Przejdź do treści
Biblioteka skilli
Nauka i badania · Konwersja i formaty

scholar-slides

/scholar-slides

Zamienia artykuł naukowy lub temat badawczy w slajdy akademickie na seminarium, obronę lub wystąpienie, z dokładnymi wzorami i cytowaniami.

Repozytorium
agents-sync ↗
Licencja
MIT
Ostatnia zmiana repozytorium
05.11.2026

Ocena i bezpieczeństwo

95/ 100Klasa: dobry
Format25 / 25
Frontmatter z nazwą i opisem, poprawna nazwa, opis do 1024 znaków, nazwa zgodna z folderem.
Odkrywalność25 / 25
Opis mówi, kiedy użyć skilla, ma sensowną długość i jest w 3. osobie.
Treść35 / 35
Co najmniej 150 słów i do 500 linii, nagłówki, listy kroków, przykłady albo kod.
Higiena10 / 15
Bez zaślepek typu TODO, bez ścieżek Windows, z licencją.
Ryzykobrak sygnałów w skanie
Pliki w paczce
71
Skrypty
27
Komentarze HTML
0
Adresy URL w SKILL.md
0
Przestarzałość

Brak sygnałów. Repozytorium zmienione 05.11.2026, bez odwołań do wycofanych modeli i API.

Licencja

MIT, otwarta: można używać i zmieniać, z zachowaniem informacji o autorze.

Co sprawdzić przed instalacją
  • 27 skryptów w paczce. Przejrzyj je przed instalacją, bo Claude może je uruchomić.

Skrypty: install.sh, scripts/build_deck.mjs, scripts/crop_figure.py, scripts/detect_figures.py, scripts/export_pptx.mjs, scripts/fetch_bib.py, scripts/ingest_pdf.py, scripts/lib/emphasis.mjs, scripts/lib/escape.mjs, scripts/lib/figure.mjs, scripts/lib/img_size.mjs, scripts/lib/layouts.mjs, scripts/lib/math.mjs, scripts/lib/notes.mjs, scripts/lib/pptx_layout.mjs, scripts/lib/qa.mjs, scripts/lib/render_check

Co robi i kiedy

Zamienia artykuł naukowy lub temat badawczy w slajdy akademickie na seminarium, obronę lub wystąpienie, z dokładnymi wzorami i cytowaniami.

Wyzwalacze z opisu

make slides / 做成 PPT / 讲一下这篇论文
Oryginalny opis (EN)

Use when the user wants a research paper, arXiv/DOI link, or research topic turned into academic slides — journal club / lab meeting (组会汇报), thesis defense (答辩幻灯片), or conference / job talk (学术报告) — or shares a paper and asks to "make slides / 做成 PPT / 讲一下这篇论文", especially when equations, numbers, figures, and citations must stay accurate and editable. NOT for marketing/pitch decks with no scholarly source, NOT for writing the paper itself, and NOT for stylized visual/image decks (use paper-slide-deck for those). No poster or grant-pitch workflow exists in this skill.

Jak działa

  1. Locked stack (see docs/03-design-proposal.md §0)
  2. The non-negotiable integrity gate (ALWAYS on — see references/integrity.md)
  3. Pipeline (7 stages, 3 human checkpoints)
  4. Scripts (scripts/)
  5. Learnings: read before changing pipeline behavior
  6. Build status
Sekcje z nagłówków SKILL.md, w oryginale · 109 linii, 908 słów
Podgląd SKILL.md
---
name: scholar-slides
description: >-
  Use when the user wants a research paper, arXiv/DOI link, or research topic turned
  into academic slides — journal club / lab meeting (组会汇报), thesis defense (答辩幻灯片),
  or conference / job talk (学术报告) — or shares a paper and asks
  to "make slides / 做成 PPT / 讲一下这篇论文", especially when equations, numbers,
  figures, and citations must stay accurate and editable. NOT for marketing/pitch decks
  with no scholarly source, NOT for writing the paper itself, and NOT for stylized
  visual/image decks (use paper-slide-deck for those). No poster or grant-pitch
  workflow exists in this skill.
---

# scholar-slides

Turn a research paper (or topic) into a **fidelity-first** academic slide deck. Academia
inverts the priorities of a normal deck tool: **source fidelity > polish, evidence >
persuasion, editability > flash.** Equations, tables, numbers, figures, and citations stay
**true text/vector and traceable to the source** — never rasterized by an image model,
never fabricated.

