PPT Master — Cross-Agent Public Edition

🎯 把文档扔进来，拿走一个真正的、精美的、可编辑的 PowerPoint。
本地执行，不依赖任何外部 API。SVG→DrawingML 管线，原生形状、动画、演讲备注全保留。
在 Hermes、Claude Code、Codex、OpenClaw 等 skill-compatible runtime 上均可运行。

🎯 场景选择指南 — 该走哪条路？

你的场景	推荐路线	输出
从文字/文档做正式PPT	主线：Step1→Step7（标准7步）	.pptx
只有一个话题，没素材	先跑 topic-research 工作流（标准/深度两档）→ 再进主线	.pptx
有PPT/海报截图，想还原成可编辑PPTX	跑 image-to-pptx 工作流	.pptx
需要批量生成固定格式的报告	主线 + brand/layout模板锁定	.pptx
内部分享/技术演示（不一定要.pptx）	主线 + 附带 presenter-view 工作流	.pptx + HTML演讲者视图
已有成品，想检查视觉质量	主线结束后跑 visual-quality-check 工作流	质量报告
品牌统一、多模板输出	主线 + design-tokens.md 约束求解	风格一致的多版本
已有.d2/架构图要嵌入PPT	主线中 SVG渲染阶段嵌入	.pptx+图表
需要自定义动画/音乐/旁白	主线后跑 customize-animations/generate-audio 工作流	.pptx 带动画/旁白
只想读/分析已有的.pptx	不要用ppt-master → 用 powerpoint skill	内容提取

对比同类工具：

场景	用别人的？	用我们的？
纯HTML演示（多人在线分享）	html-ppt-skill	ppt-master做.pptx + presenter-view做HTML
文档→多格式（含PDF）	huashu-slides	ppt-master（.pptx为主，未来加PDF输出）
设计系统化/本地部署	Open Design	ppt-master + design-tokens.md
视觉质量优化	guizang-ppt-skill	ppt-master + visual-quality-check工作流
批量自动化生成	—	ppt-master 主打路线

Core Pipeline: Source → Init Project → [Template] → Strategist → [Images] → Executor (SVG) → Quality Check → Post-process → PPTX

[!CAUTION]

🚨 Global Execution Discipline (MANDATORY)

This workflow is a strict serial pipeline. The following rules have the highest priority — violating any one constitutes execution failure:

1. SERIAL EXECUTION — Steps MUST execute in order; each step's output is the next step's input. Non-BLOCKING adjacent steps may proceed continuously once prerequisites are met, without waiting for "continue"
2. BLOCKING = HARD STOP — Steps marked ⛔ BLOCKING require a full stop; MUST wait for explicit user response before proceeding
3. NO CROSS-PHASE BUNDLING — The Eight Confirmations (Step 4) are ⛔ BLOCKING. After user confirms, all subsequent non-BLOCKING steps proceed automatically
4. GATE BEFORE ENTRY — Each Step has prerequisites (🚧 GATE); MUST verify before starting that Step
5. NO SPECULATIVE EXECUTION — "Pre-preparing" content for subsequent Steps is FORBIDDEN (e.g., writing SVG during Strategist phase)
6. NO SUB-AGENT SVG GENERATION — Executor SVG generation MUST be completed by the current main agent end-to-end
7. SEQUENTIAL PAGE GENERATION — SVG pages MUST be generated sequentially one-by-one
8. SPEC_LOCK RE-READ PER PAGE — Before each SVG page, re-read spec_lock.md. All colors/fonts/icons/images from this file only
9. SVG MUST BE HAND-WRITTEN — Every SVG page is written by the main agent directly. NO script-based batch generation

[!IMPORTANT]

🌐 Language & Communication

• Response language: match the user's input and source materials
• Template format: design_spec.md follows English template structure; content in user's language

[!IMPORTANT]

🔌 Cross-Runtime Compatibility

• ppt-master runs on Hermes, Claude Code, Codex, OpenClaw, and any skill-compatible runtime
• Script paths use ${SKILL_DIR} (resolved at runtime by each agent)
• On Hermes: ${SKILL_DIR} = skill installation directory under ~/.hermes/skills/
• On Claude Code: ${SKILL_DIR} = skill directory under ~/.claude/skills/ppt-master/
• On Windows: use python instead of python3 if python3 not found

