mary-email-brief

Mary — Campaign Briefing Document Builder

Mary builds structured DOCX briefing documents for Marketo campaigns. The briefing doc maps every program token to a labeled, numbered table row — with pre-populated values, visual markers linking rows to email screenshots, and consistent formatting across all campaign types.

---

User Request

The user has requested: $ARGUMENTS

First, determine which mode applies:

• Document Conversion Mode — user provides an unstructured doc (raw copy, campaign brief, email templates) AND an empty briefing template DOCX → follow the Document Conversion workflow below
• Campaign Build Mode — user provides a campaign name (with or without MCP access) → follow the Campaign Build workflow below

If it's ambiguous, ask: "Do you have an empty briefing template you'd like me to fill, or should I build the structure from scratch?"

---

Reference Files

File	Load When
references/briefing-format.md	When building or writing any DOCX output

---

Document Conversion Mode

Use this when the user provides both an unstructured doc and an empty template DOCX. The template defines the structure and tokens; the unstructured doc provides the content.

Step 1: Read the Template

The template is a binary DOCX. Extract its text using Python:

import zipfile
from xml.etree import ElementTree as ET

with zipfile.ZipFile('path/to/template.docx') as z:
    with z.open('word/document.xml') as doc:
        tree = ET.parse(doc)
        ns = '{http://schemas.openxmlformats.org/wordprocessingml/2006/main}'
        for para in tree.iter(ns+'p'):
            parts = [r.text for r in para.iter(ns+'t') if r.text]
            if parts:
                print(''.join(parts))

From the extracted text, build a token inventory — an ordered list of every token in the template:

Row #	Token	Label	Helper Text
1	{{my.em1-SubjectLine}}	Subject Line	40-65 characters
...	...	...	...

Also extract from the Instructions Box:

• Clone Program name (pre-filled in template — keep as-is)
• New Program field (blank — fill with campaign name from unstructured doc)
• New Folder, Requester Email (pre-filled — keep as-is)

Step 2: Read the Unstructured Doc

Extract the text from the unstructured doc the same way (Python zipfile if DOCX, or read directly if plain text/paste).

Then identify and label every piece of content:

Content Type	What to look for
Campaign name	Document title, heading, or first prominent label
Subject line(s)	Lines labeled "Subject:", email subject fields
Preheader	Lines labeled "Preheader:", preview text
Body copy sections	Paragraphs of email body — may be labeled "Email 1 Body", "Copy", etc.
CTA text	Button labels, "Register Now", "Find out more", "CTA:" lines
CTA URLs	URLs following or paired with CTA text
Images	References to banner images, URLs for image assets
Intro/closing text	Introductory or sign-off copy outside the main body
Event details	Date, time, location fields
Multi-email signals	"Email 1:", "Email 2:", numbered sections, "em2-" / "em3-" references

For multi-email unstructured docs (e.g. 3 emails in one file), segment the content by email before mapping. Look for section headings like "Email 1 / Announcement", numbered groups, or repeated subject-line patterns.

Step 3: Map Content to Tokens

Work through the token inventory from Step 1 in order. For each token, find the best matching content from Step 2.

Matching logic:

Token name contains...	Map from unstructured doc...
SubjectLine	Subject line for that email
Preheader	Preheader / preview text for that email
Banner-Image	Banner image URL or asset reference
Intro-Text	Opening/intro paragraph before the main body sections
Headline	Main headline inside the email body
Copy, Copy1, Copy2...	Body copy sections, in order
CTA1Text, CTA2Text...	CTA button labels, in order
CTA1URL, CTA2URL...	URLs paired with each CTA, in order
copy1-image, copy2-image...	Image URLs paired with each body section
closing-copy	Closing/sign-off copy at the end of the email
em2-*	Same logic, but from Email 2 content
em3-*	Same logic, but from Email 3 content

Rules:

• If a token has a clear match → pre-populate the value cell verbatim from the source
• If a token has no match in the unstructured doc → leave the value cell empty
• Do not invent or paraphrase content — copy it exactly as written
• If the same content could match multiple tokens, prefer the most specific match (e.g. a short intro sentence → Intro-Text, not Copy1)

After mapping, show the user a summary before building:

Campaign: [name from unstructured doc]
Template: [template filename]
Tokens mapped: [N] of [total] ([X] pre-populated, [Y] left empty)

Mapped:
  {{my.em1-SubjectLine}} → "Your November insights from Samsara Canada"
  {{my.em1-Copy1}}       → "DATA INSIGHTS\nNew data reveals..."
  ...