Design rationale and the survey it is built on live in `docs/`; build plan in `IMPLEMENTATION_PLAN.md`.

## Locked stack (see docs/03-design-proposal.md §0)
- **Render backend:** reveal.js + KaTeX → vector PDF (Playwright) + editable PPTX (parity-tested). Beamer later.
- **Default deck type:** lab meeting / journal club (组会) — reading-first, high density.
- **Citations:** Zotero-first (`mcp__zotero__*`), Crossref/arXiv/DOI fallback.
- **Language:** bilingual, English-default.

## The non-negotiable integrity gate (ALWAYS on — see references/integrity.md)
Apply at every stage; enforced at Checkpoint 3:
- **Never fabricate** numbers, citations, or figures. Everything traces to the source digest.
- **Math/tables/citations stay vector/text** — never an image model's pixels.
- **Reuse real figures** (bbox-cropped + `Figure N [cite]`); a conceptual schematic may be
  redrawn only if labeled "redrawn, not from source".
- **Flag, don't invent:** an unresolved figure/number/citation emits a visible
  `[MISSING: …]` / `[UNVERIFIED: …]` placeholder surfaced to the user — never a silent fill.
- **Provenance is mandatory:** every reused asset carries its source; every claim is traceable.

## Pipeline (7 stages, 3 human checkpoints)

```
INPUT → 1.INGEST/DIGEST → [CKPT-1] → 2.DECK-TYPE → 3.OUTLINE → [CKPT-2]
      → 4.PER-SLIDE SPEC → 5.RENDER → 6.SELF-REVIEW(QA) → [CKPT-3] → 7.EXPORT
```

1. **Ingest → typed digest** — `references/ingestion.md`. Run
   `scripts/prepare_source.py <pdf|arXiv id|arXiv URL>` to build the digest-input bundle
   (text, figure/table inventory with bboxes, cropped figures), then synthesize the typed
   **paper digest** from it (grounded in the bundle, never invented).
   **→ [CKPT-1]** confirm the digest with the user (contributions, headline result,
   figure→slide map, any FLAGGED asset).
2. **Deck-type & parameters** — `references/deck-types.md`. Pick deck type (default 组会),
   audience, time budget → slide budget, density register, required archetypes.
3. **Narrative outline** — `references/narrative.md`. Re-sequence the paper into talk order;
   action-title per slide; arc-tension check.
   **→ [CKPT-2]** approve the story arc.
4. **Per-slide spec** — `references/slide-spec.md`. One self-contained, on-disk spec per slide
   (layout, content, equations, figures, citations, speaker notes).
5. **Render** — `references/math.md`, `references/figures-tables.md`, `references/citations.md`,
   `references/charts.md`, `references/design-system.md`, `references/export.md`. reveal.js + KaTeX;
   real tables; real bbox-cropped figures (furniture-stripped; one panel per slide for multi-panel);
   data-bound redrawn charts; Zotero/BibTeX citations. Pick the visual theme via `meta.theme`
   (`journal-club` default / `conference`).
6. **Self-review QA** — `references/qa-self-review.md` + `references/integrity.md`, then the
   **aesthetics loop** in `references/aesthetics-review.md`. Integrity scan + static validator +
   render screenshots + narrative/timing checks; the render pass also measures geometry
   deterministically (canvas voids, figure-text projected below the 12px legibility floor). Then
   score the rendered *pixels* on the 6-dimension rubric (adversarial persona), **write the scores
   to `<deck>/aesthetics_report.json`** (schema in aesthetics-review.md), and rework any slide with
   a dimension ≤ 2 or total < 18 until the report's `rework` list is empty — the QA gate flags a
   missing report (P3) and an unspent rework list (P2). It also nudges on bullet-ratio and layout
   monotony.
   **→ [CKPT-3]** truth sign-off: review every `[MISSING]`/`[UNVERIFIED]` and generated asset.
7. **Export** — `references/export.md`. Vector PDF (projection) + speaker notes + editable PPTX;
   verify the PPTX preserves the spec natively with `scripts/verify_pptx_parity.py` (protects the
   minimal-manual-edits promise). Regress the whole corpus with `node scripts/run_benchmark.mjs`.

