← ClaudeAtlas

academic-de-ailisted

Review English social-science and psychology prose for formulaic, over-performed, mechanically polished, or evidence-straining features. Use only when explicitly invoked. Review and mark only; never estimate AI authorship or edit the manuscript.
xburner23412/academic-prose-pattern-review · ★ 0 · AI & Automation · score 70
Install: claude install-skill xburner23412/academic-prose-pattern-review
# Academic De-AI Review prose features without claiming to detect authorship. A mark means “read this in context,” not “AI wrote this.” Prefer abstention to a weak mark. Source prose stays English. Explanations follow the user's language. JSON field names stay English. ## Scope - English social-science and psychology prose. - Markdown, TXT, DOCX and PDF. - Version 1 reviews and marks only. It does not rewrite, apply edits, compare versions, profile an author, or optimize detector scores. - Invoke only when the user explicitly asks for `$academic-de-ai`. ## Required reference Read [references/patterns.md](references/patterns.md) before reviewing. Treat `rules.json` as the executable rule table. ## Run 1. Run `scripts/mark_patterns.py SOURCE --out review.json`. 2. Inspect extraction status, excluded roles, source hash and partial reading coverage before interpreting marks. 3. Read all included prose. Apply the five reading rules in `patterns.md`, adjudicate automatic marks in context, and run the protection pass. 4. Set every automatic mark to `keep`, `protected_or_functional`, or `rule_misfire` with a brief rationale. Add reading marks with `origin: model_reading`; retain exact paragraph, offsets, quote and source locator. Change `reading_coverage.status` to `complete` only after every included section has been read. 5. Render with `scripts/render_marked_report.py review.json --out report.html --lang LANG`, where `LANG` follows the user's languag