Author Methodology Analysis

Problem to Solve

Analyzing an author's writing method can miss stable signals if done only by reading: opening patterns, paragraph rhythm, evidence density, dialogue feel, and structural moves. Multi-author comparison needs auditable indicators, not impressions alone. This skill turns Markdown/TXT samples into offline style analysis for imitation briefs and comparison artifacts.

How It Works

It uses lightweight local feature extraction and produces reviewable outputs:
- Article similarity clustering: uses term vectors and cosine similarity, generating similarity-matrix.csv and clusters.json to identify style-consistent text groups.
- Style radar: outputs style-radar.png around clarity, density, rhythm, evidence, dialogue, and structure, with metric boundaries explained instead of absolute quality scores.
- Imitation brief: generates imitation-brief.md, capturing language DNA, common openings, transitions, paragraph rhythm, evidence habits, reusable prompts, and guardrails.
- Multi-author comparison: supports author subdirectories and outputs author-comparison.csv, comparing structure, density, dialogue feel, and evidence density.
Default settings live in config/default.yaml; thresholds, minimum article counts, outputs, and preferences can be adjusted. A successful run typically produces local files such as report.md and manifest.json.

Boundaries

This is heuristic text analysis, not literary judgment or fact verification. Metrics are useful for initial screening, comparison, and drafting guidance. Before publishing or high-fidelity imitation, reviewers should check the original text for tone, factual accuracy, and copyright limits.