WIT

WIT (Writing Is Thinking) is a human–LLM collaborative skill for scientific thinking and writing.

Advance the research. Grow the researcher.

WIT is designed not only to improve scientific work, but also to strengthen the researcher's ability to ask questions, interpret evidence, design experiments, and make scientific judgments.

Automate labor; augment judgment.

WIT uses the LLM to extend, challenge, scaffold, and audit the researcher's thinking—not simply to replace it. Core principle:

WIT specifies scientific reasoning functions, not a rigid surface template for papers.

It helps researchers:

  • open vague ideas into a testable scientific question space;
  • organize Results around evidence and restrained interpretation;
  • distinguish local Results-level interpretation from integrated Discussion-level interpretation;
  • move from findings toward broader meaning or general principles;
  • generate competing hypotheses and choose high-information next experiments;
  • decide which new questions deserve further experiments and which belong to boundaries, limitations, or future studies;
  • audit the logical chain and claim–evidence alignment of a paper;
  • keep the researcher actively involved in the scientific judgments that matter most.

Use

If your agent supports Agent Skills / SKILL.md, install or expose the wit/ directory as the skill package and let the agent load wit/SKILL.md.

Otherwise, ask the AI to read:

wit/references/WIT-Scientific-thinking-and-writing-skill.md

Then invoke it directly:

Use WIT to open up this question: XXX.

This is a finding: XXX. Use WIT to expand it.

Use WIT to review the Results and Discussion.

Use WIT to audit the logical chain of this paper.

Use WIT in human–LLM collaboration mode: help me think, not just finish the task for me.

Paper Logic Audit

  • Introduction: does it explain why the study is necessary?
  • Results titles: do they form a coherent progression? Question-driven and pipeline-driven forms can both work.
  • Results subsection: is the underlying Motivation / Question → Fact → restrained 1-hop Opinion → Next Step chain recoverable?
  • Discussion core: Integrated Interpretation → Broader Meaning / justified abstraction
  • Discussion optional: New Questions → Boundary / Limitations → Future Studies, when scientifically useful

Audit the reasoning, not the template.

Three ideas summarize WIT:

Writing is thinking.
WIT is a question generator, not a checklist completer.
Advance the research. Grow the researcher.

For details, see WIT-Scientific-thinking-and-writing-skill.md.

中文说明见 README-cn.md.

License

WIT uses split licensing:

  • The executable Agent Skill and software components — especially wit/SKILL.md and executable code/scripts — are licensed under the Apache License 2.0 (Apache-2.0).
  • The WIT framework and human-readable content — including the README files, wit/references/, wit/tests/, and wit/case-studies/ — are licensed under the Creative Commons Attribution 4.0 International License (CC-BY-4.0).

See LICENSE for the exact scope and attribution guidance. Full license texts are provided in LICENSE-APACHE-2.0 and LICENSE-CC-BY-4.0.