An editing skill that scrubs AI fluff from papers and grants

What it does

This is essentially a carefully tuned SKILL.md file—plus examples—that you point at an AI coding assistant. It runs an audit→rewrite loop that scrubs generic AI phrasing like “In recent years…”, “paves the way,” and “delve” from academic drafts. It is calibrated against real accepted papers and funded NSF/NIH proposals, and it comes in two flavors: one for manuscripts and one for grants.

The interesting bit

Instead of blunting the prose like blog-focused humanizers, it preserves the boring-but-necessary parts of scholarship: passive voice where appropriate, evidence-tied hedging, definitions, symbols, and every citation and number. The grant mode even knows that NSF and NIH reviewers score the first pages hardest, so it spends its effort there.

Key highlights

• Targets academic-specific AI tells (rule-of-three, em-dash abuse, inflated verbs like “prove”) and downgrades claims to match evidence (prove becomes show empirically).
• Leaves real scholarship untouched: citations, numbers, references, and field-specific conventions stay exactly as written.
• Includes a dedicated NSF/NIH grant mode that preserves vision statements and prioritizes first-page clarity.
• Designed to be forked and personalized against your own prior accepted work, not used as a one-size-fits-all filter.
• Runs as a skill for Claude Code, Codex, or MorphMind; there is no standalone app to install.

Caveats

• It is a prompt configuration, not software, so you need an AI assistant that can load SKILL.md files to use it.
• The authors explicitly note it does not generate findings, invent data, or evade AI-use detection policies.

Verdict

Grab this if you are tired of manually de-bloating AI drafts and want an editing pass that respects scholarly precision. Skip it if you are looking for a standalone app or a tool to game peer review.