SVG PPT Generator

SVG-based PPT generator with 9 themes, 8 layouts, 30+ charts, and 600+ icons

Security Scan

The skill appears to implement an SVG→PPTX generator that matches its description, but the bundled code includes many optional image-backend connectors (OpenAI, Replicate, Stability, etc.) while the SKILL.md declares no required credentials — that mismatch and the potential for those backends to be invoked at runtime is a concern.

Purpose & Capability

The name/description (SVG-based PPT generator) matches the included templates and conversion scripts (svg_to_pptx, templates/layouts/*). However, the package also contains multiple image backend adapters (backend_openai.py, backend_replicate.py, backend_stability.py, backend_qwen.py, etc.) and an Image_Generator role — these go beyond pure SVG→PPTX conversion and imply optional networked image-generation capabilities that are not declared in the skill's metadata.

Instruction Scope

SKILL.md's runtime instructions are explicit and limited to reading bundled design_spec.md and template SVGs, writing SVG files to /tmp, running local conversion scripts, and optionally running an image-generation pipeline. However, the Technical Flow instructs importing and executing bundled Python scripts (via a workspace path) which can call the included image backends and post-processing tools. The instructions do not mention or require any API keys but the codebase contains modules that will attempt external API calls if invoked — the scope is broader than the SKILL.md declares and could trigger network activity or require credentials not listed.

Install Mechanism

No install spec (instruction-only), so nothing is downloaded at install time. That reduces risk from arbitrary remote installers. However, the skill bundle already contains many Python scripts and templates that will be present on disk when the skill is installed — these files can be executed by the agent at runtime. No external download URLs or extract steps are present in the manifest.

Credentials

Registry metadata lists no required environment variables, yet the codebase includes many image backend modules (backend_openai.py, backend_replicate.py, backend_stability.py, backend_zhipu.py, backend_volcengine.py, etc.) which typically require API keys (e.g., OPENAI_API_KEY, REPLICATE_API_TOKEN, STABILITY_KEY). This is a mismatch: either those backends are never used (benign) or they will try to read credentials from the environment at runtime (surprising to users). The SKILL.md does not document what credentials are needed or how external services are selected, so credential access is under-specified and potentially disproportionate.

Persistence & Privilege

always: false and user-invocable: true. The skill does not request to be always-enabled or to modify other skills' configs. It instructs writing temporary files under /tmp and using its own bundled scripts — normal for this functionality. The optional reviewer subagent flow increases the operational scope (spawning subagents) but autonomous invocation is not forced by the skill metadata.

What to consider before installing

This skill's description and bundled templates/scripts match — it can generate editable PPTX from SVGs — but exercise caution before enabling it: 1) The package contains many image-backend connectors (OpenAI, Replicate, Stability, Qwen, Volcengine, Zhipu, etc.). If you or the agent trigger image-generation features, those modules may attempt outbound network calls and will look for API keys in the environment; the SKILL.md does not declare those credentials. 2) If you plan to install/use this skill, review the backend_* files to see what providers and env var names are referenced, and only provide API keys you trust the skill to use (or avoid supplying keys and disable image generation). 3) Prefer limiting the skill to manual, user-invoked operations rather than allowing autonomous invocation; if possible, run it in a sandboxed environment (no sensitive env vars) or inspect/strip unused backends. 4) If you need a second opinion, ask the skill author for a minimal README that lists exactly which env vars or external services the skill will use and under what conditions; absence of such a list is a red flag.

PPT Generator Skill

Professional presentation engine. Generates SVG pages and converts them to native editable PPTX via svg_to_pptx. Includes 8 layout templates covering dark, light, consulting, tech, and more.

When to Use

• User asks to create a PPT / presentation / slides
• User provides content, outline, or data that needs to become a PPTX
• User mentions "make a PPT", "generate slides", "presentation", etc.

Interaction Flow (must follow)

After receiving a PPT request, do not generate immediately. Guide the user step by step:

Step 1: Confirm Topic and Duration

Got it, I'll make this PPT for you. Topic: "{extracted from user message}" — correct? How long is the presentation? This determines page count:

• 10 min → 10-12 pages
• 20 min → 15-18 pages
• 30 min → 22-25 pages
• 45 min → 28-35 pages

Duration is required — do not skip this. Page count directly depends on it.

If the user already provided a detailed outline, skip topic confirmation but still ask duration.

Step 2: Pick a Style

Pick a style (just reply with the number):

1. dark_warm (default) — dark warm tone, AI/tech feel

2. consultant — white + blue, consulting style

3. cloud_orange — deep navy + orange, cloud/tech architecture

4. ai_ops — full dark, ops/DevOps style

5. tech_blue — blue tech, formal business

6. smart_red — red business

7. exhibit — light showcase, data-heavy

8. pixel_retro — pixel retro, creative/fun

Step 3: Confirm Outline

Propose a structure based on the topic and style:

Based on your needs, here's a suggested structure ({N} pages):

P1 — Cover: {title}

P2 — Table of Contents

P3 — {Section 1 title}

...

