writing-dna-skill
Writing DNA Distiller analyzes at least 20 complete articles across language, structure, topics, sources, cognitive frames, and visuals, producing four layered documents and Writing-DNA.md for an agent to reread before drafting.
Project overview
Writing DNA Distiller (writing-dna-skill) turns a corpus of complete articles into writing rules an AI agent can reread. It analyzes language, structure, topics, sources, cognitive frames, and visual presentation, then produces layered documents and Writing-DNA.md. It is intended for building a personal style asset or comparing authors, not for promising faithful imitation from a handful of excerpts.
Core capabilities
Turn style into six layers of rules
Combines vocabulary, sentence length, and punctuation with openings, transitions, source selection, and recurring judgments rather than listing favorite words alone.
Capture visual presentation
For visual corpora, sample 5–10 articles and inspect every image for evidence, explanation, or narrative roles, alongside layout, hierarchy, and color.
Create reusable pre-draft context
Four layered documents plus an integrated file form the style asset. Reread them and 5 relevant originals before drafting, without transferring source facts into the new piece.
Limit wording cleanup to explicit rules
The bundled whitelist targets matched writing tells after drafting, leaves the article structure intact, and does not override expressions supported by the target DNA.
Installation and usage
This is a file-driven agent workflow. Make the repository accessible to an agent that can read directories, load the complete SKILL.md, and point it at your corpus. This guide uses the README's direct-reading route rather than inventing a universal CLI installer.
Let an AI Agent install it
Send this prompt to Codex, Claude Code, or another AI agent that can work with your local environment.
Help me set up and use writing-dna-skill (Writing DNA Distiller) from https://github.com/larashero3-dotcom/writing-dna-skill. Read README.md, SKILL.md, and docs/usage-boundaries.md completely; for English artifacts also read references/workflow.en.md. Use the README's direct-reading method and retain the templates, references, and bundled skills/lieflat-less-ai-tone/. Confirm my corpus path, output path, and artifact language, and check for at least 20 complete articles. Produce _meta/, the four layered artifacts, and Writing-DNA.md, then check metadata coverage and source-backed rules. If the corpus is insufficient, report readiness only, not verified results. Ask before requesting credentials, additional permissions, or overwriting files. Do not publicly upload my corpus.- An AI agent that can read local directories and Markdown; image inspection is also needed for visual corpora.
- At least 20 complete .md or .txt articles you have permission to process, preferably spanning dates, topics, and formats.
- Separate corpus and output directories, with images and meaningful formatting preserved.
Project URL: https://github.com/larashero3-dotcom/writing-dna-skill/blob/main/SKILL.md
Copy the URL and follow the steps below to complete setup.- 1Prepare a complete corpus
Create an author directory and put at least 20 complete .md or .txt articles in raw/ or raw-corpus/. Preserve dates, titles, images, and formatting. Do not substitute examples/format-only/ directory placeholders for a real corpus.
- 2Load the workflow and choose a language
Download and extract the repository into a location the agent can read, retaining its relative directory structure. Have it read the full SKILL.md and, for English artifacts, references/workflow.en.md. Specify corpus and output paths and select templates/author-corpus/zh/ or templates/author-corpus/en/ for the requested artifact language.
- 3Run the analysis and inspect five artifacts
Use the README's invocation: Use writing-dna-skill to distill the writing style from every article in this directory. Ask the agent to build _meta/, complete all six analysis layers, and produce four layered documents plus Writing-DNA.md. Check coverage and source evidence using the verification criteria above.
- 4Test reuse with a new topic
Provide the new subject, audience, and verified facts. Have the agent reread all five artifacts, select 5 relevant originals using _meta/, and describe their shared voice before drafting. Apply the bundled lieflat-less-ai-tone rules afterward, preserve DNA priority, and review the result manually.
Check that _meta/ covers at least 80% of the corpus and that all four layered documents plus Writing-DNA.md contain rules traceable to originals, not empty templates. Try a new topic, confirm that the agent rereads every artifact and 5 relevant originals, and manually review style and factual accuracy.
- Chinese and English artifact filenames are defined in their respective templates; choose the set matching the requested output language.
- English outputs: language-dna.md, structure-patterns.md, cognitive-framework.md, visual-style-guide.md, and Writing-DNA.md.
- Public documentation does not supply a universal CLI installer. Automatic discovery by a particular host requires separate verification.
Use cases
Build a personal writing asset
Turn an archive of blog or publication articles into reusable structure, wording, and image guidelines for future drafts.
Reduce drift across a team
Create shared rules from authorized brand content so writers and agents can consult the same documents before drafting.
Compare authors' approaches
Analyze each author's corpus separately, then compare topic angles, evidence choices, and structure without merging them into a single style.
Assessment
Based on the README, SKILL.md, and template structure, the main editorial value is a set of inspectable style artifacts and mandatory recalibration before each draft. The workload is also explicit: complete source articles and repeated reading. AgentMatter reviewed public materials only and did not perform an installation or real-corpus test.
Why it may be useful
- Six layers cover language, editorial judgment, and visual presentation.
- Explicit metadata, layered artifacts, and pre-draft reading requirements.
- Chinese and English templates with corpus-permission and authorship boundaries.
What to know first
- Requires at least 20 complete articles; a few short samples only demonstrate the workflow.
- Distillation and repeated pre-draft reading consume context; cost and results depend on the host, model, and corpus quality.
- Repository examples show directory format, not measured results; AgentMatter has not installed or tested a real corpus.
README
writing-dna-skill
Overview
Writing DNA Distiller analyzes at least 20 complete articles across language, structure, topics, sources, cognitive frames, and visuals, producing four layered documents and Writing-DNA.md for an agent to reread before drafting. Writing DNA Distiller - distill and recreate any writing style as an agent skill
Getting started
- This is a file-driven agent workflow. Make the repository accessible to an agent that can read directories, load the complete SKILL.md, and point it at your corpus. This guide uses the README's direct-reading route rather than inventing a universal CLI installer.
- The README defines six layers: language, article structure, topic logic, source strategy, cognitive frames, and visual style.
- Input consists of at least 20 complete .md or .txt articles. Preserve images and meaningful formatting, and build _meta/ before analysis.
- Check that _meta/ covers at least 80% of the corpus and that all four layered documents plus Writing-DNA.md contain rules traceable to originals, not empty templates. Try a new topic, confirm that the agent rereads every artifact and 5 relevant originals, and manually review style and factual accuracy.
Configuration
Project URL: https://github.com/larashero3-dotcom/writing-dna-skill/blob/main/SKILL.md
Copy the URL and follow the steps below to complete setup.