一键在 Manus 中运行任何 Skill

nature-figure

星标4

分支0

更新时间2026年7月1日 03:14

Create, revise, or audit Nature-style scientific figures for RL, MARL, LLM, and LLM Agent papers. Use for Python/R plotting, training curves, ablations, benchmark bars, architecture diagrams, agent workflow figures, trajectory visualizations, and publication export QA.

安装

用 Codex 或 Claude 帮你安装复制这段 Prompt，粘贴到 Codex、Claude 或其他助手里，让它检查 Skill 页面并帮你完成安装。

在 Manus 中运行

来源

vcis-wangchenxu

vcis-wangchenxu/agent_nature-skills

打开 GitHub 仓库查看创作者相关仓库

下载

在 Manus 中运行

AI/RL publication figure workflow

Blocking gate

If the user asks to generate executable plotting code and has not chosen a backend, ask: Python or R? If the user provides Python code or matplotlib context, proceed with Python. For image/design-only requests, focus on figure logic and layout.

Figure contract

Before drawing, define:

core conclusion of the figure;
evidence units: seeds, tasks, agents, models, datasets, environments;
comparison target: baselines, ablations, scaling, generalization, cost, failure mode;
uncertainty display: CI, standard error, bootstrapping, per-seed scatter;
export target: PDF/SVG for vector, PNG/TIFF for raster.

Field-specific figure archetypes

RL: learning curve with shaded uncertainty, final-return bar with per-seed dots, sample-efficiency curve, hyperparameter sensitivity.
MARL: partner/opponent generalization matrix, zero-shot coordination heatmap, population diversity map, role/policy trajectory visualization.
LLM: benchmark grouped bars, scaling/cost Pareto plot, error taxonomy, ablation matrix, prompt sensitivity.
LLM Agent: workflow graph, tool-call timeline, success/failure trajectory, memory/retrieval diagram, latency-cost-quality Pareto plot.

Quality rules

Do not plot mock data unless the user explicitly asks for synthetic illustration.
Use consistent fonts, readable labels, direct legends, restrained color and high contrast.
Every multi-panel figure must have a single narrative order.
State what raw data columns are required if data is missing.
For RL curves, avoid hiding instability: show seeds or uncertainty.
For benchmark tables, include rank, mean, uncertainty and compute/cost where relevant.

Output format

Return: figure plan → required data → code or layout → export instructions → QA checklist.

同仓库更多 Skills

同仓库

nature-academic-search

vcis-wangchenxu/agent_nature-skills

Search, plan, and organize literature for RL, MARL, LLM, and LLM Agent research. Use for finding papers, building related-work maps, identifying baselines, benchmark provenance, citation gaps, and recent work tracking.

2026-07-014

nature-citation

vcis-wangchenxu/agent_nature-skills

Build, audit, and organize citation evidence for RL, MARL, LLM, and LLM Agent manuscripts. Use for related work matrices, citation gaps, DOI/BibTeX checks, baseline provenance, benchmark provenance, and Nature-style support for claims.

2026-07-014

nature-data

vcis-wangchenxu/agent_nature-skills

Prepare Data Availability, Code Availability, Model Availability, Prompt Availability, and reproducibility statements for RL, MARL, LLM, and LLM Agent manuscripts.

2026-07-014

nature-downloader

vcis-wangchenxu/agent_nature-skills

Help legally obtain, organize, and archive academic papers, PDFs, BibTeX, supplementary files, code links, and benchmark documents for RL, MARL, LLM, and LLM Agent research. Do not bypass paywalls or access controls.

2026-07-014

nature-literature-pipeline

vcis-wangchenxu/agent_nature-skills

Build a repeatable literature discovery, scoring, reading, and archiving pipeline for RL, MARL, LLM, and LLM Agent research. Use for daily/weekly paper tracking, topic watchlists, benchmark monitoring, and research-note automation.

2026-07-014

nature-paper-to-patent

vcis-wangchenxu/agent_nature-skills

Convert RL, MARL, LLM, or LLM Agent research into a preliminary technical invention disclosure or Chinese patent-style draft. Use for invention points, claims, embodiments, flowcharts, and technical effects. Not legal advice.

2026-07-014

name	nature-figure
description	Create, revise, or audit Nature-style scientific figures for RL, MARL, LLM, and LLM Agent papers. Use for Python/R plotting, training curves, ablations, benchmark bars, architecture diagrams, agent workflow figures, trajectory visualizations, and publication export QA.
version	2.0.0-agent-rl

AI/RL publication figure workflow

Blocking gate

Figure contract

Before drawing, define:

core conclusion of the figure;
evidence units: seeds, tasks, agents, models, datasets, environments;
comparison target: baselines, ablations, scaling, generalization, cost, failure mode;
uncertainty display: CI, standard error, bootstrapping, per-seed scatter;
export target: PDF/SVG for vector, PNG/TIFF for raster.

Field-specific figure archetypes

RL: learning curve with shaded uncertainty, final-return bar with per-seed dots, sample-efficiency curve, hyperparameter sensitivity.
MARL: partner/opponent generalization matrix, zero-shot coordination heatmap, population diversity map, role/policy trajectory visualization.
LLM: benchmark grouped bars, scaling/cost Pareto plot, error taxonomy, ablation matrix, prompt sensitivity.
LLM Agent: workflow graph, tool-call timeline, success/failure trajectory, memory/retrieval diagram, latency-cost-quality Pareto plot.

Quality rules

Do not plot mock data unless the user explicitly asks for synthetic illustration.
Use consistent fonts, readable labels, direct legends, restrained color and high contrast.
Every multi-panel figure must have a single narrative order.
State what raw data columns are required if data is missing.
For RL curves, avoid hiding instability: show seeds or uncertainty.
For benchmark tables, include rank, mean, uncertainty and compute/cost where relevant.

Output format

Return: figure plan → required data → code or layout → export instructions → QA checklist.