infographic-powerpoint-deck

What this skill does

# Infographic PowerPoint Deck

## Default input model (raw article first)

This skill should work directly from the user's raw article content by default.

If the user provides a full article, sermon, report, or long notes:
1. Read the source content and infer the audience, burden, density, and argument/story structure.
2. Build `slides_message_plan.md` from the article structure.
3. Turn `slides_message_plan.md` into `slides_display_plan.md`.
4. Turn `slides_display_plan.md` into `slides_visual_plan.md`.
5. Write one image prompt per slide and generate the deck.

If the user already provides `slides_message_plan.md` or an equivalent content-first slide table, you can skip the article-intake step and use that directly.

If the user provides one or more reference slides/screenshots, use `references/reference_slide_intake.md` to extract the reusable visual grammar before writing prompts.

## Artifact model

Use three distinct planning artifacts:

- `slides_message_plan.md`
  - content / presenter planning only
  - what each slide is meant to teach, explain, or say in full
  - the detailed speaking content or narrative burden behind the slide
  - source/article anchors, scripture/evidence anchors, and must-preserve claims
  - what beats must stay separate

- `slides_display_plan.md`
  - what will actually be shown on the slide
  - title, verse fragment, module headers, bullets, labels, captions, footer references
  - explicit on-slide text after text-budget and archetype decisions
  - can be much shorter than the message plan

- `slides_visual_plan.md`
  - downstream visual-production planning
  - deck archetype
  - style pack choice
  - layout routing
  - scene choice
  - text budget and other production constraints

`slides_message_plan.md` is the source of truth for what the presenter means to communicate.
`slides_display_plan.md` is the source of truth for what text the audience actually sees.
`slides_visual_plan.md` is the source of truth for design and prompt production.

## Quick start (default: images-only deck, structured planning)

1. Start from either:
   - raw article content / notes and use `references/article_intake_workflow.md`, or
   - a prebuilt `slides_message_plan.md`, or
   - an underspecified content slide table that you will normalize first
   - optional reference slide images / screenshots and use `references/reference_slide_intake.md`
2. If starting from a raw article, create `slides_message_plan.md` first.
3. Infer one **deck archetype** from the approved message plan using `references/deck_archetype_routing.md`.
   - If the user explicitly requests a style, use that as an override.
4. Pick one **style pack folder**.
   - If user does not specify style and archetype does not strongly imply a different choice, default to `editorial-light`.
   - Style pack chooses the deck's visual language (palette, lighting, typography attitude, scene bias). It does **not** force one fixed layout for every slide.
5. Create `slides_display_plan.md` from `slides_message_plan.md`.
   - Decide what the audience should actually see on each slide after considering archetype, density, and likely layout family.
   - Do not force the full speaking content onto the slide.
6. Create `slides_visual_plan.md` from `slides_display_plan.md`.
   - Add style, layout, scene, and text-budget decisions here, not in the message plan.
7. Gather the style-pack material:
   - You may run `scripts/compose_style_pack_blocks.py --pack-id <id>` as a helper, or manually read the style-pack files.
   - Treat the script output as helper context only, not as the final prompt body.
   - The concrete prompt should still be authored by the model in plain instruction prose.
8. For each slide in `slides_visual_plan.md`, fill `references/prompt_template.md`:
   - Use the style-pack material as helper context and rewrite it into plain instruction prose.
   - Write the prompt instructions in English by default.
   - Paste exact required on-slide text from `slides_display_plan.md` in the user-specified language(s).
   - Add slide-specific scene layer details and icons using the visual plan.
   - Keep internal routing IDs such as `L1` or `L10` in planning metadata only. Translate the routed layout into plain composition language inside the concrete image prompt.
   - If using a reference image or edit-style tool call, keep that mechanic in the tool call. The concrete prompt should still read as result-oriented art direction such as `Using the provided image as the compositional base...`, not tool language such as `edit this image`.
9. Generate each slide as one **16:9 image**.
   - Allowed slide-creation modes in this skill are **image generation** and **image editing/reference** only.
   - In the slide prompt text, explicitly include a ratio lock sentence (e.g., `Hard canvas constraint: 16:9 widescreen. Do not generate a square image.`).
   - In the tool call, explicitly set ratio config when supported (e.g., `generation_config` with `aspect_ratio: "16:9"`).
   - If consistency needs reinforcement, you may use an approved prior slide image or background exploration image as an input/edit reference, but the image tool must still render the final full slide image.
10. QA readability + text accuracy; regenerate only failed slides.
   - If any slide is not 16:9 or has text defects, regenerate the image directly.
   - Do **not** crop, pad, resize, or post-process generated images unless the user explicitly asks.
   - Do **not** render text afterward with Python/PIL/PPT overlays as part of the normal skill workflow.
11. Assemble images-only PPTX with `scripts/build_images_only_pptx.py`.

Outputs to produce in the user’s workspace:
- `slides_message_plan.md` (detailed per-slide speaking / teaching plan; source of truth for what each slide is meant to communicate)
- `slides_display_plan.md` (audience-facing on-slide text plan; source of truth for displayed text and labels)
- `slides_visual_plan.md` (visual-production plan; source of truth for archetype/style/layout/scene decisions)
- `prompts.md` (the exact prompts used for each slide)
- `slides/slide01.png ...` (final 16:9 images)
- `deck_images_only.pptx` (concatenated images)

## Optional structured slide table input

For higher-quality multi-step workflows, you may start from either a raw article or an already-structured message slide table:
- Raw article content is a valid first-class input to this skill. Use `references/article_intake_workflow.md`.
- If the upstream artifact is already `slides_message_plan.md`, preserve it as the message source of truth.
- If the upstream artifact already decides deck length, section coverage, or which beats must stay on separate slides, treat that as the controlling message plan unless the user explicitly asks for further compression or expansion.
- If the upstream content table is simple or underspecified, normalize it into `slides_message_plan.md` using `references/slide_table_normalization.md` before display planning.

## Prompt stack (minimal vs full)

- Minimal stack (recommended for most decks):
  - `references/article_intake_workflow.md` (when starting from raw article content)
  - `references/deck_archetype_routing.md` (infer the overall deck form first)
  - `references/style-pack-catalog.md` (choose pack ID)
  - `references/style-pack-system.md` (load order + composition rules)
  - `references/layout_routing_policy.md` (auto-route layouts by default)
  - `references/prompt_example_library.md` (use examples and anti-examples to avoid panel drift)
  - `references/high_fidelity_prompt_playbook.md` (when the user wants a more precise prompt)
  - `references/reference_slide_intake.md` (when user wants the deck to learn from a reference slide)
  - `references/style-packs/<pack-id>/` (pack-local blocks)
  - `references/typography_spacing_lock.md`
  - `references/text_fidelity_block.md`
  - `references/negative_prompt_block.md`
- Full stack (use when you want maximum narrative + visual control):
  -