Present Paper

Present-Paper Skill

Phase 0: Init & Outline

Step 0a — Load design references (read before drafting outline)

Read three files now, in full. Read the rest only when the answer to Q0/Q2 tells you which one you need — a talk has one venue and one style, and reading the others teaches nothing you will use.

Read now (always):

A. references/ai_slide_tells.md — the marks a generated deck leaves. Read all of it, first. Building against it is why the deck does not need catching later; scripts/check_slide_tells.py catches what slips through (Step 3.6). Where another reference conflicts with it, it wins. Eyebrow labels and brand footers on every slide are the single most-cited visual tell.

B. references/presentation_archetypes.md — the skeleton, chosen by where the speaker is standing: conference oral, journal-club critique, case-anchored grand rounds, didactic lecture, defence, keynote (Duarte's sparkline, the Jobs STAR moment, Takahashi/Lessig), lay talk, decision brief (Minto's pyramid, action titles, Kawasaki's 10/20/30). The archetype (what the talk has to do) and the visual style (what it looks like) are independent choices. The skin is a preference; the skeleton is not. Its mechanical half is scripts/check_deck_budget.py.

C. references/presentation_design_guidelines.md — the enforceable rules (assertion headlines, 24-pt floor, negative space, ≤3 colours, colourblind-safe palettes, redraw-don't-screenshot, animation discipline) plus the G1–G10 self-check the Phase 3.5 critic scores against.

Read on demand — after Q0/Q2 tell you which one:

Required Inputs

Collect these before starting. Do not draft until the target audience is defined.

Step 0b — Template & visual style selection

Before drafting the outline, settle how the deck will look. Ask (use AskUserQuestion; skip a question the user already answered in their request):

Q0 — "Where are you standing, and for how long?" (venue + minutes)

This decides the archetype — the skeleton — before any question about looks. Map the answer with the selector table in references/presentation_archetypes.md, and carry archetype + minutes forward: Step 3.6 checks the built deck against them.

If the user gives only a topic and no venue, ask. Do not guess: a deck built for no particular room comes out generic in exactly the way every reviewer can see.

Q1 — "Do you have an institutional or branded template to use?"

• Yes → the user supplies a .pptx/.potx. Switch to Mode C (Phase 3, "Fill an institutional template"). Do not also ask Q2 — the template's theme is the style.
• No / none → ask Q2.

Q2 — "Which visual style should I generate in?" Offer the CATALOG.md menu with a one-line preview each (recommended option first, labelled):

Record the choice and pass the matching style spec to Phase 3. With no preference and a medical academic talk, default to Nature / Lancet. Style changes only Phase 3 rendering — not the outline, script, or Q&A.

Q3 — conference decks only: is slide 1 a submission requirement? Many societies require the title slide to carry the title, authors, affiliations and country exactly as entered in the abstract submission. Do not shorten them to satisfy the density check — the requirement wins. Copy the fields from the submission portal, and record SLIDE_TOO_DENSE on slide 1 as consciously overruled with that reason.

Paper Analysis

Read the paper and produce a structured analysis:

## Paper Analysis

### Citation
[Full citation with DOI]

### Background
- What gap does this paper address?
- What was known vs. unknown before this study?

### Study Design
- Type: [RCT / cohort / case series / meta-analysis / etc.]
- Subjects: [n, inclusion/exclusion]
- Methods: [key methodological choices]
- Primary outcome: [what was measured]

### Key Results
1. [Finding 1 with effect size and CI/p-value]
2. [Finding 2]

### Patient/Case Summary Table
[If applicable — structured table of individual cases or subgroups]

### Limitations
1. [Limitation 1]

### Significance
- Why does this matter? What changes because of this paper?

Slide Outline

Create a slide-by-slide outline with time allocation:

## Slide Outline ([N] slides, [M] minutes)

| # | Title | Time | Key Content |
|---|-------|------|-------------|
| 1 | Title slide | 0:30 | Paper citation, presenter |
| 2 | Context / Prior sessions | 1:00 | How this connects to prior knowledge |
| ... | ... | ... | ... |
| N | Take-home messages | 0:30 | 3-5 key points |

Gate: User approves outline before proceeding.

Phase 1: Supporting Research

Limit supporting references to 5–8, and search only the categories the approved outline needs: follow-up studies (replicated or extended?), large clinical-trial data that contextualizes the findings, review articles that frame the topic, and contradicting evidence (for balanced Q&A). Do not summarize every paper found — extract only the data points the slides need (incidence, OR/HR, AUC), findings that support or challenge the main paper, and context for its significance.

