English | 中文
Writing rules for AI agents that need to produce factual, restrained, and reviewable text.
AI agent writing often fails in predictable ways:
- unsupported claims;
- inflated tone;
- business jargon and abstract stock phrases;
- weak structure;
- renamed identifiers;
- repeated information;
- missing references;
- conclusions made from insufficient material.
This repository keeps a short rule set that can be pasted into prompts, agent instructions, review checklists, or writing workflows.
- HumanWrite.flat.en.md: English rules.
- HumanWrite.flat.md: Chinese rules.
Use the rules as constraints before asking an agent to write, rewrite, summarize, review, or translate.
Example instruction:
Follow HumanWrite.flat.en.md strictly when writing or reviewing the output.
These rules focus on writing behavior. They do not define a style brand, content strategy, or project-specific review process.
The goal is simple: if a claim cannot be checked, it must not stay in the text.