Academic Slide Minimalist

Academic Slide Minimalist

Non-Bypassable Execution Gates

These gates override every other workflow section in this skill. Run them before any slide-generation or PPTX-packaging action.

Gate A: Image2 availability gate

Before creating final slide images, sample redesigned pages, or a PPTX, explicitly determine whether the current environment has an Image2-style native full-slide image-generation backend.

The backend counts as available only if all are true:

• it can generate one complete 16:9 academic slide image per page;
• the full-slide image can include title, navigation, real source figure crops, annotations, labels, notes, and page number inside the image;
• it can preserve scientific source crops faithfully without redrawing, restyling, or semantically changing them;
• its outputs can be saved as accepted final slide images and then mechanically packaged into an image-only PPTX.

If availability is unavailable, unclear, unconfirmed, or blocked by the current environment, do not create or deliver an Image2-only PPTX. First explain the boundary, then ask whether the user accepts a lower-fidelity fallback PPTX route. Until the user explicitly accepts that fallback, deliver only the valid non-image artifacts: paper reading outputs, adaptive navigation, deck_order_map.md, figure_source_manifest.md, page_briefs.md, Image2 prompts, and assembly notes.

Use this boundary statement when Image2 is unavailable:

当前环境未确认可用 Image2-style 全页幻灯片生成后端，因此无法交付严格 Image2-only 高保真 PPTX。请选择：
A. 只交付可用于后续 Image2 生成的页面方案、figure manifest、page briefs、prompts 与 assembly notes；
B. 接受降级版 PPTX：使用现有 `assets/sample-literature-report.pptx` 的模板、版式节奏和红黑灰学术风格生成可编辑 fallback PPTX；它不是 Image2-only 高保真交付，也不能声称通过 Image2 manifest 与 image-only 验证。

Gate B: Image2 manifest gate

Do not package a PPTX unless image2_manifest.json exists and records every final slide image. The manifest must include:

{
  "image2_backend_confirmed": true,
  "slides": [
    {
      "slide_no": 1,
      "final_image": "final_images/slide_001.png",
      "generation_route": "image2_full_slide",
      "accepted": true,
      "source_figures": []
    }
  ]
}

Every delivered Image2-only slide must have generation_route: "image2_full_slide" and accepted: true. If the manifest is missing, incomplete, or says any page used another route, do not deliver an Image2-only PPTX.

Gate C: PPTX validation gate

Before delivering a PPTX, run scripts/validate_image_only_pptx.py on the assembled file. If a manifest exists, validate with --manifest image2_manifest.json. Delivery is allowed only when the validator passes.

A passing PPTX must have exactly one full-slide image per slide and no editable text boxes, shapes, charts, tables, SmartArt, icons, or extra PowerPoint annotations.

Gate D: mandatory delivery statement

When delivering a PPTX, include this factual checklist in the final response:

Image2 backend used: yes
Full-slide images generated: yes
PPTX assembly mode: one full-slide image per slide
Editable slide elements: none
Source scientific visuals: real paper/SI/user-provided figures only
Validation completed: yes

If any line is not true, do not deliver the Image2-only PPTX.

Gate E: explicit fallback consent gate

Fallback PPTX creation is allowed only after all of these are true:

• Image2-style full-slide generation is unavailable, unclear, unconfirmed, or blocked;
• the user has been shown the boundary statement above;
• the user explicitly chooses the fallback route or otherwise clearly states that a lower-fidelity PPTX is acceptable;
• the fallback deck uses the bundled assets/sample-literature-report.pptx through a template-following workflow: inspect the sample deck, map every output slide to a source slide, duplicate the mapped slides, then edit inherited objects rather than drawing a parallel layout;
• the fallback deck is grounded in the current main paper, SI, or user-provided source figures; do not use mock figures, decorative substitutes, or scientific content copied from the sample deck;
• the fallback deck passes a render-based layout QA pass with no unintended overlaps, no clipped text, and no sparse pages that use only a small fraction of the slide canvas;
• the fallback deck has no unused template placeholders or leftover shapes occupying figure/text space;
• the fallback navigation follows the sample deck's quiet academic style rather than generic oversized button tabs;
• the final response and file names clearly distinguish the fallback deck from the authoritative Image2-only deck.