Verify every reference via /search-lit (confirmed DOI or PMID). Mark any you cannot verify [UNVERIFIED - NEEDS MANUAL CHECK].

## Verified References

### Main Paper
1. [Citation] — PMID: XXXXX, DOI: XX.XXXX/XXXXX

### Supporting References
2. [Citation] — PMID: XXXXX
   → Used for: [specific data point or context]

### Key Data for Slides
- [Statistic 1]: [value] — Source: [Ref #]

Phase 2: Script & Content

Speaker Script

Draft a complete speaker script:

1. Language: the user's preferred language for narration; English for technical terms.
2. Audience adaptation: depth set by the Phase 0 audience profile. For mixed audiences, add a one-line plain-language gloss for specialty terms ("FLAIR sequence — an MRI technique that suppresses fluid signal to highlight edema").
3. Pronunciation guide in the user's language for drug names and abbreviations ("lecanemab (leh-KAN-eh-mab)").
4. Timing markers per slide. Script length must match the allocated time (roughly 130–150 words per minute for academic talks).
5. Transition phrases connecting each slide to the narrative arc.

Never invent clinical definitions, diagnostic criteria, or guideline recommendations; flag an uncertain one with [VERIFY] and ask the user.

How the sentences are built — notes are spoken, not read — and the language split for a deck that is not monolingual are in references/spoken_notes_and_bilingual.md. Read it before drafting if either applies.

## Speaker Script

### Slide 1: Title (0:30)
"[Opening — introduce yourself and the paper]"

### Slide N: Take-home Messages (0:30)
"[Summarize 3-5 key points. Thank audience. Invite questions.]"

Extension Section (Optional)

Only if the user requested it in Phase 0 — e.g. AI/computational directions, clinical practice or policy implications, connections to the user's own research.

Gate: User reviews script before proceeding.

Phase 3: Slides & Notes

Three Modes

Mode A = generate a new deck in a chosen visual style. Mode B = add notes to an existing deck. Mode C = fill the user's institutional/branded template (chosen at Step 0b). Pick the mode from the Step 0b answer.

Mode A: Generate new slide deck

Generate a fully-editable PPTX from structured inline data using python-pptx. Two template libraries:

• ${CLAUDE_SKILL_DIR}/references/generate_pptx_templates.py — generic T_lead / T_text / T_table / T_image_right / … templates (its docstring lists them; build_demo_slides() exercises each). Use for journal club, grand rounds, conference talk, and short paper talks.
• ${CLAUDE_SKILL_DIR}/templates/build_pptx_nature_lancet.py — Nature/Lancet style (spec: references/slide_visual_styles/nature_lancet.md). Use for an academic lecture multi-paper survey (template #5). Functions: new_presentation, add_title_slide, add_toc_slide, add_section_divider, add_transition_slide, add_content_slide, add_glossary_slide, add_closing_slide, plus apply_fonts(prs, en=..., ko=...) and fix_app_xml().

The Nature/Lancet defaults use 20 pt body text, subtitles and glossary entries. They suit the conference_oral, critique, case_anchored, didactic and defence budget profiles with concise content. keynote, lay_talk and decision_brief require larger type and layout adaptation; selecting a profile in the checker does not restyle the deck. Capacity bounds do not guarantee readable density or fit — render the actual content and inspect it.

For figures pulled from PDFs (rather than /make-figures output), use ${CLAUDE_SKILL_DIR}/scripts/extract_pdf_figures.py (pdftoppm + PIL crop with normalized 0–1 boxes; single-crop CLI or YAML batch config), then strip journal headers, captions and surrounding whitespace with trim_caption.py, which keeps multi-panel figures and table rows intact:

python3 "${CLAUDE_SKILL_DIR}/scripts/trim_caption.py" \
  --in-dir  figures/extracted \
  --out-dir figures/cropped

When the slot expects only the figure body (the default for build_pptx_nature_lancet.py), point FIG_DIR at the cropped output dir.

