nCaption: A faceted GO enrichment bubble plot with Human and Mouse columns, biological process labels on the left, x axis as -log10(p.value), bubble size encoding log10(count), and colors encoding biological groups.\n\nRequirements:\n1. Use the codex-panel-reproduce skill.\n2. Do not use Qwen scoring; this is a local user-supplied image.\n3. Do not manually modify images; generate executable plotting code only.\n4. Run the NaturePanelForge single-panel reproduction workflow.\n5. Generate reproduce_panel.py, reproduce_panel.png, and reproduce_panel.pdf.\n6. Generate review notes, review summary, and run log.\n7. Finally report the output directory, review_passed, contract_passed, final PNG size, and rerender command.\n```\n\n**中文 System Prompt：**\n\n```text\n请使用已安装的 codex-panel-reproduce skill，帮我复现这个科学论文 panel 图像为可编辑的 Python\u002Fmatplotlib 代码。\n\n目标图像：docs\u002Fdemo\u002Fquick_start_skill_bubble_plot\u002Ftarget.png\n可选 PDF：docs\u002Fdemo\u002Fquick_start_skill_bubble_plot\u002Ftarget.pdf\n输出根目录：UserRuns\u002Fmy_skill_test\npanel id：my_skill_test\n图类型：bubble_plot\ncaption：A faceted GO enrichment bubble plot with Human and Mouse columns, biological process labels on the left, x axis as -log10(p.value), bubble size encoding log10(count), and colors encoding biological groups.\n\n要求：\n1. 使用 codex-panel-reproduce skill。\n2. 不要使用 Qwen scoring，这是本地用户提供的图片。\n3. 不要手工修改图片，只生成可执行绘图代码。\n4. 运行 NaturePanelForge 的单图复现 workflow。\n5. 生成 reproduce_panel.py、reproduce_panel.png、reproduce_panel.pdf。\n6. 生成 review notes、review summary、run log。\n7. 最后告诉我输出目录、review_passed、contract_passed、最终 PNG 尺寸和重新渲染命令。\n```\n\nThe target image is the input, and the reproduction image is the expected editable-code output. For the full prompts saved as files, see [English prompt](docs\u002Fdemo\u002Fquick_start_skill_bubble_plot\u002Fprompt.en.md) and [中文 prompt](docs\u002Fdemo\u002Fquick_start_skill_bubble_plot\u002Fprompt.zh-CN.md).\n\nOffline rerender for the checked-in Skill Mode result:\n\n```bash\nDEMO_CASE=quick_start_skill_bubble_plot bash scripts\u002Frun_quick_start_demo.sh\n```\n\nTo use your own data in the same visual style, edit the data arrays and labels in one of the demo `reproduce_panel.py` scripts, then rerun it. If your goal is specifically “take a reference plot style and redraw my new data in that style,” FigMirror is the more product-like path; NaturePanelForge focuses on building scientific panel-to-code benchmark examples from real papers.\n\n## Agent Workflow\n\nNaturePanelForge uses three executable agent stages. Each stage writes machine-checkable artifacts and can be resumed.\n\n![NaturePanelForge complete workflow](docs\u002Fassets\u002Fnature_panel_forge_overview.png)\n\n\n**Panel Split** converts a compound full figure into complete panel crops. A split agent writes executable crop\u002Fspec logic, and a review agent checks panel letters, axis labels, ticks, legends, colorbars, titles, annotations, and edge visibility.\n\n**Code Reproduce** turns a target panel into `reproduce_panel.py`, `reproduce_panel.png`, and `reproduce_panel.pdf`. A code-writing agent renders the plot, and a review agent compares the output against `target.png`; fixable issues trigger code edits and rerendering.\n\n**Final Refine** starts from first-pass reproductions that already passed review. A polish agent edits the existing code, while an audit agent checks Arial typography, label\u002Ftick\u002Flegend overlap, scientific symbols, edge clipping, compactness, complexity score, and caption-based description.\n\n![Agent loops for high-fidelity panel-to-code reproduction](docs\u002Fassets\u002Fagent_loop.png)\n\nThis loop is the core mechanism for high-quality panel-to-code reproduction: each stage separates execution from review, records artifacts on disk, and iterates until the panel split, executable reproduction, or final refine result passes the corresponding audit.\n\n## Codex Reproduction Examples\n\nTarget panels from real paper figures are shown beside Codex-rendered outputs. Each reproduction is generated from executable plotting code, not manual image editing.\n\n![Target panels beside Codex reproductions](docs\u002Fassets\u002Freproduction_examples\u002Fcodex_reproduction_pairs.png)\n\n## Usage\n\nUse `forge.py` for the four public workflows. Each mode can be started with one command.\n\n1. **Single cropped panel image -> plotting code**\n\n```bash\npython3 forge.py single-panel-image --image \u002Fpath\u002Fto\u002Ftarget_panel.png --panel-id demo_panel --chart-type bar --caption \"A grouped bar chart with error bars and a legend.\" --out-root UserRuns\u002Fpanel_demo --model gpt-5.4 --reasoning-effort medium --review-rounds 4 --skip-existing\n```\n\n2. **Single full figure image -> reviewed panel crops**\n\n```bash\npython3 forge.py single-full-image --image \u002Fpath\u002Fto\u002Ffull_figure.png --paper-id demo_paper --caption \"A complete multi-panel scientific figure.