Source: https://github.com/aipoch/medical-research-skills

When to Use

Use this skill when you need to produce slide images as SVG files (plus outline/prompts), especially when:

1. A user asks for a “slide deck”, “PPT”, or “presentation” but explicitly wants images (SVG) rather than PPTX/PDF.
2. You need a readable, shareable deck optimized for scrolling/swiping (each slide stands alone without a presenter).
3. You want a style-driven deck using presets (e.g., corporate, minimal, blueprint) or custom style dimensions.
4. You need a structured workflow with optional review gates (outline review, prompt review) before generating images.
5. You need to regenerate specific slides (e.g., only slide 3 or slides 2,5,8) without rebuilding everything.

Key Features

• SVG-only output: Generates .svg slide images (no PPTX/PDF generation).
• Deterministic file structure:
  • Output directory: slides-svg/
  • Naming: NN-slide-[slug].svg
• End-to-end workflow: Analyze → confirm → outline → prompts → SVG render → summary.
• Style system:
  • Preset styles (e.g., blueprint, corporate, sketch-notes, notion, etc.)
  • Custom style dimensions: texture, mood, typography, density
  • Automatic preset recommendation based on content signals
• Review controls:
  • --outline-only, --prompts-only, --images-only
  • Optional outline/prompt review steps via confirmation questions
• Selective regeneration: --regenerate <N> or --regenerate 2,5,8
• No image-model dependency: SVG rendering is performed by a Python script using configuration parameters.

Dependencies

• Python 3.10+
• Project script: scripts/generate_svg.py (invoked locally)
• Reference/spec files (used by the workflow):
  • references/analysis-framework.md
  • references/outline-template.md
  • references/base-prompt.md
  • references/layouts.md
  • references/design-guidelines.md
  • references/dimensions/*.md
  • references/styles/<style>.md
  • references/config/preferences-schema.md

Example Usage

CLI (skill invocation)

/slide-deck "Explain OAuth 2.0 for product managers"
/slide-deck "Explain OAuth 2.0 for product managers" --style corporate
/slide-deck "Explain OAuth 2.0 for product managers" --audience executives
/slide-deck "Explain OAuth 2.0 for product managers" --lang en
/slide-deck "Explain OAuth 2.0 for product managers" --slides 10
/slide-deck "Explain OAuth 2.0 for product managers" --outline-only
/slide-deck "Explain OAuth 2.0 for product managers" --prompts-only
/slide-deck "Explain OAuth 2.0 for product managers" --images-only
/slide-deck "Explain OAuth 2.0 for product managers" --regenerate 3
/slide-deck "Explain OAuth 2.0 for product managers" --regenerate 2,5,8
/slide-deck  # then paste content

Generate SVG files locally

The SVG renderer reads parameters from its configuration area and does not use CLI arguments:

python scripts/generate_svg.py

Expected outputs:

• outline.md
• prompts/ (one prompt file per slide)
• slides-svg/NN-slide-[slug].svg

Implementation Details

Output constraints

• Format: .svg
• Output directory: slides-svg/
• Filename pattern: NN-slide-[slug].svg

Style system

Presets

Style dimensions

Full dimension specs: references/dimensions/*.md

Automatic preset recommendation (signals → preset)

Design philosophy (reader-first decks)

Decks are designed for reading and sharing, not live presenting:

• Each slide is understandable without narration.
• Logical flow when scrolling/swiping.
• Each slide includes necessary context.
• Optimized for social sharing.

See: references/design-guidelines.md (visual hierarchy, density, colors, fonts, recommendations).
Layouts: references/layouts.md.

Workflow and state files

Step 1: Setup & Analyze

1. Load preferences from EXTEND.md (if present) and summarize:

   • Style (preset/custom)
   • Audience (or auto-detect)
   • Language (or auto-detect)
   • Review enabled/disabled
     Schema: references/config/preferences-schema.md

2. Analyze content (per references/analysis-framework.md):

   • Save input as source.md (backup existing as source-backup-YYYYMMDD-HHMMSS.md)
   • Detect language
   • Recommend style preset from content signals
   • Estimate slide count
   • Generate topic slug

3. Check existing content (required) before confirmation:

   • If output directory exists, ask how to proceed:
     • Regenerate outline (keep images)
     • Regenerate images (keep outline)
     • Backup and regenerate all ({slug}-backup-{timestamp})
     • Exit
   • Save analysis.md with topic, audience, signals, recommended style, slide count, language.

Step 2: Confirmation (required)

• Two rounds:

  • Round 1: always
  • Round 2: only if “Custom dimensions” is selected

• Round 1 questions (AskUserQuestion):

  1. Style (recommended preset / alternative / custom dimensions)
  2. Audience (general/beginners/experts/executives)
  3. Slide count (recommended / fewer / more)
  4. Review outline? (yes/no)
  5. Review prompts? (yes/no)

• Round 2 (custom dimensions): AskUserQuestion for texture, mood, typography, density.

After confirmation:

• Write results back to analysis.md
• Record skip_outline_review and skip_prompt_review

Step 3: Generate outline

• If preset: read references/styles/{preset}.md
• If custom: combine from references/dimensions/
• Follow references/outline-template.md
• Save as outline.md
• Stop if --outline-only
• If skip_outline_review → skip to Step 5, else Step 4

Step 4: Review outline (conditional)

• Show slide summary table (title/type/layout)
• Ask whether to proceed, edit outline.md, or regenerate outline

Step 5: Generate prompts

• Base prompt: references/base-prompt.md
• For each slide:
  • Extract style instructions from outline
  • Merge layout guidance from references/layouts.md when specified
• Save to prompts/
  • Backup existing prompt files as prompts/NN-slide-{slug}-backup-YYYYMMDD-HHMMSS.md
• Stop if --prompts-only
• If skip_prompt_review → skip to Step 7, else Step 6

Step 6: Review prompts (conditional)

• Show prompt list table and prompts/ path
• Ask whether to proceed, edit prompts, or regenerate prompts

Step 7: Generate SVG images

• If --images-only, start here
• If --regenerate, render only specified pages
• Render via scripts/generate_svg.py
• Create session id: slides-{topic-slug}-{timestamp}
• Backup existing images as NN-slide-{slug}-backup-YYYYMMDD-HHMMSS.svg
• Progress reporting: Generated X/N (in user language)
• Retry once on failure, then report error

Step 8: Output summary

Include:

• Topic
• Style (preset or custom dimensions)
• Total slides
• List of generated SVG filenames
• outline.md location