Word-boundary aware markdown parser (mandatory for HLA-rich decks)

A build script that parses inline **bold** / *italic* in slide body or notes must use word-boundary lookarounds, or asterisk-bearing scientific tokens (DRB1*07:01, HLA-A*02:01, SNP IDs, footnote markers) are eaten as italic delimiters and every allele in the deck is silently corrupted:

import re
pattern = re.compile(
    r"(\*\*(?:(?!\*\*).)+?\*\*"                           # bold; inner single * allowed
    r"|(?<![A-Za-z0-9])\*[^*\n]+?\*(?![A-Za-z0-9]))"      # italic (word-boundary)
)

The bold rule tolerates an inner single * so **DRB1*04:02** stays one bold span. Put this regex in add_styled() (or equivalent) in every Nature/Lancet-style build script.

Pronunciation auto-augment for non-native presenters

When the presenter is uncomfortable pronouncing acronyms, author names, drug names, or gene symbols, append a per-slide [ Pronunciation ] section to the notes (Presenter View only):

python3 "${CLAUDE_SKILL_DIR}/scripts/inject_pronunciation_notes.py" \
  input.pptx output.pptx \
  --dict pron_dict.yaml \
  --header "[ 발음 ]"            # or any header you like

Supply a YAML/JSON dict (term → [reading, full_name]) assembled for the audience's language. Matching is word-boundary, so short acronyms (AE, OR) match only standalone; allele tokens get a reading synthesized from their base allele entry. Slides already carrying the header are skipped, so re-runs are safe.

Speaker notes statistics density

When the slide body already shows exact OR / 95% CI / p-values, the notes must not repeat them — the presenter ends up reading statistics aloud and the audience cannot keep up. Notes are a narrative with a one-line "see the slide body for the exact numbers"; exact numbers live in the body and footnotes. More than 1,000 characters with ≥5 stat tokens → compress. The measurement snippet is in references/spoken_notes_and_bilingual.md Part D §8.

Numbers in the notes are numbers you will say out loud

• Every external number in the notes is checked against the source, exactly like a number in the body. A benchmark typed from memory ("95% of 152 models rated high risk" where the source said 87%, of 148 of 171) survives a correctly cited reference list — only the digits were remembered.
• When you edit generated text, the count of injection expressions must not fall. Where a build script draws numbers from an artifact (f"{N['primary']}"), compressing the prose is where they get flattened into literals — the body stays gated and the notes silently disagree with it. Before and after:

import re
len(re.findall(r"\{N\[", src))   # must not decrease across a rewrite

Invert the same regex to pull out anything numeric not inside an injection expression — that is a literal somebody typed, and it gets checked.

Sharing-ready notes-stripped variant

Stripping notes is mandatory before a deck circulates (e.g. a professor asks for the slides), because they hold presenter-only material — second-language narrative, pronunciation hints, self-referential reminders ("Prof. ○○ will likely ask about …"):

python3 "${CLAUDE_SKILL_DIR}/scripts/strip_notes_for_sharing.py" \
  presenter_v9.pptx share/<topic>_<initials>.pptx

It clears the text of every shape on every notes page (not only the notes placeholder), removes review comments, blanks the author / last-modified-by / comments fields of docProps/core.xml, syncs the docProps/app.xml counts, and verifies all of that against the raw XML of the written file (body and figures untouched). A deck with hidden slides stops with exit 2 until you choose --drop-hidden (remove them from the shared copy) or --keep-hidden. Share <topic>_<initials>.pptx (say in the cover email that it is there for slide reuse), <topic>_<initials>.pdf (LibreOffice --convert-to pdf drops the cleared notes pages), and optionally <topic>_<initials>_references.zip (a Drive link if it exceeds the attachment limit).

The companion documents leave with the deck — check them too

_qa_prep.md, _quick_review.md and any handout are drafted from the same material as the notes, and they travel further. Before any of them goes out:

• Retired numbers. A gate on the deck is not a gate on its siblings — slides once said 26.3% while the review sheet the speaker would answer questions from still said 54.1%. Whenever the deck's numbers move, sweep the sibling .md files for the superseded values. Fence a deliberately quoted old figure (<!-- superseded-quotation -->) rather than exempting the file.
• Operator material in an audience document. A brief written to prepare you ("the point they will push back on", "the fallback position", a minute-by-minute plan) reads, handed to the person it describes, as a strategy for managing them. Make it two documents: an internal one with the contingencies, and a neutral one with the findings and what you ask them to confirm.
• Live-only devices, and what points at them. A redistributed deck is read at a desk: remove the break slide, the timer, "as we just saw in the exercise" — then grep the whole deck, bodies and notes, for the sentences that referred to them. Re-verify anything the material asserts about an upstream project (install commands, counts, versions) against that project's own source, not your months-old copy.