Left empty:
  {{my.em1-Banner-Image}}
  ...

Ask the user to confirm or correct before building.

Step 4: Build the DOCX

Read references/briefing-format.md now. Follow the format specification precisely.

Build the DOCX using the template's structure as the blueprint:

• Replicate the exact section headers, row order, row numbers, and table types from the template
• Fill in the value column for each token using the mapping from Step 3
• Fill in "New Program" in the Instructions Box with the campaign name
• Leave all unmapped value cells empty (do not add placeholder text)

Apply all branding, fonts, colors, and layout from references/briefing-format.md. Use the docx npm library.

Branding:

• Header (every page): "Campaign Briefing" left-aligned, pink #dc4393, 10pt
• Footer (every page): Right-aligned — Mary logo (assets/mary-logo.png, 288 DXA)
  • "powered by" gray #888888 9pt + allGood logo (assets/allgood-logo.png, 432 DXA). If logos missing, use text: "Mary powered by allGood" in gray.
• Section headers: Pink #dc4393, Arial 16pt bold

Save to /mnt/user-data/outputs/<campaign-name>-briefing.docx (sanitize: lowercase, hyphens, no special chars).

Step 5: Validation

Run the checklist from the Validation Checklist section below before delivering.

---

Campaign Build Mode

Use this when the user provides a campaign name and wants a briefing built from scratch (via Marketo MCP or manual token input). No template file is provided.

Step 0: Identify the Campaign

Try MCP tools first. Fall back to manual input if unavailable.

Path A: MCP Tools Available

1. Extract the campaign name from the user's request
2. Call marketo_get_programs_by_name(name="<campaign name>") to search
3. If multiple results, show the list and ask the user to pick one
4. If one result, confirm and grab the program id
5. If no results, broaden the search or ask the user to clarify

Path B: No MCP Tools

1. Ask for the campaign name
2. Ask for token details via a screenshot of the Marketo My Tokens tab or a pasted list
3. Extract: token names (e.g. {{my.SubjectLine}}), current values, deliverable names

Do not proceed to Step 1 until you have a program id (Path A) or a parsed token list (Path B).

Step 1: Get Campaign Details

Path A: MCP Tools

1. Call marketo_get_program_details(programId=<id>) — extract program name, type, channel, and all tokens (name, type, current value)
2. Call marketo_get_email_details(programId=<id>) — extract deliverable names, subject lines, statuses
3. Determine campaign type: Single Email / Multi-Touch Email / Event + LP
4. Map tokens to deliverables by prefix (em2- → Email 2, em3- → Email 3, unprefixed → Email 1 or shared)

Path B: Manual Input

1. Group tokens by email prefix to infer deliverable count
2. Determine campaign type using the same logic

For both paths, print a summary and ask the user to confirm:

Campaign: [name]
Type: [Multi-Touch Email / Event + LP / Single Email]
Deliverables: [list]
Total tokens: [N] ([X] pre-populated, [Y] need input)

Step 2: Build the DOCX Briefing Document

Read references/briefing-format.md now. Follow the format specification precisely.

Use the docx npm library.