A fallback PPTX must not be named or described as final_presentation.pptx unless the user explicitly asks for that filename. Prefer names such as:

fallback_presentation.pptx
editable_fallback_presentation.pptx

When delivering a fallback PPTX, include this factual checklist instead of the Image2-only checklist:

Image2 backend used: no
Delivery mode: fallback PPTX accepted by user
Image2-only validation: not applicable
Scientific visuals: real paper/SI/user-provided figures only
Editable or code-rendered elements may exist: yes
Template source: assets/sample-literature-report.pptx
Render/layout QA completed: yes

Never imply that the fallback PPTX satisfies Gates B-D. If the user does not accept fallback output, continue with the non-image planning artifacts only.

Gate F: fallback template gate

When producing a fallback PPTX, use the bundled assets/sample-literature-report.pptx as the required template source. Preserve its 16:9 format, red-black-gray academic tone, title hierarchy, navigation rhythm, figure/caption treatment, and page-number/footer habits as much as the fallback tooling allows.

The fallback deck may be editable and may contain PowerPoint text boxes, shapes, and normal picture objects, but it must not be a hand-redrawn imitation of the template. Use a strict template-following workflow:

1. inspect all slides in assets/sample-literature-report.pptx;
2. use references/sample-template-slot-manifest.json as the bundled template slot reference, or regenerate it with scripts/extract_template_slot_manifest.py if the sample PPTX changes;
3. choose a source slide for every output slide;
4. create a fallback_edit_plan.json or template-frame-map.json that lists each source slide and explicit edit targets;
5. duplicate the mapped source slides into a starter deck;
6. prepare every real source figure with scripts/prepare_fallback_figure.py or an equivalent deterministic crop step before insertion, and record prepared_figure_manifest.json;
7. edit inherited text, image, table, and placeholder objects in place while preserving their original position, size, typography, crop, and frame treatment;
8. replace sample scientific content only with current paper/SI/user-provided content;
9. render the result and check overlap, clipping, sparse layout, placeholder leftovers, off-center/margin-dominated figures, and inherited-template overflow.

Use scripts/draft_fallback_edit_plan.py to create a first-pass fallback_edit_plan.json from structured slide briefs when no hand-authored plan exists. Then use scripts/build_fallback_template_pptx.py for fallback construction when possible. The builder accepts the edit plan, applies inherited slot edits from references/sample-template-slot-manifest.json, prepares source figures, emits prepared_figure_manifest.json, and writes a build report with overflow, ellipsis, untouched-slot, and image-preparation warnings. If this builder reports status: "needs_review", fix the plan/content before delivery.

Fallback image placement must be figure-aware, not a rigid slot fill. The draft plan should profile each real paper figure or table by aspect ratio, density, and layout hint. When a wide table, wide plot, or dense multi-panel figure does not fit a single inherited image slot, the edit plan may use fit_strategy: "adaptive_contain" and frame_scope: "all_image_slots" to merge inherited image slots into one larger figure region. This is allowed only inside the mapped template slide's inherited image area and must not redraw the page from scratch, stretch the scientific image, or crop away scientific content.

For literature-report fallback decks, every content slide should include at least one real paper/SI/user-provided scientific figure, table, or source crop unless it is a true cover, ending, section divider, or explicitly documented logic-only transition. Do not let background, method, result, application, or summary pages become generic text-only slides when relevant real figures or tables are available.

Do not build fallback pages with a fresh freeform layout engine, generic card grid, oversized UI navigation, or new PowerPoint objects over the copied template when an inherited source object should be edited instead. If the tooling cannot identify editable targets or cleanly replace inherited objects, stop with planning artifacts instead of delivering a loosely redrawn deck.

Never copy the sample deck's scientific content into a new paper deck. Treat assets/sample-literature-report.pptx as a template/style source only; all scientific figures, claims, captions, and conclusions must come from the current paper, SI, or user-provided material.

Read references/sample-template-intro.md, references/fallback-template-pptx.md, references/fallback-template-edit-spec.md, and references/sample-template-slot-manifest.json before creating any fallback PPTX.

The fallback workflow should mirror the useful contract of gorden-ppt-skill: use slot IDs or explicit object addresses, preserve template typography and same-level font sizes, respect capacity hints, rewrite overly long text instead of shrinking individual boxes, and keep every output slide traceable to a source template slide.

Core Mission

Create academic literature-report PPTs that are clear before they are beautiful.

The default target is a complete Chinese academic presentation deck that follows the user's sample deck rhythm while adapting to the current paper's argument.

Use the sample PPT as a visual and rhythm reference, not as a fixed set of section names. Derive the actual navigation labels and page order from the paper's domain, evidence chain, and story structure. A simple paper can use a default rhythm such as basic information, background, research idea, results, summary, and closing; a result-heavy paper should split results into meaningful sections such as design strategy, performance validation, structure evidence, mechanism explanation, application validation, or limitations.

