email-designer（电子邮件设计师）

Email Designer

Generate Outlook-compatible email templates through conversation. The user describes what they want, you produce a pixel-perfect EML file that opens as a draft in Outlook.

How It Works

This skill has three layers:

1. Rules — Outlook compatibility constraints you follow when generating HTML
2. Templates — Pre-built components and layouts you assemble from
3. Code blocks — Python scripts you execute to produce EML files

The core (HTML + EML) uses Python stdlib only. Optional features (charts, header banners, image optimization) require additional packages that are auto-installed when needed — see Step 0.

Step 0: Environment Check

Before starting, verify that Python 3 is available. Try these commands (use whichever works on the current platform):

python3 --version    # macOS / Linux
python --version     # Windows (check that output shows 3.x, not 2.x)

• Exists and Python ≥ 3.8 → continue silently (do not mention the check to the user).
• Missing or version too low → do NOT proceed. Instead:

1. Run uname -s and uname -m (or systeminfo on Windows) to detect the user's OS and architecture. 2. If you have web search capability, search for the latest recommended way to install Python 3 on the user's platform, and provide the specific commands. 3. If you cannot search, show this message in the user's language: 中文: Email Designer 需要 Python 3.8 或更高版本。请访问 https://www.python.org/downloads/ 下载并安装适合你操作系统的版本，安装完成后重新运行。 English: Email Designer requires Python 3.8 or later. Please visit https://www.python.org/downloads/ to download and install the version for your OS, then try again. 4. Stop here and wait for the user to install before continuing.

Optional Dependencies (auto-installed when needed)

If Step 1 determines the email needs charts or image processing:

1. Execute code-blocks/deps-checker.py → check_and_install(features)

- features is ['charts'], ['images'], or ['charts', 'images']

1. All available → continue silently
2. Just installed → continue silently (installation is quiet)
3. Install failed → inform user and offer alternatives:

- Charts unavailable → "I'll use HTML tables and stats-grid components instead" - Images unavailable → "I'll use the HTML header component and skip compression"

Mode Detection

After environment checks pass, determine the appropriate mode based on user input and existing projects. The three modes are:

1. Import Mode — if the user provides a .eml file (or mentions importing/复刻 an existing email):

1. Execute code-blocks/eml-to-html.py → extract_from_eml(eml_path) to extract HTML and embedded images
2. Execute save_extracted(result, output_dir) to save HTML + images with CID references converted to relative paths
3. Auto-open the extracted HTML in browser for preview
4. Ask the user: "已提取邮件内容并预览。接下来您想： A) 在此基础上调整设计（进入设计模式 Step 3） B) 直接沉淀为可复用量产项目（进入沉淀流程）"
5. If A → validate the extracted HTML (code-blocks/html-validator.py → validate(html)), organize output via code-blocks/output-manager.py → create_project(name), then skip to Step 3 (Preview & Adjust) with the extracted HTML as starting point
6. If B → read rules/production-mode.md § "Crystallization Process" and follow steps C0-C5

2. Production Mode — if email-projects/ directory exists with crystallized projects:

1. Scan for subdirectories containing both template.html and template.xlsx
2. If projects exist, present them: "检测到以下邮件模板项目： {project-name-1} {project-name-2}... 请选择项目编号并提供 Excel 数据文件路径（如：1 /path/to/data.xlsx）， 或输入 'new' 创建新邮件。"
3. If user selects a project → first run code-blocks/deps-checker.py → check_and_install(features=['excel']) to ensure openpyxl is available, then read rules/production-mode.md and follow steps P0-P4
4. If user types new → continue with Design Mode

3. Design Mode — default when no EML file provided and no existing projects (or user chose new). Continue with Step 1 below.

Adaptive Flow

Not every request needs the full wizard. Match your approach to the user's input:

If the user's request already contains enough info (layout, colors, width, content type are clear from context), skip ahead. For example, "帮我做一个蓝色的产品更新邮件 模板，600px" already tells you: single-column, blue (#2563eb), 600px. Go straight to generating.

If the request needs clarification, use the guided interaction below to gather what you need. The goal is to make decisions easy — especially for non-technical users who may not know email design terminology.