Read **only** the reference file for the stage you are in (progressive disclosure — never glob
`references/`). Each stage reads the **confirmed artifact of the prior stage** — the digest
(Stage 1, user-confirmed at CKPT-1) and then `deck.json` — rather than re-deriving from the paper.

## Scripts (`scripts/`)
Ingestion — pipeline stage 1 (Python, via `./.venv/bin/python`; deps in `requirements.txt`):
- `prepare_source.py` — **Stage 1 entry point**: PDF/arXiv → digest-input bundle.
- `ingest_pdf.py` — PyMuPDF text+layout extraction; arXiv-id detection.
- `detect_figures.py` — figure/table caption inventory + bounding-box localization.
- `crop_figure.py` — clean single-figure bbox crop (no neighbor columns).

Deck build & render — pipeline stage 5 (Node, via `node`; deps in `package.json`, `npm install` + `npx playwright install chromium`):
- `build_deck.mjs` — **render entry point**: `deck.json` → self-contained reveal.js deck.
- `render_deck.mjs` — deck → one-page-per-slide **vector PDF** (and per-slide PNGs for QA).
- `lib/` — `layouts` (registered-layout lock), `math` (KaTeX), `table`, `figure`, `escape`, `qa`.

Citations & QA — pipeline stages 5–6 (citations = Python; QA = Node):
- `fetch_bib.py` — citation resolver (arXiv/DOI/Crossref) with order-sensitive title verification.
- `qa_report.mjs` — **CKPT-3 gate**: integrity scan + `validate_deck.mjs` + `verify_slides.mjs` + timing.

Speaker notes (Node):
- `speaker_notes.mjs` — `deck.json` → `notes.md` handout + bilingual talk-length estimate.
- `lib/notes.mjs` — timing + handout pure logic; design system in `assets/templates/themes/`.

Export & charts — pipeline stage 7 (export = Node; charts = Python):
- `export_pptx.mjs` — `deck.json` → **editable PPTX** (native text/tables/notes; figures+equations as images).
- `make_chart.py` — chart spec → **data-bound** Okabe–Ito plot (values plotted verbatim).

Tests: `./.venv/bin/python -m pytest` (Python) and
`node --test tests/*.test.mjs` (Node — 逐个点名会静默漏跑新测试文件,用 glob).

## Learnings: read before changing pipeline behavior

[`learnings/`](learnings/) holds post-incident notes with measured numbers (figure-size
vertical budget, craft rules vs looks). Scan it before changing rendering, sizing, or QA
behavior; append a dated note after any rework-worthy surprise.