\" --out-root UserRuns\u002Ffull_demo --model gpt-5.4 --reasoning-effort medium --review-rounds 4 --skip-existing\n```\n\n3. **Single paper -> paper metadata and full figures**\n\n```bash\npython3 forge.py single-paper --doi 10.1038\u002Fs41467-025-12345-6 --subject biology --topic AI_biology --figures-per-paper 5 --download-only\n```\n\n4. **Batched papers -> full paper-to-panel-to-code workflow**\n\n```bash\npython3 forge.py batched-paper --subject materials --topic AI_materials --target-papers 20 --batch-size 20 --figures-per-paper 5 --years 2024,2025,2026 --codex-model gpt-5.4 --codex-jobs 8\n```\n\nThe repository root intentionally keeps only one Python entry point, `forge.py`. Internal pipeline modules live under `nature_panel_forge\u002F`, while stage wrappers live under `scripts\u002F`.\n\n`single-panel-image` is the direct user-facing image-to-code path. A live run returns:\n\n```text\ntarget.png\nmetadata.json\nqwen_score.json\nreproduce_panel.py\nreproduce_panel.png\nreproduce_panel.pdf\nuser_reproduce_summary.json\nresult.json\n```\n\n`user_reproduce_summary.json` and `result.json` include the generated code text, output paths, review status, missing-output checks, and whether the live contract passed. Dry-runs intentionally set `live_contract_checked=false` and `contract_passed=false`.\n\n## Setup\n\n```bash\ncd NaturePanelForge\npython3 -m venv .venv\nsource .venv\u002Fbin\u002Factivate\npip install -r requirements.txt\ncp configs\u002Fdemo.env.example .env\nsource .env\n```\n\nRequired external pieces:\n\n- network access for paper and figure download\n- Codex CLI for panel splitting, reproduction, and refine\n- local Qwen vision model for panel classification\u002Fscoring\n\nCheck Codex:\n\n```bash\ncodex --version\n```\n\nSet Qwen:\n\n```bash\nexport QWEN_MODEL_PATH=\u002Fpath\u002Fto\u002FQwen3.6-27B\nexport CUDA_VISIBLE_DEVICES=0\n```\n\nThe default Qwen path is local `transformers` loading through `--qwen-backend transformers`. `--qwen-backend openai` is only an optional compatibility hook for users who intentionally run an OpenAI-compatible vision endpoint.\n\n## Install The Codex Skill\n\nThe easiest way is to let Codex install it from this GitHub repository. Already inside Codex? Paste this prompt:\n\n**Install Skill Prompt:**\n\n```text\nInstall the NaturePanelForge Codex skill for me: https:\u002F\u002Fgithub.com\u002Flittlepeachs\u002FNaturePanelForge\n```\n\nCodex should clone or open the repository, install `skills\u002Fcodex-panel-reproduce\u002FSKILL.md` into the local Codex skills directory, and verify that this file exists:\n\n```text\n${CODEX_HOME:-$HOME\u002F.codex}\u002Fskills\u002Fcodex-panel-reproduce\u002FSKILL.md\n```\n\nAfter installation, use the full English or Chinese System Prompt shown in Quick Start Mode 2.\n\nLocal Codex can then follow the single-panel reproduction\u002Frefine workflow without re-learning the prompt structure from scratch.\n\n## Repository Layout\n\n```text\nforge.py                            # single public Python CLI entry point\nnature_panel_forge\u002F                 # internal paper, figure, export, refine, and image-to-code modules\nagent_loop\u002F                         # Codex panel splitting, manifest building, Qwen scoring\nexamples\u002F                           # Codex reproduction and final-refine batch drivers\nscripts\u002F                            # portable stage wrappers and skill installer\ngallery\u002F                            # static gallery shell and catalog builder\nskills\u002F                             # local Codex skill packages\nconfigs\u002F                            # environment templates\ndocs\u002F                               # architecture, methods, and visual assets\nprompts\u002F                            # agent prompt blueprints\n```\n\nGenerated run directories look like:\n\n```text\nPipelineRuns\u002F\u003Csubject>\u002F\u003Ctopic>\u002Frun_YYYYMMDD_HHMMSS_batch001\u002F\n  Papers\u002F\n  FullFigures\u002Ffull_figures.csv\n  Panels_codex_full\u002F\n  PanelReviews_codex_full\u002F\n  QwenPanelScore\u002F\n  Final_Schematic\u002F\n  Reproduce_Statistical\u002F\n  Reproduce_Statistical_Reviews\u002F\n  Reproduce_Statistical_Refined\u002F\n```\n\n## More Documentation\n\n- [Tutorial](TUTORIAL.md)\n- [Architecture](docs\u002Farchitecture.md)\n- [Introduction and Methods](docs\u002Fintro_and_methods.md)\n\n## Citation\n\nIf you use this workflow in a paper or benchmark, cite the project repository and describe the exact paper sources, model versions, scoring thresholds, and Codex review-round settings used in your run.\n",2,"2026-06-11 04:11:36","CREATED_QUERY"]