Directory Structure

ppt-master/
├── SKILL.md                    ← This file — complete workflow
├── README.md                   ← Public-facing README (sells the skill)
├── test-prompts.json           ← Acceptance test prompts
├── requirements.txt            ← Python dependencies
│
├── scripts/                    ← 42+ Python scripts (pipelines, utilities)
│   ├── source_to_md/           ← Document conversion (pdf, docx, excel, web)
│   ├── project_manager.py      ← Init / validate / manage projects
│   ├── svg_quality_checker.py  ← SVG validation
│   ├── svg_to_pptx.py          ← SVG → native PPTX export
│   ├── finalize_svg.py         ← Post-processing
│   ├── image_gen.py            ← AI image generation
│   ├── image_search.py         ← Web image search
│   ├── ...                     ← 28+ more (see scripts/ directory)
│
├── workflows/                  ← 13 standalone workflows
│   ├── topic-research.md       ← Web research (standard + Deep Research)
│   ├── template-fill-pptx.md   ← Fill existing PPTX template
│   ├── create-template.md      ← Create new layout/brand/deck template
│   ├── resume-execute.md       ← Split-mode execution (Phase A → B)
│   ├── verify-charts.md        ← Chart coordinate calibration
│   ├── customize-animations.md ← Animation tuning
│   ├── live-preview.md         ← Browser preview during gen
│   ├── visual-quality-check.md ← Visual quality inspection
│   ├── presenter-view.md       ← HTML presenter view
│   ├── visual-review.md        ← Per-page rubric check
│   ├── generate-audio.md       ← Voiceover generation
│   ├── create-brand.md         ← Brand-only identity preset
│   └── image-to-pptx.md        ← Image/screenshot → editable PPTX
│
├── references/                 ← Role definitions & technical constraints
│   ├── strategist.md, executor-base.md, shared-standards.md
│   ├── canvas-formats.md, image-layout-patterns.md
│   └── verified-pipeline.md    ← Pipeline verification record
│
├── templates/                  ← Layouts, brands, decks, charts
│   ├── layouts/                ← 7+ page layout templates
│   ├── brands/                 ← Brand identity presets
│   ├── decks/                  ← Full deck templates
│   ├── charts/                 ← Chart SVG templates
│   └── icons/                  ← Icon libraries (README.md for usage)
│
└── projects/                   ← Generated projects (gitignored)

Main Scripts

Script	Purpose
${SKILL_DIR}/scripts/source_to_md/pdf_to_md.py	PDF → Markdown
${SKILL_DIR}/scripts/source_to_md/doc_to_md.py	DOCX/HTML/EPUB → Markdown (pandoc fallback for legacy)
${SKILL_DIR}/scripts/source_to_md/excel_to_md.py	XLSX → Markdown
${SKILL_DIR}/scripts/source_to_md/ppt_to_md.py	PPTX → Markdown
${SKILL_DIR}/scripts/source_to_md/web_to_md.py	URL → Markdown (supports WeChat via curl_cffi)
${SKILL_DIR}/scripts/project_manager.py	Project init / validate / manage
${SKILL_DIR}/scripts/analyze_images.py	Image analysis
${SKILL_DIR}/scripts/latex_render.py	LaTeX formula → PNG
${SKILL_DIR}/scripts/image_gen.py	AI image generation (multi-provider)
${SKILL_DIR}/scripts/image_search.py	Web image search
${SKILL_DIR}/scripts/svg_quality_checker.py	SVG quality check
${SKILL_DIR}/scripts/total_md_split.py	Speaker notes splitting
${SKILL_DIR}/scripts/finalize_svg.py	SVG post-processing (unified entry)
${SKILL_DIR}/scripts/svg_to_pptx.py	SVG → native PPTX export
${SKILL_DIR}/scripts/update_spec.py	Propagate spec_lock color/font changes across SVGs

Windows note: if python3 fails (common on python.org installs), use python.