Guided Interaction

Email design has 6 decision dimensions. The agent's job is to infer what it can from the user's request, then ask about the rest — one question at a time, always with a recommended default.

The 6 dimensions (ask in this order — earlier answers inform later defaults):

Interaction principles:

CRITICAL — DO NOT ask multiple questions in one message. Each message MUST ask about ONE dimension only. Listing all questions at once overwhelms the user and defeats the guided interaction purpose. ❌ BAD (dumping all questions at once): "请告诉我：1. 报告内容范围 2. 视觉风格 3. 布局偏好 4. 邮件宽度 5. 其他" ✅ GOOD (one question per message, with recommendation): Message 1: "季度商机报告 — 我推荐 Dashboard 布局（800px）。通常包含：A) KPI 指标卡 B) 趋势图表..." Message 2: "配色风格？A) 深蓝商务 B)..." Message 3: "你有现成的数据吗？A) 有 B) 先用示例数据"

• Infer first, confirm second. Parse everything the user already said. For "商机季度报告邮件", you already know: Dashboard layout, likely needs charts and KPI cards, 800px is appropriate. State your inference and ask for confirmation — don't present all 7 layouts and ask them to choose.
• One question per message. Each message asks about one dimension. This keeps the user focused and makes it easy to answer. If a dimension is already known (inferred or stated by user), skip it entirely.
• Always recommend. Every question includes a labeled recommendation with brief reasoning: "推荐 A（深蓝商务），适合正式的季度报告". Users who don't have a preference can just accept.
• Multiple choice preferred. Use A/B/C/D options with short descriptions rather than open-ended questions. Include an open option for customization (e.g., "D) 自定义颜色").
• 3 questions maximum. If you've asked 3 questions and still have unknowns, use sensible defaults for the rest. Most users lose patience beyond 3 rounds. Combine questions only when two dimensions are tightly related (e.g., layout + width).

Example flow for "帮我做一个商机季度报告邮件":

Agent infers: Dashboard layout, 800px, needs charts/KPI.

Question 1 (confirm inference + ask modules):

"季度商机报告 — 我推荐 Dashboard 布局（800px）。这类报告通常包含： A) KPI 指标卡 B) 趋势图表 C) 漏斗图 D) 分布图 E) Top 列表 F) 总结文字 推荐：全选，季度报告信息量大。你要哪些？"

Question 2 (visual style):

"配色风格？ A) 深蓝商务 B) 深绿增长 C) 紫色科技 D) 自定义颜色 推荐：A 深蓝商务，适合正式的季度报告"

Question 3 (data):

"你有现成的数据吗？ A) 有，我提供数据 B) 先用示例数据，之后替换"

Then start generating — no more questions.

Step 1: Understand the Request

Using the guided interaction above, resolve these dimensions:

• Layout: Check templates/layouts/ for options:

- Single Column (单栏) — product updates, announcements, weekly digests - Two Column (双栏) — reports with sidebar data - Magazine (杂志) — editorial with hero image and articles - Announcement (公告) — minimal, centered, call-to-action - Dashboard (数据面板) — KPI reports, analytics digests, weekly reviews - Transactional (事务通知) — order confirmations, password resets, system alerts - Onboarding (引导流程) — welcome emails, getting-started guides - Custom — user describes, you design from components

• Width: 600px (mobile-first) / 800px (balanced) / 1200px (desktop) / custom
• Colors: Brand image provided? Analyze it for colors (read the image with vision). Otherwise, offer presets: Blue #2563eb, Green #059669, Orange #ea580c, Purple #7c3aed, Gray #374151. (Full palettes in rules/brand-color-extraction.md)
• Saved templates: Run code-blocks/template-manager.py → list_templates() to check for reusable templates. If any exist, offer them first.

Material Assessment

After understanding layout/width/colors, assess what visual materials are needed:

• User provided data (Excel, CSV, table, numbers in conversation) → plan chart generation (Step 1.5a)
• User provided background image → plan header banner compositing (Step 1.5b)
• User provided local images → plan image optimization (Step 1.5c)
• Email type is Dashboard/Weekly Report/Data Report but no data provided → ask: "Do you have data you'd like visualized as charts?"

