Teach agents to blueprint a book-length thesis before layout: one defensible central claim, institution-aligned chapter architecture, merged and attributed source material, and design intent that reads like a serious degree submission—not a stack of papers or a generic essay.
This skill produces thesis manuscripts as
packages. Follow the
k2f skill for authoring: unpack or init a workspace, edit the semantic tree, run pack/verify, then export. K2F keeps agents on
meaning (chapter roles, arguments, figures)—not pixel coordinates—while the engine compiles layout, typography, and theme into consistent output across PDF, DOCX, and other formats.
Before you build, install the core skill:
npx skills add suzheng/k2f --skill k2f
.
One blueprint for semantic content and aesthetic intent. Use labeled blocks the k2f skill can map to chapters and front matter later.
-
Institution & scope lock — Degree level, discipline, template or handbook rules, word/page limits, mandatory front matter (title page, abstract, acknowledgments, declarations) and back matter (appendices, bibliography). Citation style as metadata only. Cut material outside rules before drafting.
-
Central thesis claim — One paragraph: problem, gap, and what the degree work proves or establishes (empirical, theoretical, or methodological). Cap at one to three thesis-level contributions; park extras in a future-work chapter or appendix.
-
Source synthesis — Treat user inputs (drafts, PDFs, notes, published chapters) as raw results: deduplicate repeated facts across files; cluster by theme, not by filename; prefer the most complete and recent version when sources agree; keep conflicting interpretations separate and flag them. Attribute each major claim to a source tag (
,
,
). Lead the outline with the integrated story, not a file-by-file dump.
-
Chapter architecture — List chapters in submission order with word budgets summing to the limit (±10%). Typical patterns: introduction; literature/context; methods (once); one or more results chapters; synthesis discussion; conclusion—or discipline-specific variants. Per chapter: purpose, link to the central claim, and 3–6 bullet claims. Mark which chapters reuse published paper content vs net-new narrative.
-
Evidence map per chapter — Tag each bullet (
,
,
,
,
). Separate
observations (what was found) from
interpretations (what it means); empirical results chapters report; cross-chapter discussion interprets. Flag missing data, pending citations, or hypothetical results—never invent statistics, participants, or references.
-
Abstract & keywords — Structured summary (context → gap → approach → main outcomes → implications) within institutional word cap. Keywords as concepts. If multiple languages are required, specify parallel structure, not word-for-word translation.
-
Visual plan — Assign heavy quantitative or structural ideas to figures, tables, or schematics across chapters. Note caption takeaway (one sentence), consistent numbering intent across chapters, and where charts replace prose.
-
Design intent — Tone, hierarchy, density, palette for the engine:
- Tone: discipline-appropriate register; precise over promotional; examiners scan for contribution early.
- Hierarchy: chapter titles and H1s conclusion-oriented where allowed; part dividers only if the institution expects them; methods subheads procedural.
- Density: introduction and discussion narrative with short paragraphs; methods and results bullet- or table-led; no block longer than ~6 lines without a break, list, or visual.
- Palette: calm academic (neutral body, one accent for chapter titles, figure callouts, key statistics); print-friendly contrast; minimal decoration.