The interface is localized. Skill descriptions and documents are machine-translated by AI into your language; code, commands, paths and technical terms stay in the original, and the author's document governs.
Explore skills
Data & research · SKILL.md

nature-data

nature-data

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

Independent GitHub snapshot · Automated classification · Original documentation governs

Skill documentation

GitHub

name: nature-data description: Prepare Data Availability, Code Availability, Model Availability, Prompt Availability, and reproducibility statements for RL, MARL, LLM, and LLM Agent manuscripts. version: 2.0.0-agent-rl

Data, code and artifact availability

Scope

Use for availability statements, artifact checklists, repository plans, supplementary material, reproducibility claims and reviewer requests about data/code/prompt release.

Required artifact inventory

  • RL/MARL: environment versions, wrappers, configs, seeds, logs, checkpoints, evaluation scripts, plotting scripts.
  • LLM: model identifiers, API version/date, prompts, decoding parameters, datasets, contamination controls, evaluation scripts.
  • LLM Agent: tool schemas, sandbox configuration, memory/retrieval corpus, trajectories, failure logs, human annotation or feedback protocol.
  • All papers: license, privacy constraints, compute resources, hardware, dependencies and archive DOI if possible.

Workflow

  1. Identify what can be released, what cannot, and why.
  2. Separate data, code, model checkpoints, prompts, trajectories and logs.
  3. Draft concise Nature-style availability statements.
  4. Add a reproducibility checklist for supplementary material.
  5. Flag claims that require artifacts not yet planned.

Output format

Return:

Artifact inventory:
- releasable:
- restricted:
- missing:

Data Availability statement:
...

Code Availability statement:
...

Model/Prompt/Trajectory Availability statement:
...

Supplementary reproducibility checklist:
- ...

Red lines

Do not promise public release if the user has not confirmed it. Do not expose API keys, private datasets, personal data, proprietary model weights or credentials.

Saved skills