If any material preparation is needed:

1. Run dependency check (Step 0 optional deps)
2. Proceed to Step 1.5

If no materials needed → skip directly to Step 2.

Step 1.5: Material Preparation (when materials identified)

Prepare visual assets before generating HTML. All outputs go to the project's images/ directory.

1.5a: Chart Generation (if data provided)

1. Read rules/chart-design-system.md for visual constraints
2. Analyze data structure → recommend chart type:

- Categories × single value → horizontal bar - Categories × multiple series → stacked bar - Time × values → line chart - 2D matrix → heatmap - Parts of whole → pie/donut - ≤ 5 data points → suggest HTML table or stats-grid instead

1. Confirm with user: "Based on your data, I recommend a [type] chart. OK?"
2. Execute code-blocks/chart-generator.py: gen = EmailChartGenerator(container_width=WIDTH, output_dir='OUTPUT/images') path = gen.bar_chart(categories=[...], series={...}, title='...', filename='chart_name.png')
3. Preview chart image with user, iterate if needed
4. Final PNG ready at images/{chart_name}.png

1.5b: Header Banner (if background image provided)

1. Execute code-blocks/header-generator.py: gen = HeaderGenerator() path = gen.generate(title='...', subtitle='...', output_path='OUTPUT/images/header_banner.jpg', bg_image='path/to/bg.png') # or bg_color='#2563eb' for solid
2. If no background image but user wants a visual banner → offer solid-color option
3. If user doesn't want a banner image → use the HTML header component instead (no Pillow needed)
4. Output: images/header_banner.jpg

1.5c: Image Optimization (if local images provided)

1. Ensure images are in the project's images/ directory
2. Check filenames — non-ASCII names must be renamed to ASCII (see rules/design-system.md § Image filename rules)
3. Execute code-blocks/image-optimizer.py: results = optimize_directory('OUTPUT/images/', threshold_kb=200) # results = [(filename, original_kb, compressed_kb),...]
4. Report savings to user if any files were optimized

Step 2: Generate HTML

Before generating HTML, read these files for guidance:

• rules/outlook-compatibility.md — the Outlook compatibility rules (essential)
• rules/design-system.md — universal design foundation (colors, typography, spacing) (this is what makes emails look professional and modern, not just compatible)
• rules/design-system-data-report.md — ONLY for data-heavy emails (KPI dashboards, weekly reports, status updates). Skip this for simple newsletters/announcements.
• rules/chart-design-system.md — ONLY when generating charts. Color system, typography, sizing, and data label conventions for email-embedded charts.
• rules/email-best-practices.md — design guidelines
• rules/style-presets.md — design style presets (Corporate, Editorial, Minimal). Choose a style matching the email's purpose to guide spacing, font sizes, and color usage.
• rules/placeholder-i18n.md — use placeholders matching the user's language
• templates/components/*.html — proven Outlook-safe component patterns (includes stats-grid, nav-bar, status-badge for advanced layouts)
• examples/example-*.html — complete working examples for reference

Design tip — keep headers compact: The header should feel like a navigation bar, not a hero banner. Logo + title on one line, total height around 60-70px. Avoid tall spacer rows and oversized logos in the header — save vertical space for content.

Why Outlook compatibility matters: Outlook 2007-2019 on Windows uses Microsoft Word's rendering engine instead of a browser engine. This means most modern CSS (flexbox, grid, border-radius, margin) is ignored or broken. The rules file contains battle-tested patterns from a production email system that handles this correctly. Every HTML pattern in templates/components/ has been verified to render correctly across Outlook, Gmail, Apple Mail, and web clients.

Key principles when generating:

• NEVER insert content the user did not request — no "generated by AI" watermarks, no tool credits, no disclaimers. The footer should only include what the user's design calls for (copyright, unsubscribe link, etc.).
• Follow rules/outlook-compatibility.md strictly — it covers table layout, inline styles, VML dividers, spacer rows, MSO conditionals, and all other Outlook quirks. Read that file before generating any HTML.
• Gmail size limit — keep HTML under 102KB or Gmail clips the email.
• Preheader text — add a hidden <div style="display:none;">Preview text</div> right after <body>.

