Prompt bundles that teach AI agents Nature journal standards

What it does — nature-skills is a repository of reusable instruction bundles—each centered on a SKILL.md file—that plug into Codex and Claude Code. The skills cover the full academic paper lifecycle, from generating multi-panel matplotlib figures and polishing prose to drafting reviewer responses, searching literature, and creating bilingual markdown readers. Each skill is a self-contained directory with references, static assets, and scripts that the agent consumes as context.

The interesting bit — This is essentially prompt engineering at scale: instead of one-off chats, the author has codified Nature journal conventions into structured agent workflows with manifest files and shared support directories. The nature-figure skill even ships with a chart-type atlas and production scripts from actual papers, so the agent has concrete visual standards rather than vague style requests.

Key highlights —

• 10 skills span figure generation, academic writing, peer-review simulation, citation management, data-availability statements, and paper-to-PPT conversion
• nature-figure produces multi-panel matplotlib plots with Nature typography, semantic colour palettes, and editable SVG output
• Packaged as a Codex marketplace plugin and thin Claude Code subagent wrappers that read from a local clone
• Only nature-figure and nature-polishing are marked Stable; the majority are Draft or Beta
• Shared _shared/, references/, and static/ directories let skills reuse assets and avoid broken relative links

Caveats —

• Most skills carry Draft or Beta status, so the breadth is wide but maturity is uneven across the catalog
• The project is essentially structured prompts and configuration files consumed by Codex or Claude Code, not a standalone application, so it is only useful if you already use those specific agents

Verdict — Researchers who want to automate Nature-format busywork through Codex or Claude Code will find the Stable skills genuinely useful; everyone else should treat this as a well-organized prompt library rather than a finished product.