P{N} — Closing: {key takeaway / CTA}

Want to adjust anything, or shall I start generating?

Step 4: Generate

Only start after user confirms. Send a status message:

Starting generation, estimated X minutes.

Then execute the Technical Flow below.

Step 5: Deliver and Iterate

Send the final PPTX with a brief note:

PPT is ready, {N} pages total. Want changes? Just say:

• "Update the data on page 3"
• "Add a page about XX"
• "Make the colors darker"

---

Technical Flow (executed in Step 4)

Read design_spec.md + reference SVGs → Write SVG files → svg_to_pptx → Deliver

Phase 1: Read Design Spec

Load the target style's design spec and reference templates:

ppt-master-assets/templates/layouts/{style}/design_spec.md   ← colors, fonts, layout rules
ppt-master-assets/templates/layouts/{style}/01_cover.svg      ← cover reference
ppt-master-assets/templates/layouts/{style}/02_chapter.svg    ← chapter page reference
ppt-master-assets/templates/layouts/{style}/03_content.svg    ← content page reference
ppt-master-assets/templates/layouts/{style}/04_ending.svg     ← ending page reference

Phase 2: Generate SVG Files

Use the write tool to create SVG files page by page in /tmp/ppt_svgs/{style}/.

SVG Rules:

1. viewBox must be 0 0 1280 720 (16:9)
2. Strictly follow design_spec.md for colors, fonts, and layout
3. Do not use: foreignObject, clipPath, mask, <style>, class
4. File naming: 01_cover.svg, 02_toc.svg, 03_chapter1.svg... in order
5. All text uses <text> elements with font-family, font-size, fill
6. Background: full-coverage <rect>. Decorations: <rect>/<circle>/<line>/<path>
7. Tables: manual <rect> + <text> layout (no HTML tables)
8. Fill the page — avoid large empty areas
9. Titles should state insights, not category labels

Suggested order:

• Cover and ending first (set the tone)
• Chapter dividers next (consistent style)
• Content pages last (data-heavy, one at a time)

Phase 3: SVG → PPTX Conversion

svg_to_pptx converts SVG elements into native DrawingML shapes (not images):

• <text> → editable text boxes (double-click to edit)
• <rect> → native rectangles (drag, recolor)
• <circle> / <ellipse> → native circles
• <line> / <path> → native lines/paths
• Tables (rect + text combos) → editable shape groups

The output PPTX is fully editable in PowerPoint, just like a manually created file.

import sys, os
sys.path.insert(0, os.path.expanduser(
    "~/.openclaw/workspace/skills/ppt-generator/ppt-master-assets/scripts"))
from svg_to_pptx import create_pptx_with_native_svg
from pathlib import Path

svgs = sorted(Path("/tmp/ppt_svgs/{style}").glob("*.svg"))
create_pptx_with_native_svg(svgs, Path("/tmp/output.pptx"),
                            canvas_format="ppt169",
                            use_native_shapes=True,  # required! otherwise SVG is embedded as image
                            verbose=True)

Phase 4: Cross Review (optional, requires user confirmation)

After generating the first draft, ask the user:

First draft is ready. Want to run a cross-review? Multiple reviewers check in parallel — more thorough but takes a few minutes.

If the user agrees and the agent supports subagents, run a review-fix cycle:

1. Round 1 (full review): Launch 5 reviewer subagents in parallel

   • 🎤 Presentation Coach — narrative arc, flow, pacing
   • 👥 Target Audience — simulated audience reaction, comprehension
   • 🔬 Domain Expert — factual accuracy, technical depth
   • 📋 Content Auditor — structure, typos, data consistency
   • 👁️ Visual Inspector — layout, alignment, readability

2. Fix: Aggregate all issues, fix by priority (🔴 before ⚠️)

3. Round 2 (regression): 3 reviewers verify the fixes

4. Exit criteria: 🔴=0 and ⚠️≤3, or round≥4 force exit

Skip if user says "no review" or "just a quick draft".

Phase 5: Deliver

Only send the final version. Do not send intermediate versions unless the user asks.

---

Style Directory Mapping

#	Style	Directory	Background
1	Dark Warm	dark_warm	Dark cover + light content
2	Consultant	consultant	White + blue
3	Cloud Orange	cloud_orange	Deep navy + orange
4	AI Ops	ai_ops	Dark
5	Tech Blue	科技蓝商务	Blue
6	Smart Red	smart_red	Red
7	Exhibit	exhibit	Light
8	Pixel Retro	pixel_retro	Dark

Default recommendation: dark_warm.

Design Spec Location

Each template in ppt-master-assets/templates/layouts/{directory}/ contains:

• design_spec.md — full design parameters
• 01_cover.svg — cover template
• 02_chapter.svg / 02_toc.svg — chapter/TOC page
• 03_content.svg — content page
• 04_ending.svg — ending page

Reference Documents

Key docs in ppt-master-assets/references/:

• strategist.md — strategist role
• executor-consultant-top.md — top-tier consulting executor
• shared-standards.md — SVG technical constraints

---