Template Index

Index	Path	Purpose
Layouts	${SKILL_DIR}/templates/layouts/layouts_index.json	Available page layout templates
Brands	${SKILL_DIR}/templates/brands/brands_index.json	Brand identity presets (color/typography/logo/voice)
Charts	${SKILL_DIR}/templates/charts/charts_index.json	Chart SVG templates
Icons	${SKILL_DIR}/templates/icons/	Icon libraries; search with ls ... | grep <keyword>

Standalone Workflows

Workflow	Path	When to Use
Topic Research	workflows/topic-research.md	User supplies only a topic — gather web sources first
Template Fill	workflows/template-fill-pptx.md	User has an existing PPTX template to fill
Create Template	workflows/create-template.md	Create a new layout/brand/deck template
Create Brand	workflows/create-brand.md	Brand-only identity preset
Resume Execute	workflows/resume-execute.md	Split mode — resume Phase B in a fresh chat
Verify Charts	workflows/verify-charts.md	Chart coordinate calibration after SVG gen
Customize Animations	workflows/customize-animations.md	User wants to tune per-object animation order/effects
Live Preview	workflows/live-preview.md	Browser-based live preview during generation
Visual Review	workflows/visual-review.md	Per-page rubric visual self-check (opt-in)
Image to PPTX	workflows/image-to-pptx.md	图片/截图 → 可编辑PPTX逆向还原
Visual Quality Check	workflows/visual-quality-check.md	视觉质量检查（对比度/留白/对齐/层级）
Presenter View	workflows/presenter-view.md	演讲者视图HTML生成（内部分享/演示用）

⚡ Topic Research 工作流支持两档搜索：标准模式（快速覆盖基本信息）和 Deep Research 模式（多源交叉→深度抓取→结构化综合），用户说"深入了解/全面调研"时自动触发。

---

Workflow

Step 1: Source Content Processing

🚧 GATE: User has provided source material (PDF / DOCX / EPUB / URL / Markdown / text description / conversation content).

No source content? If the user supplies only a topic name without files or substantial description, run the `topic-research` workflow first, then return here with its products.

Convert non-Markdown content immediately:

User Provides	Command
PDF file	python3 ${SKILL_DIR}/scripts/source_to_md/pdf_to_md.py <file>
DOCX / Word / Office	python3 ${SKILL_DIR}/scripts/source_to_md/doc_to_md.py <file>
XLSX / XLSM / Excel	python3 ${SKILL_DIR}/scripts/source_to_md/excel_to_md.py <file>
CSV / TSV	Read directly as plain-text table
PPTX / PowerPoint	python3 ${SKILL_DIR}/scripts/source_to_md/ppt_to_md.py <file>
EPUB / HTML / LaTeX / RST	python3 ${SKILL_DIR}/scripts/source_to_md/doc_to_md.py <file>
Web link	python3 ${SKILL_DIR}/scripts/source_to_md/web_to_md.py <URL>
WeChat / high-security site	python3 ${SKILL_DIR}/scripts/source_to_md/web_to_md.py <URL> (requires curl_cffi)
Markdown	Read directly

Office vector assets (EMF/WMF): doc_to_md.py / ppt_to_md.py extract embedded vector images alongside bitmaps. After import-sources, these land in images/ with image_manifest.json. Do NOT convert EMF/WMF to PNG — svg_to_pptx.py embeds them natively. Browser preview cannot render EMF (expected — PPTX is source of truth).

Failure handling: If a converter fails (missing dependency, unsupported format, corrupted file), report the specific error and suggest alternatives (e.g., "Try saving as .docx first" for legacy .doc, or "Install pandoc" for RST/LaTeX).

✅ Checkpoint — Proceed to Step 2.

---

Step 2: Project Initialization

🚧 GATE: Step 1 complete; source content is ready.

python3 ${SKILL_DIR}/scripts/project_manager.py init <project_name> --format <format>

Format options: ppt169 (default, 16:9), ppt43 (4:3), xhs (小红书), story, etc. Full list: references/canvas-formats.md.

Import source content:

Situation	Action
Has source files (PDF/MD/etc.)	python3 ${SKILL_DIR}/scripts/project_manager.py import-sources <project_path> <source_files...> --move
User-provided text in conversation	No import needed — content is in context

⚠️ MUST use --move (not copy): all source files go into sources/. After execution they no longer exist at the original location.

Failure handling: If project_manager.py reports "project already exists", prompt user for a different project name or --force. If the converter produced no output (sources/ empty), revisit Step 1.

✅ Checkpoint — Proceed to Step 3.

---

Step 3: Template Option

🚧 GATE: Step 2 complete; project directory structure ready.

Default — free design. Proceed directly to Step 4. Do NOT query any *_index.json unless triggered. Do NOT ask the user. Do NOT proactively suggest templates.

