Stop handing reviewers the knife before they ask for it

What it does

This repo codifies one idea: a paper is a product launch, not a project status report. It ships a set of twelve writing rules (in Chinese and English) as both copy-paste prompts and installable Skills for Claude Code, Cursor, or Codex. Feed it a paragraph of your paper and it rewrites away the “we must admit”, “results should be interpreted with caution”, and “still lags behind” phrasing that pre-emptively concedes every fight.

The interesting bit

The core move is reframing: metrics you don’t win aren’t contests you enter — they’re just not the competition. The README’s before/after table is genuinely instructive, e.g. turning “we must admit we haven’t tested cross-domain generalization, so interpret cautiously” into “cross-domain generalization has not been evaluated.” Same fact, zero apology. It also assigns each experiment a specific argumentative job instead of treating the results section as a warehouse.

Key highlights

• Zero dependencies, no MCP services — it’s prompts and Markdown, nothing else
• Works with any AI via a plain copy-paste prompt, or as a Skill for agent tools
• Bilingual: full Chinese and English versions of both the skill and the quick prompt
• Includes a pre-rebuttal self-check list (the “don’t hand reviewers the knife” checklist)
• Gives decision priorities for what to cut when compressing a paper

Caveats

• The Codex plugin isn’t confirmed in the official marketplace — the README is explicit that a GitHub release is not an official listing
• It’s glue code in the best sense: the entire value is the writing philosophy, not the Python (one build script for packaging)

Verdict

Worth a look if you write papers with AI assistance and keep finding your drafts sound like confessions. If your writing is already assertive, this is a twelve-rule lecture you don’t need.