`nature-figure` 技能SAFE

From the repository's own README, as read at the audited commit. Badges and raw HTML are left out.

The instruction file, verbatim from the audited commit — this is the text the model reads, and the surface the audit's instruction layer examines. Quoted here so you can judge it without cloning anything.

---
name: nature-figure
description: >-
  Create, revise, audit, and export manuscript scientific figures in Python or R.
  Use for 论文配图、科研绘图、多面板图 and submission-ready plots, or explicitly
  requested AI-generated graphical abstracts and mechanism schematics. Not for
  interactive dashboards, data cleaning, or statistics-only analysis.
---

# Nature Figure Making — Router

## Routing protocol

For a new task, load the core and matching resources below. Reuse already loaded guidance on follow-ups; load more only when the task needs it.

### 0. Check for graphical-abstract and AI-schematic routes

For every graphical-abstract planning, generation, revision, or audit task that
uses AI, read
[references/ai-graphical-abstract-workflow.md](references/ai-graphical-abstract-workflow.md)
first. It owns the message/audience brief, composition and palette workflow,
policy gate, human scientific review, disclosure boundary, and provenance
requirements. A Nature Careers article is practitioner advice, not submission
clearance; verify the current official policy for the exact target journal.

If the request is planning or auditing only, do not ask for Python or R unless
the user also asks to render or revise a data-driven figure.

If the user explicitly asks to generate a manuscript schematic, graphical abstract, mechanism diagram, concept illustration, or paper schematic with OpenRouter, GPT Image 2, an image-generation API, or similar wording, do **not** ask "Python or R?". This is a non-plotting AI-schematic route.

For this route:

1. Read [manifest.yaml](manifest.yaml) and the `always_load` files.
2. Read [references/ai-graphical-abstract-workflow.md](references/ai-graphical-abstract-workflow.md).
3. Read [references/openrouter-image-generation.md](references/openrouter-image-generation.md).
4. Use [scripts/generate_openrouter_schematic.py](scripts/generate_openrouter_schematic.py) when the user wants a real API call or a reproducible payload.
5. Treat output as a draft schematic / graphical abstract, not as a quantitative data panel. Do not invent experimental values, author logos, institutional marks, or unsupported mechanisms. Keep internal usefulness separate from submission eligibility.

If the user wants a hand-built, editable vector schematic (architecture, pipeline, agent or model block diagram, or mechanism flow) rather than an AI-generated image, do **not** ask "Python or R?". Read [references/vector-schematic-workflow.md](references/vector-schematic-workflow.md) and [references/vector-schematic-style-rules.md](references/vector-schematic-style-rules.md), confirm the brief with the user before drawing, and run `scripts/audit_svg_schematic.py` on the SVG before delivery. This route never plots data.

Only continue to the Python/R backend gate for plotting, charting, data visualization, or manuscript figure assembly tasks that are not explicit OpenRouter AI image-generation requests.

### 1. Load the manifest and the core layer

Read [manifest.yaml](manifest.yaml). It declares the `backend` axis, the allowed values, and the file paths each value maps to.

Also read every file listed under `always_load` (`static/core/contract.md` and `static/core/stance.md`). These hold the figure contract, the backend gate, the missing-runtime rule, the privacy rule, and the default operating stance that apply to every figure job.

### 2. Resolve the plotting backend

Backend selection applies only to rendering or editing plotting code. Reuse a choice already established in the same task and its follow-ups; do not ask again merely because a new message omits the language. Read-only figure review and backend-independent data inspection may proceed without this choice. If the backend remains unresolved, retain the one-time Python/R question and pause only dependent plotting steps. Explicit approval requirements and backend exclusivity remain in force.

Resolve the plotting backend from the current task before consulting the saved default. Decide the `backend` value in this order:

1. If the current request explicitly chooses Python or R, use that backend and save it with `scripts/nature_figure_backend.py set python` or `scripts/nature_figure_backend.py set r`.
2. If the request provides a clearly language-specific input file/workflow, use that backend and save it.
3. Otherwise reuse a Python/R choice already established in this task. If none exists, run `scripts/nature_figure_backend.py get` and use a returned `python` or `r` preference.
4. If neither a task choice nor a saved preference exists, ask exactly one concise question — **Python or R? I will remember this as your default.** — and pause only dependent plotting steps. After the user answers, save the answer before proceeding.

- `python` — matplotlib / seaborn.
- `r` — ggplot2 / patchwork / ComplexHeatmap.

Do not guess or choose a backend by aesthetics alone. Only recommend a backend when the user explicitly asks you to choose; then use `references/backend-selection.md`, state the reason, save the selected backend, and proceed. Once selected, the backend is **exclusive** for all drawing, previewing, exporting, and visual QA (see `core/contract.md`). This gate does not apply to the explicit OpenRouter AI-schematic route above.

### 3. Load the matching backend fragment

After the backend is resolved, Read the mapped fragment (`static/fragments/backend/python.md` or `static/fragments/backend/r.md`). It carries the backend-only execution rule and the publication quick-start (rcParams/theme and export helper). Do **not** load the other backend's fragment.

### 4. Build the figure using the loaded material

Apply the loaded material in this order:

1. Figure contract (`core/contract.md`) — write the core conclusion, map the evidence chain, classify the archetype, set the journal/export contract, before any code.
2. Multi-panel evidence architecture — when planning, restructuring, or auditing a labelled multi-panel figure, load `references/multipanel-evi

SAFEgrade B · trust 89/100 Nothing in the source contradicts what it says it does. Grade A is reserved for packages that have also passed the behavioural sandbox.

What the source does

Filesystem

none-observed

Network

declared (4 observation(s))

Shell

none-observed

Dependencies

not all pinned

Secrets in source

none-found

Findings (8)

response = {"data": [{"media_type": "image/png", "url": "http://127.0.0.1/x.png"}]}

return base64.b64decode(value)

PyMuPDF

assets/figures4papers/assets/ImmunoStruct_results_CEDAR.png

assets/figures4papers/assets/RNAGenScape_schematic.png

assets/figures4papers/figure_Cflows/figures/diffusion_swiss_roll.png

assets/figures4papers/figure_ImmunoStruct/figures/bars_ablation_IEDB.png

assets/figures4papers/figure_RNAGenScape/figures/manifold_holes.png

Gates applied: no_behavioural_pass.

[![ModelsAtlas audit: `nature-figure` 技能](https://modelsatlas.com/skills/yuan1z0825/nature-figure/badge.svg)](https://modelsatlas.com/skills/yuan1z0825/nature-figure/)

Not your listing, or something out of date? Claim or update your listing.