Template flow triggers ONLY on explicit directory paths from the user's initial message. A bare name without a resolvable directory path does not trigger.

User input contains	Step 3 action
One or more explicit template paths (each resolves to a design_spec.md with kind: brand/layout/deck)	Read each spec's kind, dispatch per kind matrix below, fuse if multiple
Anything else — bare names, style descriptions, or silence	Skip Step 3, free design

Three template kinds (schema: docs/zh/templates-architecture.md):

Kind	Physical dir	Contains
brand	templates/brands/<id>/	Identity: color/typography/logo/voice/icon style
layout	templates/layouts/<id>/	Structure: canvas/page structure/page types/SVG roster
deck	templates/decks/<id>/	Full replica: identity + structure + overview

Single-path dispatch:

• kind: brand → Copy design_spec.md + logo + assets. Identity locked; structure free.
• kind: layout → Copy spec + SVG roster + assets. Structure locked; identity free.
• kind: deck → Copy everything. All locked.

Multi-path fusion (different kinds): Fuse segments by priority — brand wins identity, layout wins structure, deck fills middle. See full fusion matrix in docs/zh/templates-architecture.md.

Failure handling: If template directory doesn't contain a valid design_spec.md, report the invalid path and suggest listing available templates via ls ${SKILL_DIR}/templates/layouts/.

✅ Checkpoint — Default path proceeds to Step 4 automatically.

---

Step 4: Strategist Phase (MANDATORY — cannot be skipped)

🚧 GATE: Step 3 complete.

First, read the role definition:

Read references/strategist.md

⚠️ Mandatory gate: before writing design_spec.md, Strategist MUST read_file templates/design_spec_reference.md and follow its I–XI section structure.

Eight Confirmations

⛔ BLOCKING: Present the Eight Confirmations as a bundled recommendation set and wait for explicit user confirmation before outputting design spec.

1. Canvas format — 16:9 / 4:3 / story / custom
2. Page count range — estimated based on source material
3. Target audience — who will see this deck
4. Style objective — professional / academic / creative / minimalist / etc.
5. Color scheme — from template or recommended palette
6. Icon usage approach — icon library, stroke vs filled, density
7. Typography plan — including formula rendering policy (mixed/render-all/text-only)
8. Image usage approach — AI-generated / web-sourced / user-provided / placeholder

Split-mode note (mandatory, appended after the eight items):

• Heavy (long deck / bulky sources): Recommend split mode — stop this chat, open fresh window, use 继续生成 projects/<name> for Phase B
• Normal: Default continuous mode; can switch to split mode any time with 继续生成 projects/<name>

Formula rendering policy (inside item 7):

Policy	Behavior
mixed (default)	Complex formulas → PNG; inline expressions → editable text
render-all	All formula-worthy expressions → PNG
text-only	No rendering; formulas stay as editable text

After confirmation and before outputting design_spec.md, if formula policy is mixed or render-all:

1. Identify expressions for LaTeX rendering
2. Write <project_path>/images/formula_manifest.json
3. Run: python3 ${SKILL_DIR}/scripts/latex_render.py <project_path>
4. Include rendered PNGs in design_spec.md §VIII

Output:

• <project_path>/design_spec.md — Human-readable design narrative
• <project_path>/spec_lock.md — Machine-readable execution contract

Checkpoint:

## ✅ Strategist Phase Complete
- [x] Eight Confirmations completed (user confirmed)
- [x] Design Specification & Content Outline generated
- [x] spec_lock.md generated
- [ ] **Next**: Auto-proceed to Image Acquisition / Executor

---

Step 5: Image Acquisition Phase (Conditional)

🚧 GATE: Step 4 complete. Any formula rows have Status: Rendered.

Trigger: At least one row in design_spec.md §VIII has Acquire Via: ai or web. If all are user/formula/placeholder, skip to Step 6.

Load reference:

Read references/image-base.md

Then dispatch per Acquire Via:

Via	Load	Run
ai	references/image-generator.md	python3 ${SKILL_DIR}/scripts/image_gen.py --manifest <project_path>/images/image_prompts.json
ai (agnes)	references/agnes-ai-image-provider.md	⚠️ Do NOT use image_gen.py — its OpenAI backend sends response_format param that Agnes rejects (400). Call Agnes API directly via Python requests.post() with minimal params (model/prompt/n/size only). See key handling below.
web	references/image-searcher.md	python3 ${SKILL_DIR}/scripts/image_search.py ...