When the user says to follow their sample PPT, do not invent a generic commercial deck and do not hard-code one universal navigation bar. Preserve the sample deck's navigation style, page rhythm, academic tone, and restrained red-black-gray visual language, while adapting section names to the current paper. Use references/sample-deck-rhythm.md, references/adaptive-navigation.md, and, when needed, the bundled assets/sample-literature-report.pptx as style references only.

Image2-Only Route Lock

This skill is an Image2-style full-slide academic deck workflow, not a code-rendered slide engine and not an editable PowerPoint layout workflow.

For any final slide page image, redesigned page, sample page, full-deck page, or delivered PPTX:

1. Image2-style native full-slide generation is the only allowed visual generation route.
2. Each slide must be generated as one complete 16:9 full-slide image.
3. The generated image itself must already contain the title, navigation, real source figure crops, annotations, labels, notes, and page number.
4. A final PPTX, when delivered, is only a mechanical container for those accepted full-slide images.
5. Each PPT slide must contain exactly one full-slide image and no editable slide objects.

Hard stop for the authoritative Image2-only route: if Image2-style full-slide generation or an equivalent native full-slide image-generation backend is unavailable, stop before Image2-only PPTX assembly and ask whether the user accepts a fallback PPTX. Do not create a substitute deck unless the user explicitly accepts the fallback route. If accepted, the substitute must be template-based using assets/sample-literature-report.pptx.

Forbidden routes for the authoritative Image2-only deck include:

• Python, Pillow, Matplotlib, or any code-drawn full-slide page;
• HTML, CSS, React, or browser-rendered slide pages;
• SVG or canvas slide rendering;
• browser screenshots of coded layouts;
• native PowerPoint layouts built from editable text boxes, shapes, charts, tables, SmartArt, icons, or card grids;
• editable PPT pages rasterized afterward and presented as Image2 output;
• generic commercial PPT templates;
• decorative AI scientific diagrams replacing real paper figures.

Allowed code usage is mechanical only:

• extract or crop real figures from paper, SI, user screenshots, or user PPT pages;
• rename, organize, and version source assets;
• create contact sheets, montages, or render audits for QA;
• compress or validate images;
• package accepted full-slide images into an image-only PPTX;
• check slide count, aspect ratio, ordering, filenames, and whether editable elements exist.

Code must never be used to design or render the authoritative Image2-only final slide page. If the user explicitly accepts a fallback PPTX, code or editable PowerPoint construction may be used only to inspect the bundled template, duplicate selected template slides, and replace inherited template slots for that clearly labeled fallback deck, while still preserving real scientific figures and honest delivery labeling. Code must not hand-redraw a slide from approximate template coordinates.

Non-Negotiable Scientific Rule

Use only real figures from the paper, supplementary information, user-provided screenshots, or user-provided PPT pages as scientific visuals.

Never invent, regenerate, redraw, or replace scientific content, including:

• molecular structures
• crystal structures
• microscopy images
• spectra
• XRD / NMR / IR / Raman / TG / DSC data
• adsorption curves
• reaction schemes
• mechanisms
• tables
• catalytic or performance charts
• any experimental data figure

Allowed operations are deterministic only:

• crop
• enlarge / reduce
• align
• mask margins
• add simple labels
• add thin boxes / arrows / lines
• add short callouts
• place the real figure into a 16:9 slide image

If a needed figure is not available, do not fabricate it. Use a text/logic page, ask for the source, or say the figure is missing.

Real Figure Fidelity Lock

Image2-style generation may compose the academic slide page, but it must not invent, redraw, stylize, reinterpret, or semantically alter scientific visuals.

All scientific visuals must come from exact source crops extracted from:

• the main paper PDF;
• supplementary information / supporting information;
• user-provided screenshots;
• user-provided PPT pages;
• other explicit user-provided source images.

The image-generation prompt must instruct the backend to preserve source figure crops as faithful embedded evidence, not reinterpret them as decorative diagrams.

Never use Image2 or any other image model to fabricate, approximate, or restyle:

• molecular structures;
• crystal structures;
• microscopy images;
• spectra;
• XRD / NMR / IR / Raman / TG / DSC data;
• adsorption curves;
• reaction schemes;
• mechanisms;
• tables;
• catalytic or performance charts;
• any experimental data figure.

If the image-generation backend cannot preserve real figure crops reliably enough for academic use, stop after preparing the figure-source manifest, slide outline, page briefs, Image2 prompts, and assembly notes. Do not replace real figures with simulated charts, redrawn mechanisms, decorative scientific icons, or code-generated substitutes.