After generating:

1. Validate first: execute code-blocks/html-validator.py → validate(html). Checks 32 rules including: forbidden CSS, missing VML, column width overflow, text overflow, Gmail 102KB size limit, missing alt text, non-HTTPS links, missing preheader, elements exceeding container width. Fix any errors before proceeding.
2. Organize output: execute code-blocks/output-manager.py → create_project(name) to create a timestamped directory in the user's working directory (e.g., ./output/2026-03-15_product-report/). Never write output files into the skill's own installation directory.
3. Save HTML to {project_dir}/newsletter-preview.html

Step 3: Preview & Adjust

1. Auto-open in browser: execute code-blocks/preview-helper.py → open_in_browser()

- All images render directly (charts, header banner, content images) - No need to generate EML for preview

1. Show ASCII layout summary in conversation: → ascii_layout_summary(html)
2. Ask if adjustments needed, iterate until satisfied

Step 4: Fill Content (Optional)

Ask: "Want to fill in content now, or leave placeholders for editing in Outlook?"

If filling now:

1. Execute code-blocks/content-filler.py → generate_fill_template(html) to show all placeholders as a YAML-like template the user can fill at once
2. Collect content — either one at a time OR as a batch dict from the user
3. Execute fill_batch(html, content_dict, image_dir) for efficient batch filling (auto-maps images from a directory to CID placeholders)
4. Re-preview and check unfilled_placeholders(html) for any remaining

Quick Edits (anytime after Step 2)

If the user asks for a targeted change (e.g., "change the header color to red", "make the title bigger"), use code-blocks/html-patcher.py instead of regenerating:

• replace_color(html, old, new) — swap a color everywhere
• change_width(html, old_w, new_w) — resize the container
• replace_text(html, old, new) — change visible text
• add_section(html, section_html) — insert before footer
• patch_file(path, colors={...}, texts={...}) — apply multiple patches at once

Re-validate after patching: html-validator.py → validate(patched_html).

Step 5: Generate EML

This is where the HTML becomes a real email file. Two options:

Option A (recommended): Use eml-builder.py fluent API

Write a short Python script that uses the EMLBuilder class:

from pathlib import Path
# Read the eml-builder.py source, then use:
eml = (
    EMLBuilder(sender="", subject="Newsletter Title")
    .set_html(Path("output/newsletter-preview.html").read_text())
    .add_image("logo", "output/images/logo.png")   # for each CID image
    .build("output/newsletter.eml")
)

Option B: Use html-to-eml.py script template

Copy the script, edit the CONFIG section at the top, then run it:

HTML_FILE = "output/newsletter-preview.html"   # your generated HTML
OUTPUT_EML = "output/newsletter.eml"           # output path
SUBJECT = "..."                                # ask user
SENDER = ""                                    # optional
TO_ADDRS = []                                  # optional
IMAGE_DIR = "output/images"                    # if images exist

Both approaches use Python's built-in email module — no pip install needed. They create a proper MIME structure (multipart/alternative → text/plain + multipart/related → text/html + CID images) with X-Unsent: 1 so Outlook opens it in draft/compose mode.

CRITICAL — Use the provided scripts only. You MUST use eml-builder.py (Option A) or html-to-eml.py (Option B) to generate EML. DO NOT write your own temporary/inline Python script for EML generation — it will likely produce incorrect MIME structure or miss CID image embedding, resulting in broken images in Outlook.

After EML is generated, validate before proceeding:

1. Execute code-blocks/eml-validator.py → validate_eml(eml_path). Checks 14 rules including: MIME structure, text/plain fallback, CID forward completeness (every src="cid:xxx" has a matching embedded image), residual unresolved image paths, Content-ID format (RFC 2392), X-Unsent header, Content-Disposition inline, and file size. Checks are adaptive — image-related rules only fire when the HTML actually references images; pure-text emails pass cleanly.
2. Fix any errors before proceeding to Step 6.

After validation passes, you MUST proceed to Step 6 — do not end the conversation here.

Step 6: Wrap Up

Always perform all of the following, in order:

6a. Offer to Crystallize for Repeated Use

This is important — many users will want to reuse this template with different data in the future (e.g., weekly reports, periodic newsletters). Always ask:

"邮件设计完成。是否要将此模板沉淀为可复用项目？以后只需填 Excel 即可重复生成。"

If the user agrees, read rules/production-mode.md § "Crystallization Process" and follow steps C0-C5. After crystallization completes, continue with 6b and 6c below. If the user declines, continue with 6b and 6c below.