⚠️ CRITICAL — Agnes API key handling: Never let API key go through bash variable expansion. Always read from .env inside Python:

# Safe: Python reads .env directly (no bash expansion risk)
with open(".env") as f:
    for line in f:
        if line.startswith("OPENAI_API_KEY"):
            key = line.strip().split("=", 1)[1]
            break
r = requests.post(f"{base}/images/generations",
    json={"model": "agnes-image-2.1-flash", "prompt": "...", "n": 1, "size": "1024x1024"},
    headers={"Authorization": f"Bearer {key}", "Content-Type": "application/json"},
    timeout=120)
# Download returned URL:
img = requests.get(r.json()["data"][0]["url"], timeout=30)

For bash heredoc, ALWAYS use single-quoted delimiter: python3 << 'PYEOF' (NOT python3 << EOF).

Workflow:

1. Extract all rows with Status: Pending, Acquire Via ∈ {ai, web}
2. Generate prompts (ai) and/or search (web) per image-base.md §2
3. Verify each row reaches: Generated / Sourced / Needs-Manual

Failure handling: On acquisition failure, retry once, then mark Needs-Manual, report to user, and continue.

Checkpoint:

## ✅ Image Acquisition Complete
- [x] image_prompts.json / image_sources.json created
- [x] Each row: terminal status (no Pending remaining)

Default — auto-proceed to Step 6. If user opted into split mode, output Phase A hand-off marker:

## ✅ Phase A Complete
- [x] Spec: design_spec.md, spec_lock.md
- [x] Resources: sources/, images/, templates/
- [ ] **Next**: Open fresh chat → `继续生成 projects/<name>` + resume-execute workflow

---

Step 6: Executor Phase

🚧 GATE: Step 4 (and Step 5 if triggered) complete.

Read role definition:

Read references/executor-base.md
Read references/shared-standards.md
Read references/executor-general.md  # or executor-consultant.md / consultant-top.md

Design Parameter Confirmation: Before first SVG, output key design parameters (canvas, colors, fonts, body font size).

Live Preview Auto-Startup (Mandatory):

python3 ${SKILL_DIR}/scripts/svg_editor/server.py <project_path> --live

• Starts at http://localhost:5050 (port conflict → --port <other>)
• Do NOT wait for it to exit. Keeps running until user clicks Exit or says stop in chat.
• Do NOT read/apply annotations during generation — window opens after Step 7.

Pre-generation Batch Read: Before the first SVG, batch-read every distinct layout SVG in spec_lock.page_layouts and chart SVG in spec_lock.page_charts. One read per file, up front.

Per-page spec_lock re-read: Before each SVG page, read_file <project_path>/spec_lock.md. Colors/fonts/icons/images from this file only. Also read page_rhythm / page_layouts / page_charts for that page.

Visual Construction: Generate SVG pages sequentially, one at a time → <project_path>/svg_output/.

Quality Check Gate (after all SVGs):

python3 ${SKILL_DIR}/scripts/svg_quality_checker.py <project_path>

• error → Must fix before proceeding
• warning → Fix when straightforward, otherwise acknowledge and release

Logic Construction: Generate speaker notes → <project_path>/notes/total.md

Chart pages? If the deck contains data charts, run workflows/verify-charts.md before Step 7.
Visual review (opt-in)? Only if user explicitly asked — run workflows/visual-review.md.

Spec_lock drift warnings: svg_quality_checker.py may report spec_lock drift for template-native colors (#F5F7FA, #E6F3FA, etc.) not listed in spec_lock. These are safe to ignore — best practice is to add template auxiliary colors to spec_lock's colors section, but not blocking.

Checkpoint:

## ✅ Executor Phase Complete
- [x] Live preview available at URL
- [x] All SVGs generated to svg_output/
- [x] svg_quality_checker passed (0 errors)
- [x] Speaker notes at notes/total.md

---

Step 7: Post-processing & Export

🚧 GATE: Step 6 complete. Image readiness gate: every expected file must exist at <project>/images/<filename>.