Architecture

Every slide is a template-function call with explicit inline data, producing native, editable text frames. Three rules keep slides stable:

1. No markdown parsing. Markdown auto-parsed into slides drifts on every regeneration.
2. No cur_top cumulative position tracking. Use the fixed coordinate zones defined at the top of generate_pptx_templates.py — cur_top accumulates rounding errors and breaks layout after ~10 slides.
3. No Marp. Marp renders to images; the deck becomes uneditable and reviewers cannot copy text or restyle.

Figure source formats (when consuming /make-figures output)

• Preferred: PNG at ≥300 dpi via add_picture(). Set img_pct (T_image_right) so the figure occupies ≥40% of slide width on a 13.33 × 7.5-in layout.
• Vector source: convert PDF → PNG at the target DPI (pdftoppm -r 300 input.pdf out_prefix) before insertion; python-pptx PDF embedding is unreliable across PowerPoint versions.
• Forbidden: TIFF (Mac PowerPoint silently drops it); JPEG for line art (artifacts on diagonals); raw SVG (PowerPoint Mac handles it inconsistently).
• Caption: re-draft for a listener with 5–10 seconds of attention, not the journal legend.
• Put the claim in the slide, not in the raster. A conclusion or headline number drawn into the PNG stops tracking the talk the moment the bullet above it is edited, and no text search can find it. Numbers and conclusions live in the slide's own text.
• Generated illustrations: concepts and scenes only, never anything that could be mistaken for a measurement — no generated CT, MRI, histology, or radiograph, not even "as an illustration". Read references/generated_illustrations.md before the first prompt.

Diagrams and plots are drawn as CODE, then inserted (not out of autoshapes)

Hard rule, and the highest-yield rule in the skill. Agent-built slides assembled in a PPT tool almost always fail; drawing diagrams and plots in a well-known tool as code and inserting the result is what works.

Then insert the rendered PNG (≥300 dpi) with add_picture().

Check the rendered PNG before you insert it. A stroke laid on the figure's boundary is half-cut by the render and reads on the slide as a box with a side missing:

python3 scripts/check_diagram_edges.py diagrams/ --json qc/diagram_edges.json

DIAGRAM_EDGE_CLIP reports ink within a few pixels of the image border and leaves full-bleed images alone. Run it straight after savefig, where the fix is an inner margin plus bbox_inches="tight" and a pad — once the PNG exists, cropping cannot bring the stroke back.

Why the ban. An autoshape diagram produces two AI tells at once: identical rounded rectangles (SHAPE_MONOTONY) joined by unlabelled arrows (ARROW_NO_SEMANTICS). In Graphviz an edge is written with what it claims:

digraph mechanism {
  rankdir=LR; node [shape=box, fontname="Inter"];
  catheter -> tract   [label="seeds along"];
  tract    -> nodule  [label="grows into"];   // an arrow that says what it means
}

An arrow is a claim — causes, becomes, flows into, is compared with, predicts. Unlabelled, every person in the room supplies a different verb. See references/ai_slide_tells.md §4–5.

The one exception: a single, deliberate, labelled shape used as an accent (a callout box, a highlight frame). One shape is a choice; eight identical ones are a generator.

The font is a delivery decision, not a taste decision

A typeface missing on the presenting machine is substituted silently: metrics change, lines re-break, a box that fitted stops fitting — invisible on the authoring machine, visible on the projector.

python3 scripts/check_font_portability.py output/presentation.pptx --json qc/font_portability.json

FONT_NOT_PORTABLE names any typeface bundled with one operating system and absent on the other, with a count per font. It is a blocklist, not an allowlist (a licensed brand face is not its business); it exempts fonts the deck embeds, and treats a theme-level default as inert until the deck contains text of the script that slot serves. A pass does not verify font installation or renderer substitution. For the Nature/Lancet builder, call apply_fonts after adding slides and before saving to select installed Latin and East Asian faces (it does not embed fonts or touch fonts inside images, tables or charts). Verify the exported PDF's fonts as well as its layout.

Two ways to be safe:

• Embed the fonts (PowerPoint: Save > Embed fonts in the file). Licence permitting.
• Carry a PDF — but PDF drops embedded video, so a deck with a clip must ship its MP4s separately.