## Build status
The whole pipeline is implemented and validated on real papers. Not built: Beamer export,
OCR for scanned PDFs, draft-deck ingestion; conference/答辩 deck types are authoring guidance
in `references/deck-types.md` (no dedicated code). A deck is "done" only after the CKPT-3
human truth sign-off — the gate surfaces defects but cannot certify truth.
Pliki w paczce (71)
  • .gitignore 1 KB
  • IMPLEMENTATION_PLAN.md 34 KB
  • LICENSE 1 KB
  • README.md 7 KB
  • SKILL.md 8 KB
  • install.sh 3 KB
  • package-lock.json 9 KB
  • package.json 1 KB
  • pyproject.toml 1 KB
  • requirements.txt 1 KB
  • assets/templates/deck-stage/print.css 1 KB
  • assets/templates/deck-stage/tokens.css 5 KB
  • assets/templates/deck-stage/viewport-base.css 1 KB
  • assets/templates/themes/base-theme.css 13 KB
  • assets/templates/themes/conference.css 2 KB
  • assets/templates/themes/journal-club.css 1 KB
  • docs/01-landscape-research.md 127 KB
  • docs/02-academic-needs.md 26 KB
  • docs/03-design-proposal.md 30 KB
  • docs/04-panda-aesthetic-audit.md 5 KB
  • docs/README.md 2 KB
  • docs/gallery/deepseek-conf-cover.png 96 KB
  • docs/gallery/glm5-cover.png 66 KB
  • docs/gallery/glm5-critique.png 228 KB
  • docs/gallery/glm5-pipeline-figure.png 138 KB
  • docs/gallery/glm5-scoreboard-table.png 165 KB
  • docs/gallery/glm5-zh-cover.png 78 KB
  • learnings/2026-07-09-figure-size-vertical-budget.md 3 KB
  • learnings/2026-07-14-porting-craft-rules-not-looks.md 3 KB
  • learnings/2026-08-09-zotero-paper-to-journal-club-deck.md 4 KB
  • learnings/2026-10-10-默认路径错一位门就静默不存在.md 1 KB
  • learnings/README.md 1 KB
  • references/aesthetics-review.md 12 KB
  • references/charts.md 2 KB
  • references/citations.md 2 KB
  • references/deck-types.md 3 KB
  • references/design-system.md 5 KB
  • references/export.md 3 KB
  • references/figures-tables.md 7 KB
  • references/ingestion.md 6 KB
  • references/integrity.md 3 KB
  • references/math.md 3 KB
  • references/narrative.md 3 KB
  • references/qa-self-review.md 3 KB
  • references/slide-spec.md 5 KB
  • scripts/build_deck.mjs 5 KB
  • scripts/crop_figure.py 5 KB
  • scripts/detect_figures.py 19 KB
  • scripts/export_pptx.mjs 12 KB
  • scripts/fetch_bib.py 8 KB
  • scripts/ingest_pdf.py 4 KB
  • scripts/make_chart.py 5 KB
  • scripts/prepare_source.py 5 KB
  • scripts/qa_report.mjs 8 KB
  • scripts/render_deck.mjs 3 KB
  • scripts/run_benchmark.mjs 4 KB
  • scripts/speaker_notes.mjs 2 KB
  • scripts/validate_deck.mjs 2 KB
  • scripts/verify_pptx_parity.py 10 KB
  • scripts/verify_slides.mjs 5 KB
  • scripts/lib/emphasis.mjs 2 KB
  • scripts/lib/escape.mjs 1 KB
  • scripts/lib/figure.mjs 1 KB
  • scripts/lib/img_size.mjs 1 KB
  • scripts/lib/layouts.mjs 8 KB
  • scripts/lib/math.mjs 2 KB
  • scripts/lib/notes.mjs 2 KB
  • scripts/lib/pptx_layout.mjs 2 KB
  • scripts/lib/qa.mjs 9 KB
  • scripts/lib/render_checks.mjs 5 KB
  • scripts/lib/table.mjs 2 KB

Wersje

RepozytoriumWynikLicencja
Ta strona · yukeyao20-star/agents-sync95MIT
luwill/research-skills95MIT

Migawka bazy z 13.11.2026 · wersja paczki ffcae66c4387 (skrót SHA-256 całej paczki)

Pobrania w ostatnich 30 dniach: 0 · ranking

Zgłoś problem z tym skillem: szkodliwa treść, błędny opis albo prośba autora o usunięcie.

Podobne skille

  • 100academic-slides
    Pomaga tworzyć i udoskonalać prezentację naukową i samo wystąpienie: strukturę, rozkład slajdów, próby, timing i slajdy zapasowe.
  • 100paper-to-podium
    Tworzy wystąpienie na podstawie artykułu: scenariusz mówiony i slajdy, sprawdzone i gotowe do próby.
  • 100academic-pptx
    Pomaga tworzyć albo poprawiać prezentację naukową: wystąpienie konferencyjne, obronę pracy, briefing grantowy albo wykład.
  • 90k2f-research-deck
    Zamienia artykuły i notatki z badań w prezentację na zebranie zespołu lub seminarium, z wiernymi liczbami i zastrzeżeniami.
  • 90presentation-poster
    Pomaga przygotować wystąpienie konferencyjne, obronę, slajdy i plakat naukowy: strukturę narracji, projekt graficzny, czas, występ i pytania publiczności.
  • 100group-meeting-slides
    Przerabia artykuł naukowy w PDF na prezentację .pptx na zebranie grupy, zachowując oryginalne rysunki i tabele jako obrazy.

Uwagi i zgłoszenia

Wiadomość trafi do Bartosza Sokolińskiego. Adres e-mail jest potrzebny tylko do odpowiedzi. Szczegóły w polityce prywatności.