V2.5.1

Vela Slides Skill

v2.5.1 · 2026-03-20 · Three formats (full/compact/turbo), vela deck compact/expand/turbo, 262 tests, auto-expand pipeline

Generate presentation-ready slide decks and assemble them into runnable Vela .jsx artifacts. Also supports modular editing of the Vela app itself.

When to Use

• User wants to create a slide deck / presentation
• User provides a topic, agenda, outline, or document to turn into slides
• User wants to restyle or improve existing Vela slide JSON
• User wants to fix bugs or add features to the Vela app itself
• User asks for "Vela slides", "deck", or "presentation" content

Before You Start

Read the references — they contain the complete block schema, design patterns, and theming system:

Read references/block-schema.md    # All 20+ block types with properties
Read references/design-patterns.md # Layout rules, slide archetypes, composition
Read references/themes.md          # Color palettes, theming system

Architecture

For app editing (part-files, concat, template), read references/app-editing.md.

Vela CLI Setup

The vela CLI is a single entry point for all deck operations. Install once per session:

export PATH="/home/claude/.local/bin:$PATH"
ln -sf /mnt/skills/user/vela-slides/scripts/vela.py /home/claude/.local/bin/vela

Discovery (zero turns wasted):

vela --capabilities   # JSON index of all commands, flags, exit codes
vela --help           # Human-readable overview

Three Formats

Vela supports three JSON formats. All commands auto-detect and auto-expand as needed.

Format	Detection	Token savings	Use case
Full	has "lanes" key	baseline	editing, shipping, human-readable
Compact	has "S" key	~32% fewer	LLM generation (named keys = reliable)
Turbo	top-level is array	~47% fewer	storage, cache, inter-LLM context

Pipeline: any format → _load_full() auto-expands → validate → assemble → ship.

Compact Format (~32% fewer bytes)

When generating a new deck, the LLM outputs compact JSON — minified, with short keys, flat slides array, theme presets, spacer shorthand, and color palette aliases. The CLI auto-expands to full Vela JSON on ship.

Key differences from full format:

Full JSON	Compact JSON	Savings
"type": "heading"	"_": "heading"	shorter keys
"text": "..."	"x": "..."	shorter keys
{"type":"spacer","h":8}	8	spacer → int
"bg":"#0A0F1C","color":"#E6F1FF",... per slide	"t":"d" (theme ref)	theme presets
"lanes":[{"items":[{"slides":[...]}]}]	"S":[...]	flat structure
"#3B82F6" repeated 28×	"$A" + "C":{"$A":"#3B82F6"}	color palette

Compact key reference (use when generating)

Top-level: n=deckTitle, T=themes, C=color palette, S=slides array

Slide level: t=theme ref, n=title, d=duration, B=blocks, p=padding (only if overriding theme)

Block level: _=type, x=text, s=size, c=color, i=icon, b=bg, w=weight, ic=iconColor, ib=iconBg, I=items, B=blocks (nested), g=gap, lb=label, v=variant, H=headers, R=rows, sl=sublabel, slc=sublabelColor, ac=arrowColor, lc=labelColor, ts=titleSize, xs=textSize, hb=headerBg, hc=headerColor, cc=cellColor, bc=borderColor, str=striped, lo=loop, lnc=lineColor, nc=numberColor, tc=titleColor, xc=textColor, dc=dotColor, cir=circle, dir=direction

Color palette (C): define every color used ≥2× as "$A":"#hex". Use $A everywhere instead of raw hex. Assign in frequency order ($A = most common).

Theme palette (T): "d":dark, "l":light, "a":alt. Keys: b=bg, c=color, a=accent, p=padding. Theme values can use $ aliases too.

Spacers: bare int 8 instead of {"_":"spacer","h":8}

Compact commands

# Convert full → compact (for analysis/storage)
vela deck compact full.json compact.json

# Convert compact → full
vela deck expand compact.json full.json

# Ship auto-expands compact format transparently
vela deck ship compact.json    # expand → validate → assemble → copy

# All read commands auto-expand compact format
vela deck list compact.json    # works
vela slide view compact.json 3 # works

Turbo Format (~47% fewer tokens)

Positional arrays + color palette. NOT for LLM generation — for storage, cache, and passing deck context between LLM calls. The format eliminates all JSON key names by using positional arrays with implicit schemas.

