A Claude Code plugin that reviews documents and suggests edits to remove language patterns characteristic of AI-generated text.
Give it a document and ask it to review for LLM-isms. It reads the full text, identifies patterns that read as machine-generated, and produces line-level edit suggestions with concrete rewrites.
The plugin ships with a catalog of ~50 known patterns across categories like emdash overuse, inflated vocabulary ("delve", "landscape", "tapestry"), formulaic structures ("it's not X, it's Y"), hedge stacking, and sycophantic filler. The catalog is a starting point, not a boundary. The reviewer applies editorial judgment beyond the list.
review this document for LLM-isms
check my writing for AI tells
make this sound less AI-generated
The skill reads the document, flags issues by severity (high/medium/low), and suggests the minimum rewrite needed to fix each one. It preserves the author's voice and avoids over-correcting.
claude plugin marketplace add sst-marketplace
claude plugin install llm-isms@sst-marketplaceOr for local development:
git clone https://github.com/superscript/llm-isms /path/to/llm-isms
claude --plugin-dir /path/to/llm-isms/pluginThe LLM-ism catalog lives in plugin/skills/review/references/llm-isms.md.
To add a new pattern, add an entry under the appropriate category heading
with: pattern name, description, before/after examples, and notes on when
the pattern is acceptable. Create new category headings as needed.
BSD 3-Clause. See LICENSE.