de-llm

Rewrites English prose to remove LLM stylistic tells while preserving substance.

2|Updated Apr 2, 2026
One-click install
npx skills add https://github.com/nseng-ai/ns --skill de-llm-nseng-ai
Or copy as Structured Prompt for Agent
Please help me install this Agent Skill.
Skill: de-llm
Source: https://github.com/nseng-ai/ns/tree/main/skills/incubating/writing/de-llm
Command: npx skills add https://github.com/nseng-ai/ns --skill de-llm-nseng-ai

SYSTEM DOCUMENTATION & REQUIREMENTS

What problem does it solve? AI-generated and AI-assisted prose carries recognizable stylistic tells — grandiose verbs, symmetric contrasts, rhythmic triplets, bold-lead bullet scaffolding — that undermine credibility with readers. Manually hunting these patterns across a long document is slow and easy to get wrong. ## Core Features & Use Cases - Full-passage rewriting: Re-expresses prose from scratch in the author's voice rather than patching tells one by one, since the structural scaffolding is often the loudest tell. - Substance preservation: Guarantees every claim, citation, qualification, normative word (MUST/SHOULD/MAY), and logical relationship survives the rewrite, with a mandatory sentence-by-sentence verification pass. - Cataloged tell detection: Targets word-level tells ("leverage", "delve", "seamless"), sentence-mechanics tells (em-dash chains, nested appositives), and structural tells (artificial parallel sections, empty summary sentences). - Use Case: You drafted a design document with an AI assistant and it reads like generic LLM output. Paste the text and receive a rewritten version in your voice plus a short change list of the largest stylistic edits. ## Quick Start Paste your AI-drafted prose and ask the agent to rewrite it to remove LLM stylistic tells without changing any of its substance.

Frequently Asked Questions about de-llm

High-intent search queries and answers about installing and using this skill.

FAQPage Schema
How do I remove AI writing tells from my document?

Paste the prose and request a rewrite. The skill re-expresses each passage from scratch in the author's voice rather than patching individual tells, then returns the rewritten text with a short list of the largest stylistic changes.

What LLM writing patterns does this detect?

It targets word-level tells like "leverage", "delve", and "seamless", sentence-mechanics tells like em-dash chains and nested appositives, and structural tells like bold-lead bullets, artificial parallel sections, and empty summary sentences.

Does rewriting change the technical content of my text?

No. Substance is immutable: every claim, citation, qualification, and normative word such as MUST or SHOULD is preserved. A mandatory sentence-by-sentence verification pass compares the revision against the source to catch any drift.

Can I use this on text with code blocks and citations?

Yes. Code blocks, inline code, commands, quoted material, citation titles, versions, and identifiers are treated as verbatim zones and copied exactly. Listed words inside verbatim zones are not treated as tells.

When does it use surgical edits instead of a full rewrite?

Surgical edits are the fallback only when re-generation is risky and tells are purely local, such as a short span dense with figures, citations, or normative language where fidelity matters more than style.