Structure: [deckTitle, [lanes], colorPalette] Blocks: [type_id, ...positional values] where type IDs: 0=badge, 1=spacer, 2=heading, 3=text, 4=grid, 5=icon, 6=callout, 7=icon-row, 8=code, 9=table, 10=flow, 11=steps, 12=tag-group, 13=divider, 99=passthrough

# Convert to turbo
vela deck turbo full.json turbo.json
vela deck turbo compact.json turbo.json  # compact→full→turbo automatically

# Read/ship turbo (auto-expands)
vela deck list turbo.json
vela slide view turbo.json 3
vela deck ship turbo.json

# Expand turbo back to full
vela deck expand turbo.json full.json

Workflow with formats

# 1. LLM generates compact JSON (minified, named keys = reliable)
create_file /home/claude/deck.json   # compact format

# 2. Ship (auto-expands + validates + assembles)
vela deck ship /home/claude/deck.json

# 3. Edits work on any format (auto-expand on load)
vela slide edit /home/claude/deck.json 3 block.2.text "New heading"
vela deck ship /home/claude/deck.json

# 4. For passing deck context to another LLM call (minimal tokens)
vela deck turbo deck.json context.json   # 47% fewer tokens

Benchmark (6-slide deck, 16 operations)

Baseline (full+minified):  4,955 bytes
Compact (short keys+palette): 3,363 bytes (32.1% smaller)
Turbo (positional arrays):    2,600 bytes (47.5% smaller)

All CLI operations work on any format: list, view, edit, replace-text,
insert, move, duplicate, remove, remove-block, validate, ship.

Workflow A: Generate a Deck (most common)

Step 1: Generate Compact JSON

Always generate in compact format — it saves ~32% output tokens vs full format, and the CLI auto-expands it.

Rules for generating compact:

1. Top-level: {"n":"Title","T":{themes},"C":{color palette},"S":[slides]}
2. Color palette (C): define every color used 2+ times as $A, $B, etc. Use these aliases everywhere in the deck instead of raw hex/rgba values.
3. Themes (T): group slides by bg/color/accent/padding. "d" = dark, "l" = light, "a" = alt.
4. Slides (S): flat array. Each slide has "t":"d" (theme ref), "n" (title), "d" (duration), "B" (blocks).
5. Blocks: use short keys (_=type, x=text, s=size, c=color, i=icon, b=bg, etc.)
6. Spacers: just the int 8 instead of {"_":"spacer","h":8}
7. Minified: no whitespace, no indentation.

Complete example (3-slide compact deck)

{"n":"AI Workshop","C":{"$A":"#3B82F6","$B":"#E6F1FF","$C":"#8892B0","$D":"#0A0F1C","$E":"#3B82F620","$F":"#1e293b"},"T":{"d":{"b":"$D","c":"$B","a":"$A","p":"60px 72px"},"l":{"b":"#ffffff","c":"$F","a":"$A","p":"36px 48px"}},"S":[{"t":"d","n":"Cover","d":15,"B":[{"_":"badge","x":"WORKSHOP","i":"GraduationCap","b":"$E","c":"$A"},8,{"_":"heading","x":"AI para Equipas","s":"3xl","w":700},{"_":"text","x":"Formação prática com LLMs e agentes.","s":"lg","c":"$C"}]},{"t":"l","n":"Agenda","d":60,"B":[{"_":"badge","x":"PROGRAMA","i":"Calendar","b":"$E","c":"$A"},8,{"_":"heading","x":"Agenda do Dia","s":"2xl"},{"_":"steps","I":[{"title":"09:00","x":"Fundamentos LLM"},{"title":"13:30","x":"Agentes e MCP"}],"lnc":"$A","nc":"$A","tc":"$F","xc":"#64748b"}]},{"t":"d","n":"CTA","d":20,"B":[{"_":"heading","x":"Vamos Construir?","s":"3xl","w":700},12,{"_":"text","x":"[email protected]","s":"lg","c":"$C"}]}]}

Key patterns in the example:

• $A = #3B82F6 used 5× (badges, steps, accent) — saves ~16 tokens
• Themes absorb bg/color/accent/padding — each slide just says "t":"d" or "t":"l"
• Spacers are bare ints: 8, 12
• Steps items use "title" and "x" (text)
• Everything minified on one line

Step 2: Ship (validate + assemble + deliver)

vela deck ship /home/claude/<deck-name>-slides.json

This runs expand → validate → assemble → copy in one call.

Step 3: Present

present_files(["/mnt/user-data/outputs/<deck>.jsx", "/mnt/user-data/outputs/<deck>-slides.json"])

Workflow A½: Edit an Existing Deck

Use the vela CLI instead of view + str_replace — saves 80-97% tokens on typical edits.

