How to use the Nature Data Availability Helper agent skill

What is an agent skill, and what does nature-data make the agent do?

An agent skill is a folder of instructions and reference files that an AI agent loads when a task matches the skill's description. The agent reads the main SKILL.md file first and opens the referenced files only when it needs them.

The nature-data skill covers publication data policy, not data analysis. Once installed, the agent drafts or audits Data Availability and Code Availability statements, plans dataset access routes and repositories, and checks FAIR metadata. It is explicitly told not to invent DOIs, accession numbers, licences or ethics approvals, and to treat vague 'available upon request' wording as weak.

• Draft a manuscript-ready Data Availability statement.
• Map each dataset that supports a figure or conclusion to a repository, accession, DOI, licence or access condition.
• Classify each dataset into an access route: public repository, controlled access, within paper, reused public, third-party restricted, justified request, or not applicable.
• Run a FAIR metadata check and list open items that need author input.
• Return paste-ready text with unresolved fields flagged.

Install the nature-data skill

Download the ZIP for skills/nature-data from the Nature agent-skills page. Then use one of the two install routes below.

For Claude Code, unzip the folder into ~/.claude/skills/ to make it available to all your projects, or into a project's .claude/skills/ folder to limit it to one project. For the Claude apps, upload the ZIP under Settings → Capabilities → Skills.

• Personal install: unzip into ~/.claude/skills/
• Project install: unzip into the project's .claude/skills/ folder
• Claude apps: Settings → Capabilities → Skills, then upload the ZIP

What the agent needs to run the skill

The skill needs an agent that can read SKILL.md and the markdown files it references. The source lists no external software, no API keys and no accounts. It does not run scripts and does not use the network.

The agent works only with the facts you give it. Before you start, gather the supporting data sources for each figure or conclusion, the target journal, and any repository identifiers, embargo dates or access restrictions you already have.

How to trigger the skill with natural requests

You do not need special syntax. Describe the manuscript task in plain language and mention the journal and the data situation. The skill's description includes Chinese terms as well, so requests in Chinese such as 数据可用性声明 or 数据仓库选择 also match.

For a wording edit or audit of one existing statement, the agent keeps the repository identifiers and access conditions you supply and checks only the affected claims. For a new data-sharing plan, a full statement or a submission audit, it runs the complete eight-step workflow.

• “帮我把这篇稿子的 Data Availability 写成 Nature 风格。”
• “检查我的数据声明是否缺 accession、仓储或许可证。”
• “这些数据有一部分不能公开，帮我写受控访问说明。”
• “Check whether 'data are available from the corresponding author' is sufficient for my Nature Article.”

Example: working through the eight-step workflow

For a new statement, the agent confirms the target journal, lists each supporting dataset, classifies its access route, chooses a repository and identifiers, drafts the statement, adds dataset citations, runs a FAIR metadata check, and returns paste-ready text. If the target is Nature Machine Intelligence, the agent also checks that a separate Code availability section follows the Data Availability statement.

If the journal's instructions conflict with the skill, the journal takes precedence.

• Step 1: identify the journal and article type.
• Step 2: inventory every supporting dataset.
• Step 3: classify each dataset into one access route.
• Step 4: choose the repository and identifier strategy before drafting.
• Step 5: draft the statement with explicit dataset-to-location mapping.
• Step 6: add formal dataset citations.
• Step 7: run the FAIR/metadata audit.
• Step 8: return ready-to-paste text and unresolved fields.

Safety notes: scripts, network and restricted data

The skill runs no scripts and uses no network access, so it only reads its own files and the text you provide. Because it cannot verify access committees or ethics approvals, it depends on you for facts about restrictions. If you supply no facts, its wording can be misleading.

Data with legal, ethical, privacy, commercial or third-party limits needs real approvals and real conditions from you. The skill will not supply them.

• Do not ask the agent to confirm a DOI or accession it has not seen; check the repository yourself.
• Give the actual access conditions for restricted data rather than a generic phrase.
• Treat the output as a drafting aid and check final statements with the journal or editor.

Tips for better results

Provide facts before asking for wording. The more specific your input (repository name, accession, embargo end, licence), the less the agent has to flag as open.

For Chinese author notes, ask in Chinese or say you want Chinese guidance. The skill then adds a 中文核对 check block unless you ask for statement text only.

• List each figure or conclusion with its data source.
• Say which journal and article type you are submitting to.
• Ask for a short confirmation list if information is missing, rather than a full statement.