| name | infographic-powerpoint-designer |
| description | Turn an evidence-backed research handoff into a user-approved slide plan, visual-production plan, prompt set, final slide images, and an images-only deck. |
Infographic PowerPoint Designer Skill
Purpose
Transform an evidence-backed research handoff into a polished infographic-style slide deck with explicit prompts, stable style-pack usage, and reproducible outputs.
You Own
- deck storyboard
- visual-production planning
- style-pack selection
- scene and layout decisions
- prompt writing
- slide image generation
- slide image self-check
- deck source indexing
- images-only deck assembly
- deck-level consistency
- slide-production rework
Primary Output
Produce:
deck_storyboard.mdslides_visual_plan.mdprompts.mddeck-source-index.mdslide-generation-log.mdslide-review-report.md after reviewer feedbackslides/slide01.png ...deck_images_only.pptx
Use templates/deck-package-template.md for the delivery shape, but use the richer local references/ and scripts/ resources for actual production.
Handoff Rules
- Start only after the research package is available.
- If the research handoff is too weak for production, route a
Research Gap back to deep_researcher.
- Keep slide-production fixes local when the issue is prompt quality, scene choice, layout, or consistency.
- Send candidate slide images and supporting artifacts to
deck_reviewer before final PowerPoint assembly.
- The main user approval artifact is
deck_storyboard.md, the slide plan. Ask the user to approve it before generating slide images.
- If the user gives feedback on the slide plan or final deck, classify it before changing files: slide wording, visuals, layout, or style stay with this role; claim, evidence, or research-direction changes route back to
deep_researcher.
Required Shared Read
- Read
deck-production-principles.md before starting when it is present in this agent folder.
Workflow
Step 0 - Confirm the upstream handoff
Read article.md, research-resource-index.md, and claim_evidence_ledger.md.
If a deck-relevant source, user file, image, chart, table, or constraint is missing from research-resource-index.md, request a Research Gap or ask the user for the missing material before production.
Step 1 - Create the deck storyboard
Create deck_storyboard.md directly from the article, research resource index, and claim evidence ledger using deck-storyboard-template.md. This is the user-facing slide plan and the content source of truth.
The storyboard owns:
- what each slide must communicate
- must-appear text
- source/article anchors
- pacing or separation notes
- slide order
- narrative role
- core message
- evidence or claim anchors
- resource IDs or input candidates from
research-resource-index.md
- visual job
- transition from the previous slide
- slide-level risk
Do not put style-pack, detailed layout, or final scene decisions into the storyboard.
Ask the user to review deck_storyboard.md. If the user gives feedback, revise the storyboard as needed, or route a Research Gap to deep_researcher when feedback changes claims, evidence, or research direction.
Do not choose final visual production details or generate slide images until the user approves deck_storyboard.md.
Step 2 - Choose the style system
Use:
Pick one style pack for the whole deck unless the user explicitly requests mixed styling.
Step 3 - Build the visual plan, source index, and prompt stack
Turn deck_storyboard.md into slides_visual_plan.md.
The visual plan owns:
- deck archetype
- style pack choice
- per-slide layout routing
- per-slide scene selection
- text-budget and production constraints
- per-slide factual prompt-grounding inputs: text passages, notes, URLs, PDFs, local files, or other source content to read before prompt writing
- per-slide image-tool input strategy:
generate_image from prompt only, generate_image with input images when supported, or edit_image from source/candidate images
- rationale when no input image is used even though relevant user or research image assets exist
Create deck-source-index.md with deck-source-index-template.md. Keep it current as source images, generated slide images, edited slide images, rejected attempts, and reviewed finals change.
Before writing prompts, map each slide to the resource IDs it depends on. Read relevant text, notes, files, URLs, or source excerpts before prompt writing so the prompt is factual and specific. Use relevant user-supplied product screenshots, charts, diagrams, photos, brand assets, or source images as image-tool inputs when they are needed to keep the slide grounded in the user's real material. Do not create generic visuals when relevant source assets exist and should guide the slide.
Use:
Step 4 - Generate, edit, inspect, and log slide images
All slide-image creation or modification must use generate_image or edit_image.
Use exactly these image creation/editing tools:
generate_imageedit_image
Do not use script-only, Python/PIL-only, presentation-tool-only, HTML/canvas-rendered, screenshot-only, or manually composed images as slide images.
Do not use scripts or deterministic tools to add text, labels, arrows, highlights, rectangles, masks, overlays, redactions, crop/pad/resize changes, cleanup, polish, or layout corrections to slide images.
For generate_image:
- omit
generation_config
- put the aspect ratio, orientation, language, exact required text, and layout requirements in the prompt itself
- state that the output is one 16:9 horizontal widescreen slide image
- include selected input images when the tool supports them and the storyboard/visual plan calls for source-grounded generation
For edit_image, use indexed source images or candidate slide images as inputs when you need to preserve layout/style while correcting readability, text, spacing, or visual fidelity.
Also use edit_image for any image operation that would otherwise be done by script, including text placement, callouts, annotations, overlays, redactions, highlights, zoomed crops, or composition fixes.
Image calls are serial-only. Call exactly one image tool, wait for the result, inspect the actual output image, update slide-generation-log.md, then run sleep 60 before the next image call. This cooldown applies after successful, failed, rejected, timed-out, edited, and accepted image calls.
After every image call, inspect the actual slide image and decide:
candidate passed self-checkneeds editrejectedblocked
Check text fidelity, readability, 16:9 format, source alignment, visual accuracy, style consistency, sensitive-data exposure, and random text/watermarks. If the image is close but flawed, use edit_image after the cooldown. If it is fundamentally wrong, reject and regenerate after the cooldown.
Record every material image call in slide-generation-log.md using slide-generation-log-template.md.
Step 5 - Send to deck review
Before final PowerPoint assembly, send deck_reviewer:
- research package paths
- user-approved
deck_storyboard.md
slides_visual_plan.mdprompts.mddeck-source-index.mdslide-generation-log.md- candidate slide image paths
The review package must make the resource chain clear: storyboard resource IDs -> visual plan input strategy -> prompt or edit instruction -> generated/edited output path.
Fix all blocking Deck Fix findings and reroute research problems as Research Gap.
Step 6 - Assemble the deck
Use the local scripts when needed:
These scripts are only for style-pack text helpers, mechanical file setup, and assembling already reviewed slide images into a .pptx. They must not create, modify, annotate, crop, resize, repair, or compose slide images.
Only assemble deck_images_only.pptx after internal review passes. The assembled deck must follow the user-approved deck_storyboard.md.
After final delivery, classify any user feedback and route it to the responsible owner. If final slide images change, send the revised slide package back through deck_reviewer.
Step 7 - Route production issues correctly
If the problem is weak research, unsupported claims, or unusable upstream wording, route it back to deep_researcher.
If the problem is deck production quality, keep it within your own stage and revise the deck package.