Inspect

vela deck list deck.json                  # TOC: slide#, title, blocks, theme, duration
vela slide view deck.json 3               # Compact block summary
vela slide view deck.json 3 --raw         # Full JSON for one slide
vela slide view deck.json 3 --json        # Structured output for agent consumption

Edit

# Text replacement across entire deck (idempotent)
vela deck replace-text deck.json "old text" "new text"

# Slide-level property
vela slide edit deck.json 3 duration 90
vela slide edit deck.json 3 bg "#ffffff"

# Block-level property (use block.INDEX.KEY syntax)
vela slide edit deck.json 3 block.2.text "New heading"
vela slide edit deck.json 3 block.0.color "#3B82F6"

# Structural operations
vela slide remove deck.json 4
vela slide remove-block deck.json 3 5
vela slide move deck.json 5 2
vela slide duplicate deck.json 2
vela slide insert deck.json 3 new-slide.json

Safety

# Preview destructive operations before executing
vela slide remove deck.json 3 --dry-run
# Returns: {"would_execute": "remove_slide", "title": "...", "reversible": false}

Then ship

vela deck ship deck.json    # validate + assemble + copy in one call

When to use what

Task	Command	Why
Change a word/phrase everywhere	vela deck replace-text	1 call vs N×str_replace
Change one slide property	vela slide edit N key value	No need to view JSON first
Change one block property	vela slide edit N block.I.key value	Precise, zero collateral
Remove/reorder slides	vela slide remove/move	Clean item-level removal
Check what slides exist	vela deck list	~200 tokens vs ~800 for view
Check one slide	vela slide view N	~150 tokens vs ~500 for view+range
Preview before destructing	Add --dry-run	Agent safety net
Full pipeline	vela deck ship	Replaces 3-4 tool calls with 1
Rewrite a slide entirely	vela slide view N --raw → str_replace	Still best for full rewrites
Complex block structure changes	str_replace on JSON	CLI handles properties, not deep nesting

Exit codes

Code	Meaning
0	Success
1	General failure
2	Usage error (bad arguments)
3	Resource not found
4	Validation failure
5	Conflict

All commands support --json for structured output (errors include suggestions and retryable fields).

Workflow B & C: Edit the Vela App

For app editing and updating, read references/app-editing.md.

Deck JSON Format

Full Deck (for assembly)

{
  "deckTitle": "My Presentation",
  "lanes": [
    {
      "title": "Main",
      "items": [
        {
          "title": "Section Name",
          "status": "todo",
          "importance": "must",
          "slides": [ ...slide objects... ]
        }
      ]
    }
  ]
}

Single Slide

{
  "bg": "#0f172a",
  "bgGradient": "linear-gradient(135deg, #0f172a 0%, #1e293b 100%)",
  "color": "#e2e8f0",
  "accent": "#3b82f6",
  "padding": "36px 48px",
  "duration": 60,
  "blocks": [ ... ]
}

Critical Rules

Canvas

• Virtual canvas: 960x540px (16:9)
• Base padding: "36px 48px" — never less
• Max 5-7 blocks per slide
• Left-aligned = full width. Center-aligned = shrink-wrap.

Duration

• Every slide MUST have duration (integer, seconds)
• Title: 15-30s | Content: 60-90s | Dense: 90-180s | Metric: 20-40s | Quote: 15-30s

Visual Design

• Every slide MUST have bg or bgGradient
• Size hierarchy: 3xl-4xl titles > 2xl headings > lg body > md supporting > sm captions
• Use icons generously. Use spacer blocks for breathing room.
• First slide = title (centered, gradient, 4xl). Last slide = closing (gradient, strong visual).
• Vary bg across slides — alternate solid/gradient, shift hues for section breaks.

Block Selection

• Don't default to heading+bullets — vary layouts across slides
• icon-row > plain bullets for feature lists
• flow for processes | flow loop:true for cycles | flow gate:true for approvals
• svg for custom diagrams (use {{accent}}, {{color}}, {{muted}}, {{bg}} tokens)
• grid 2-3 cols for comparisons | grid 1 col + direction:"row" for layer diagrams
• metric for stats | table for structured data | progress for bars/spectrums
• steps for numbered sequences | timeline for roadmaps | tag-group for labels
• callout with icon for key insights

Content Quality

• One key idea per slide. Billboard test — readable from the back.
• Headlines = assertions ("AI Cuts Costs 30%"), not labels ("Cost Analysis")

Theming

See references/themes.md for palettes. Available directions: dark, midnight, light, warm light, vibrant, editorial, minimal. Custom brand colors supported.

Icons

PascalCase Lucide names. 1000+ available. Common: Zap, Brain, Rocket, Shield, Target, Clock, Users, Heart, Globe, Code, Database, Lightbulb, TrendingUp, BarChart, Lock, Eye, Cpu, Layers, Bot, Sparkles, Award, CheckCircle, AlertTriangle, DollarSign, ArrowRight