Default Full-Deck Rule

When the user asks for a literature-report PPT from a paper:

• default to at least 20 pages unless the user specifies fewer pages.
• page count has no hard upper limit.
• prioritize explaining the paper clearly over compressing content.
• split dense results into multiple pages rather than crowding one slide.
• use one page for one claim whenever possible.

A 20+ page deck is acceptable if the paper has enough logic, figures, and results to support it.

Global Deck Control Rules

For any full-deck or multi-batch image2 task, control the deck as a single presentation project, not as independent pages.

1. Derive paper-specific navigation before page design. Do not hard-code 基本信息 | 研究背景 | 研究思路 | 研究结果 | 总结启发 | Backup; use it only as a fallback. Follow references/adaptive-navigation.md.
2. Create or maintain deck_order_map.md before image2 generation. Page order, section, navigation highlight, slide title, source figures, and mainline/backup status must come from this map. Follow references/deck-order-map.md.
3. Separate mainline and backup pages deliberately. Dense methods, full tables, full fitting parameters, and likely Q&A belong in backup unless they are central to the main story. Follow references/mainline-backup-boundary.md.
4. In multi-batch image generation, use one style contract for every batch: same aspect ratio, navigation, title scale, page-number placement, red-black-gray palette, annotation style, and figure treatment. Follow references/batch-image-consistency.md.
5. Use real scientific figures only. Do not add decorative vector icons or invented diagrams. Only annotate real figures with boxes, arrows, shaded regions, zoom windows, labels, and key-number callouts. Follow references/real-figure-annotation-rules.md and references/trend-emphasis-rules.md.
6. Track generation status for every slide. Do not guess which pages are complete; keep image_generation_status.md. Follow references/image-generation-status.md.
7. Assemble the PPT only from the final approved image folder. Follow references/version-control-and-final-folder.md, references/pre-assembly-checklist.md, and references/post-assembly-render-audit.md.
8. Authoritative final slide pages must be Image2-style full-slide images only. Code may support extraction, cropping, file naming, contact sheets, QA, compression, and image-only PPTX packaging, but code must not design or render the Image2-only final slide pages. If no Image2-style full-slide generation route is available, ask whether the user accepts a clearly labeled fallback PPTX before producing any substitute. If accepted, build the fallback from assets/sample-literature-report.pptx, not from a freeform code layout. Follow references/fallback-template-pptx.md for source-grounding, density, and layout QA rules, and follow references/fallback-template-edit-spec.md for the slot manifest, slide mapping, inherited-object replacement, and capacity checks.

Iterative Refinement and Diagnosis Mode

When the user asks to continue optimizing an existing deck, diagnose before redrawing. Use references/deck-diagnosis-mode.md to identify page-order, navigation, style, figure-readability, claim, and speaker-readiness problems. Then use references/redraw-priority.md to classify slides as A: must redraw, B: local repair, or C: keep. Do not regenerate the whole deck unless the diagnosis shows global failure or the user explicitly requests a full rebuild.

If the user has accepted a version or says the version is good, treat the final page order as locked. Use references/final-page-lock.md; later edits should preserve slide IDs, filenames, page numbers, and navigation ranges unless the user explicitly asks to reorder the structure.

Use references/sample-layout-grammar.md to choose a repeatable layout type before designing each page. Use references/figure-readability-standard.md and references/annotation-language.md before accepting figure pages. Use references/speaker-notes-to-slide-check.md and references/question-risk-level.md to make sure important pages are speakable and defensible.

Maximum-Quality Mode

When the user asks to use the strongest model, maximum quality, best possible result, or says they will provide a main paper plus SI for a complete literature-report PPT, use references/high-quality-production-protocol.md.

In maximum-quality mode, do not jump from paper reading directly to PPT generation. Produce or internally maintain these staged artifacts:

paper_logic_tree.md
terminology_table.md
main_si_crosswalk.md
figure_source_manifest.md
adaptive_navigation_plan.md
deck_order_map.md
slide_outline.md
page_briefs.md
image2_generation_plan.md
image_generation_status.md
image2_manifest.json
speaker_notes.md
quality_check_report.md
deck_diagnosis_report.md
redraw_priority_plan.md
final_delivery_preview.md
final_page_montage.png

The default deliverable remains the final PPTX, but when feasible also return traceable support files described in references/delivery-package.md.

Reading Modes

Use two related but different reading modes.