Run three sub-steps one at a time — never combine them:

Step 7.1 — Split speaker notes:

python3 ${SKILL_DIR}/scripts/total_md_split.py <project_path>

Step 7.2 — SVG post-processing (icon/embed, image crop/embed, text flattening, rounded rect → path):

python3 ${SKILL_DIR}/scripts/finalize_svg.py <project_path>

Step 7.3 — Export PPTX (embeds speaker notes by default):

python3 ${SKILL_DIR}/scripts/svg_to_pptx.py <project_path>
# Output: exports/<project_name>_<timestamp>.pptx  (native)
#         backup/<timestamp>/svg_output/            (SVG source backup)

Optional animation flags (defaults already give rich entrance animations — adjust only when user asks):

Flag	Description
-t <effect>	Page transition. Default fade. Options: fade/push/wipe/split/strips/cover/random/none
-a <effect>	Per-element animation. Default auto. Options: fade/none/mixed
--animation-trigger {on-click,with-previous,after-previous}	Start mode. Default after-previous
--auto-advance <seconds>	Kiosk-style auto-play

File list: finalize_svg.py and svg_to_pptx.py do not detect missing image files at the script layer — if images are missing, the deck will have broken references. Verify files exist before Step 7.

Paragraph editability vs line fidelity: Default merge behavior collapses dy-stacked paragraphs into one editable text frame. Add --no-merge for strict line-layout fidelity.

Post-export annotation window: If user submitted annotations during generation, run workflows/live-preview.md Step 2 to apply and re-export.

Checkpoint:

## ✅ Post-processing Complete
- [x] Speaker notes split
- [x] SVG post-processed
- [x] Native .pptx exported at <path>

---

Role Switching Protocol

Before switching roles, read the corresponding reference file. Output marker:

## [Role Switch: <Role Name>]
📖 Reading: references/<filename>.md
📋 Current task: <brief description>

---

Reference Resources

Resource	Path
Shared technical constraints	references/shared-standards.md
Canvas format specification	references/canvas-formats.md
Image-text layout patterns	references/image-layout-patterns.md
Image layout sizing	references/image-layout-spec.md
SVG image embedding	references/svg-image-embedding.md
Icon library	templates/icons/README.md
Animations	references/animations.md
Design Tokens	references/design-tokens.md
Verified pipeline	references/verified-pipeline.md

---

🛑 Anti-Patterns (Do NOT Do)

• ❌ Do NOT combine steps 7.1/7.2/7.3 into one code block or shell invocation
• ❌ Do NOT use cp as a substitute for finalize_svg.py — finalize does critical processing
• ❌ Do NOT delegate SVG page generation to sub-agents
• ❌ Do NOT batch SVG pages (5 at a time) — generate one by one
• ❌ Do NOT pre-write SVG code during Strategist phase
• ❌ Do NOT skip the Eight Confirmations ⛔ BLOCKING
• ❌ Do NOT keep hardcoded personal paths, API keys, or credentials in the skill
• ❌ Do NOT default to calling external APIs for PPT generation — this skill runs fully local
• ❌ Do NOT use git reset --hard for rollback; use patch/revert
• ❌ Do NOT write "supports everything" or "fully automatic" in user-facing descriptions — be specific

🔒 Security & Safety

• Local-only: All scripts run locally. No data is sent to external PPT generation services
• API keys: AI image generation (Step 5) may require optional API keys — user configures their own
• File operations: PPT export writes to <project>/exports/. The skill never deletes files outside the project directory
• Template isolation: Custom templates are user-controlled; only reference files from the skill's template library
• Split mode: When using split mode, Phase A assets live in the project directory on disk — user controls access

✅ Acceptance Checklist

Before declaring the pipeline complete:

• Source content processed (Step 1)
• Project initialized with sources imported (Step 2)
• Template dispatched or free-design path taken (Step 3)
• Eight Confirmations confirmed by user (Step 4) ⛔
• Design spec + spec_lock generated (Step 4)
• Images acquired (Step 5, if applicable)
• SVG pages generated, one by one (Step 6)
• Quality check passed — 0 errors (Step 6)
• Speaker notes generated (Step 6)
• Post-processing complete (Step 7.1–7.3)
• Native .pptx exported and path reported to user

---

Notes

• Local preview of SVG output: python3 -m http.server -d <project_path>/svg_final 8000
• Troubleshooting: On layout overflow / export errors / blank images, check docs/faq.md
• Supported formats: 16:9 (default), 4:3, story, xhs, custom
• Supported source types: PDF, DOCX, XLSX, PPTX, MD, URL, EPUB, HTML, LaTeX, RST, CSV, conversation text