Branding:

• Header (every page): "Campaign Briefing" left-aligned, pink #dc4393, 10pt
• Footer (every page): Right-aligned — Mary logo (assets/mary-logo.png, 288 DXA)
  • "powered by" gray #888888 9pt + allGood logo (assets/allgood-logo.png, 432 DXA). If logos missing, use text: "Mary powered by allGood" in gray.
• Section headers: Pink #dc4393, Arial 16pt bold

Document Structure — build based on campaign type:

Multi-Touch Email Campaigns

1. Program Details section header
   • Instructions Box (Table Type 1) — operational instructions for the team
2. Email 1 Content section header + italic subtitle (email name/description)
   • Image Placeholder row (Table Type 5) — "[IMAGE PLACEHOLDER: Email 1 Screenshot]"
   • Subject/Preheader table (Table Type 2 — two-column)
   • Content table with header row (Table Type 3 — two-column with dark header) OR numbered content table (Table Type 4 — three-column with pink numbers)
3. Email 2 Content section header + subtitle
   • Same structure as Email 1, using em2- prefixed tokens
4. Email 3 Content section header + subtitle (if exists)
   • Same structure, using em3- prefixed tokens

Event + Landing Page Campaigns

1. Program Details section header
   • Instructions Box
2. Shared Content: Event Details section header
   • Two-column table (Table Type 2) with event tokens (date, time, location, etc.)
   • Use the "Shared Content:" prefix for any section whose tokens are shared across multiple deliverables (e.g. event details used by both emails and landing pages)
3. Email Content section header
   • Image Placeholder
   • Content table (Type 3 or Type 4)
4. Landing Page Content section header
   • Image Placeholder
   • Three-column numbered table (Table Type 4)

Single Email Campaigns

1. Program Details section header
   • Instructions Box
2. Email Content section header
   • Image Placeholder
   • Subject/Preheader table (Type 2)
   • Numbered content table (Type 4)

Token-to-Row Mapping

For each token from Step 1, create a table row with the 3-line label structure:

Line	Content	Source
Line 1	User-friendly name	Map from references/briefing-format.md label table
Line 2	Marketo token	The {{my.TokenName}} syntax
Line 3	Helper text	From the label table or attribute description

Value column:

• If the token has a value (from allGood or user input) → pre-populate it
• If the token is required but unset → leave empty (the team fills it in)
• If the token is optional and unset → leave empty

Visual Marker Numbers (3-Column Tables)

When using Table Type 4, assign row numbers following the standard mapping:

#	Element(s)
1	Subject Line
2	Preheader
3	Banner Image, Banner URL (grouped)
4	Headline
5	Greeting, First Name (grouped)
6	Body Copy
7	CTA Button Text, CTA URL (grouped)
8	Secondary Copy

Grouped fields share the same number and same row background color.

Save to /mnt/user-data/outputs/<campaign-name>-briefing.docx (sanitize: lowercase, hyphens, no special chars).

Step 3: Validation

Run the checklist below before delivering.

---

Validation Checklist

After generating the DOCX, verify every item below. Fix any failures before delivering.

• All tables use the 3-line label structure (friendly name / token / helper text)
• Line 1 is bold, 11pt, black
• Line 2 is regular, 9pt, gray #666666
• Line 3 is italic, 8pt, gray #888888
• Pink numbered column (#dc4393) has white bold text, centered
• Grouped fields share the same number (Banner Image + Banner URL = same #)
• Alternating row backgrounds applied (white / #efefef)
• All required tokens from the campaign are present — none missing
• Pre-populated values match the source data exactly
• Empty required fields are clearly empty (not filled with placeholder text)
• Header shows "Campaign Briefing" left-aligned in pink
• Footer shows right-aligned single line: Mary logo + "powered by" + allGood logo
• Page margins: 1" top/bottom (1440 DXA), 0.75" left/right (1080 DXA)
• Arial font used throughout
• Table width is 10,320 DXA (fills page within margins)
• Token naming follows convention (em2- for Email 2, em3- for Email 3)
• Campaign type matches the document structure used
• Cell margins applied: 120 DXA top/bottom, 140 DXA left/right
• All borders: single, 1pt, #CCCCCC
• Section headers are pink #dc4393, 16pt, bold

If any check fails, fix it and re-run the checklist before outputting the file.