Mode A: Interactive close reading

Use this mode when the user explicitly asks to read, translate, or analyze the paper paragraph by paragraph. Follow references/close-reading-rules.md strictly. Process only one original paragraph at a time, preserve the supplied English text verbatim, translate sentence-by-sentence into academically natural Mainland Chinese, then provide deep commentary. Stop after each original paragraph and wait for the user to say “继续”, “可以了”, or “下一段”.

Mode B: PPT production reading

Use this mode when the user asks for a full literature-report PPT. Apply the same rigor internally, but do not pause after every paragraph and do not reproduce the whole paper verbatim in the final answer. Instead, transform close reading into:

• a paper logic tree;
• a main-paper + SI evidence chain;
• a figure-source manifest;
• a page-by-page PPT outline;
• slide claims and Chinese presentation copy.

For PPT generation, close reading is a source-grounding method, not a requirement to display the entire original paper text.

Mandatory Workflow for Paper-to-PPT Tasks

Follow this sequence. Do not jump directly into slide design.

0. Confirm the input contract

For the user's normal workflow, expect:

• one main paper PDF;
• one corresponding supplementary information / supporting information file;
• the sample deck style, either uploaded in the conversation or bundled as assets/sample-literature-report.pptx.

If the main paper or SI is missing, state the limitation. Do not pretend missing material has been read.

1. Read the main paper and supplementary information rigorously

When the user provides both a paper PDF and supplementary information, treat them as one evidence system. Read the main paper for the argument and the SI for experimental details, controls, extra characterization, validation, and source figures.

Use close-reading discipline from references/close-reading-rules.md: preserve terminology, avoid paraphrase drift, track claims at sentence level when needed, and never replace close reading with vague summary.

Then extract the paper's logic before making slides:

research background -> unresolved problem -> author strategy -> key methods -> key evidence -> performance/application -> conclusion -> inspiration

For each major result, identify:

• what question it answers
• which figure proves it
• what conclusion the audience should remember
• whether the evidence is direct, indirect, supplementary, or preliminary
• whether supporting details or figures come from the main paper or SI

1.5. Establish terminology and domain logic

Before writing slide copy, create or internally maintain a terminology table using references/terminology-control.md. Keep technical terms, sample names, abbreviations, units, and cautious claim verbs consistent across slides and speaker notes.

Identify the paper's domain pattern using references/domain-adaptation.md so the result section follows the right evidence chain instead of a generic figure order.

2. Build a figure-source manifest

Before slide generation, create or maintain an internal manifest:

figure id | source page/section | content | used slide | evidence role | claim supported

Use this manifest to prevent fake figures, unsupported claims, and figure-message mismatch. Use references/figure-source-manifest.md for the full manifest fields and extraction notes, references/figure-extraction-and-naming.md for asset naming/cropping rules, and references/evidence-strength.md to calibrate how strongly each claim may be worded.

3. Map the paper into an adaptive sample-style deck

Use the user's sample PPT as the visual rhythm and navigation style reference, not as a fixed outline. Consult references/sample-deck-rhythm.md, then derive paper-specific navigation with references/adaptive-navigation.md and write the final page order with references/deck-order-map.md.

Rules:

• the navigation labels must match the current paper's argument structure;
• use 5-8 short Chinese navigation labels when possible;
• default labels such as 基本信息 / 研究背景 / 研究思路 / 研究结果 / 总结启发 are only a fallback;
• split long result sections into more informative labels such as 性能验证, 结构证据, 机理解释, 应用验证, or 局限讨论 when this helps the audience follow the logic;
• every page must map to one section and one navigation_highlight in deck_order_map.md;
• backup pages must have a distinct backup state and should not interrupt the mainline story.

Each section can contain multiple pages. The result and evidence sections are usually the longest.

4. Decide the page claim before layout

Every slide should have one core claim. Titles should be conclusion-style rather than topic-style.

Weak:

• 热稳定性
• 催化性能
• 结构表征

Better:

• 高温处理后样品仍保持特征衍射峰与介孔吸附特征
• Al/Ti 引入使该骨架从结构材料转向催化平台
• 结构模型与 STEM 共同证实 36R 直通孔道

5. Prepare the generation brief and page briefs

Before generating image2 pages, prepare the overall generation brief using references/generation-brief-template.md and then prepare one page brief per slide using references/page-brief-template.md.

Every page brief must specify:

• the slide's section and page type;
• its one-sentence core claim;
• exact source figure(s) or none - logic/text page;
• evidence strength;
• what to crop or preserve;
• Chinese slide title and copy;
• layout plan;
• risk/caution note.

If the user asked to review the outline first, pause after the brief. If the user asked for direct generation, maintain the brief internally and continue.