6b. Save Template and Show Output

1. Offer to save the template: execute code-blocks/template-manager.py → save_template()
2. Show the output file locations using absolute paths (not just filenames) so the user can easily locate and open the files. For example: EML 文件: /Users/name/project/output/2026-03-15_report/newsletter.eml HTML 预览: /Users/name/project/output/2026-03-15_report/newsletter-preview.html 图片目录: /Users/name/project/output/2026-03-15_report/images/ Use the actual working directory path, not relative paths like output/....

6c. Show Usage Guide

Display the appropriate Outlook usage guide:

• Chinese user → templates/guides/outlook-usage-guide-zh.md
• English user → templates/guides/outlook-usage-guide-en.md

Placeholder Systems

This skill uses two different placeholder mechanisms for different purposes:

• {{placeholder}} (Design Mode) — used by code-blocks/content-filler.py for one-time placeholder filling during interactive design. Simple string replacement. Use this when building emails from scratch in Design Mode.
• <!-- SECTION/FIELD --> comments (Production Mode) — used in crystallized template.html files as structural markers that the AI agent reads to understand where to insert Excel data. These are NOT processed by content-filler.py. See rules/production-mode.md for details.

The two systems serve different workflows and do not interact. content-filler.py is a Design Mode tool; Production Mode uses the AI agent as the rendering engine.

File Map

rules/
  outlook-compatibility.md   ← Read before generating HTML (Outlook rendering rules)
  design-system.md           ← Universal: colors, typography, spacing (ALWAYS read)
  design-system-data-report.md ← Extension: KPI cards, status badges, trends (data emails only)
  email-best-practices.md    ← Design guidance (widths, colors, typography)
  style-presets.md           ← 3 design styles: Corporate, Editorial, Minimal
  placeholder-i18n.md        ← Localized placeholder text (zh/en/ja)
  chart-design-system.md     ← Chart colors, typography, sizing (read before chart generation)
  brand-color-extraction.md  ← Color extraction + preset palettes
  production-mode.md         ← Production mode: detection, workflow, crystallization, Excel integration

templates/
  components/*.html          ← 19 Outlook-safe HTML building blocks
                                (header, section, card, table, image-placeholder,
                                 divider, footer, stats-grid, nav-bar, status-badge,
                                 progress-bar, button, callout, testimonial,
                                 feature-list, pricing-table, team-member,
                                 alert, timeline)
  layouts/*.md               ← 7 preset layout descriptions
                                (single-column, two-column, magazine, announcement,
                                 dashboard, transactional, onboarding)
  guides/*.md                ← End-user Outlook usage guides (zh/en)

code-blocks/
  html-validator.py          ← Run AFTER generating HTML, BEFORE EML (auto-check 32 rules)
  eml-validator.py           ← Run AFTER generating EML, BEFORE Step 6 (auto-check 14 rules)
  html-patcher.py            ← Targeted edits (color, width, text) without regenerating
  output-manager.py          ← Timestamped project directories for organized output
  eml-builder.py             ← EML builder class (fluent API)
  cid-embedder.py            ← Image scanning + placeholder PNG creation
  html-to-eml.py             ← HTML→EML conversion (edit CONFIG section, then execute)
  eml-to-html.py             ← Extract HTML + images from .eml files (Import Mode)
  content-filler.py          ← {{placeholder}} replacement + batch filling
  template-manager.py        ← Save/load/list custom templates
  preview-helper.py          ← Browser auto-open + ASCII layout
  deps-checker.py            ← Auto-install optional dependencies (charts, images)
  chart-generator.py         ← Plotly chart generation (bar, line, heatmap, pie)
  header-generator.py        ← Header banner image compositing (text on background)
  image-optimizer.py         ← Image compression (PNG→JPEG, resize, optimize)
  excel-template-generator.py ← Generate & load Excel data templates (openpyxl)

examples/
  example-single-column.html ← Complete 600px single-column reference
  example-two-column.html    ← Complete 800px two-column reference