What problem does it solve?
Chinese NSFC grant proposals drafted with AI often contain machine-flavored writing such as formulaic transitions, pseudo-contrast sentences, engineering-protocol jargon, and repetitive cross-section content, which weakens the expert tone reviewers expect.
Core Features & Use Cases
- Four-layer diagnosis: Detects machine-flavored patterns at the word, sentence, paragraph, and section levels, including pseudo-oppositions, specification-style field strings, subjectless process sentences, and terminology drift.
- Hard constraint protection: Preserves LaTeX commands, citation keys, math, numbers, code status tokens, and safety invariants verbatim, with a post-rewrite invariant audit and manual-confirmation fallback.
- Configurable control: Supports section type, domain field, rewrite strength, output mode (including diagnosis-only and change summary), and self-evaluation rounds via config.yaml defaults.
- Use Case: Paste a LaTeX-mixed paragraph from the Research Content section of an NSFC proposal and receive a humanized rewrite with a change summary, term table, and protected-token diff confirming zero semantic loss.
Quick Start
Ask the AI to use the nsfc-humanization skill to polish the pasted NSFC proposal paragraph while keeping all LaTeX commands, citations, numbers, and code status tokens unchanged.
