Skip to main content

​
Overview

Generate cold outreach emails from a contact CSV + prompt template. The prompt template is self-contained — it has all voice, research, value prop, proof points, and personalization rules baked in. This skill just runs it per row.

​
When to Use

Trigger this skill when you need to:
  • Generate personalized cold emails at scale
  • Run an email campaign from a contact list
  • Create outreach emails from a prompt template
  • Execute email generation pipeline
Trigger phrases: “generate emails”, “email generation”, “run emails”, “create emails”, “email pipeline”, “generate outreach”, “write emails for campaign”

​
Architectural Principle

This skill is a runner, not a reasoner. All strategic reasoning (voice, value angles, proof points, research data) was done by the email-prompt-building skill at prompt-build time and embedded in the prompt template. This skill reads the prompt + CSV and generates emails. It does NOT read the context file, hypothesis set, or research files.
prompt template (.md) ─┐
                       ├──▶ generate email per row ──▶ emails CSV
contact CSV ───────────┘

​
Inputs Required

​
contact_csv
string
required
CSV file with recipient data + enrichment columnsLocation: claude-code-gtm/csv/input/{campaign}/contacts.csv
​
prompt_template
string
required
Markdown prompt file from email-prompt-building skillLocation: claude-code-gtm/prompts/{vertical-slug}/en_first_email.md
That’s it. No context file, no hypothesis set, no research files.

​
Contact CSV Columns

The prompt template specifies which columns it needs. Check the prompt’s “Enrichment data fields” section for the expected column names.

​
Required (Always)

  • first_name
  • last_name
  • company_name
  • job_title

​
Enrichment (Campaign-Specific)

Listed in the prompt template. If the prompt references a field that’s not in the CSV, the email quality degrades.
Check column alignment before running.

​
Running the Generator

1

Option A: In-Chat Generation (<30 contacts)

For small lists:
  1. Read the prompt template
  2. Read the contact CSV
  3. For each row, apply the prompt with the row’s data and generate the email
  4. Output as JSON per row, accumulate results
  5. Save to output CSV
2

Option B: Batch Generation (30+ contacts)

For larger lists, process in batches of 10-20 rows within the conversation:
  1. Load the prompt template and contact CSV
  2. Process contacts in batches
  3. For each row, apply the prompt and generate the email JSON
  4. Accumulate results and save to output CSV
Output path: claude-code-gtm/csv/output/{campaign-slug}/emails.csv

​
Quality Checks

After generating, verify:
  • Every email is within the word limit specified in the prompt
  • No banned phrases from the prompt template appear
  • Enrichment data was actually used — not just generic text
  • Example queries in P2 are specific to each recipient’s verticals
  • Proof points vary across emails (not the same PS for everyone)
  • Subject lines meet the prompt’s length constraints

​
Segmentation-Aware Generation

When the contact CSV includes segmentation data (from list-segmentation):
  • Generate individually with full attention to enrichment data
  • Route through email-response-simulation for review before sending
  • Group by hypothesis_number
  • Generate in batches within each hypothesis group
  • Spot-check 2-3 from each group
  • Do not generate emails
  • Route back to list-enrichment or list-building

​
In-Chat Refinement Loop

After generating, you can refine:
1

Identify Issues

You identify emails you don’t like and say what to change
2

Update the Prompt Template

Important: Update the prompt template (not just the individual email) — the fix should be systemic
3

Re-run the Generator

Re-run with updated prompt
4

Iterate

Repeat until satisfied
Changes made to the prompt are tracked so you can see the evolution.

​
Building a New Prompt Template

If no prompt template exists for this campaign, use the email-prompt-building skill to build one. That skill reads the context file and research, then synthesizes a self-contained prompt. Do not build prompts ad hoc in this skill.

​
Output Format

Generated emails are saved to: 
claude-code-gtm/csv/output/{campaign-slug}/emails.csv
Columns: 
recipient_name,recipient_company,recipient_email,subject,greeting,p1,p2,p3,p4,hypothesis,tier
JSON format per email: 
{
  "recipient_name": "Jane Doe",
  "recipient_company": "Acme Corp",
  "recipient_email": "[email protected]",
  "subject": "Acme's supplier data gap",
  "greeting": "Hey Jane,",
  "p1": "First paragraph text...",
  "p2": "Second paragraph text...",
  "p3": "Third paragraph text...",
  "p4": "Fourth paragraph text...",
  "hypothesis": "Database blind spot",
  "tier": "1"
}

​
Example Workflow

# 1. Load prompt template
claude-code-gtm/prompts/pe-rollup-blue-collar/en_first_email.md

# 2. Load contact CSV
claude-code-gtm/csv/input/pe-rollup-blue-collar/contacts.csv

# 3. Generate emails (batch mode for 100+ contacts)
# Process in batches of 20
# Apply prompt template to each row
# Validate against quality checks

# 4. Save output
claude-code-gtm/csv/output/pe-rollup-blue-collar/emails.csv

​
Quality Check Example

Good:
Hey Jane, I noticed Acme’s platform connects 50K+ HVAC contractors…
Uses actual company name and enrichment data (platform_type, contractor_count).Bad:
Hey Jane, I noticed your platform connects contractors…
Generic, could apply to anyone.
Good:
“HVAC companies in Texas under 50 employees”
References prospect’s actual vertical (HVAC contractors).Bad:
“suppliers in your industry”
Generic, not tailored.
Good:
FieldNation uses our data to keep 20K+ contractors verified across 40 states.
Peer company (similar platform type and size).Bad:
A small startup uses our data…
Not a peer (smaller company cited to larger prospect).

​
Troubleshooting

Emails too generic?
  • Check that enrichment columns are populated in the contact CSV
  • Verify the prompt template references enrichment fields by name
  • Re-run email-prompt-building with more specific research data
Banned phrases appearing?
  • Update the prompt template’s “Banned phrasing” section
  • Re-run generation after updating prompt
Word count violations?
  • Check the structural variant selection logic in the prompt
  • Verify the prompt specifies word limits per variant
  • Adjust the prompt’s P1-P4 length constraints
Proof points repeating?
  • Add non-redundancy rules to the prompt template
  • Specify multiple proof points with conditions for when to use each

​
Next Steps

After email generation completes:
  1. Tier 1 emails → Proceed to email-response-simulation for review
  2. Tier 2 emails → Spot-check 5-10 and proceed to campaign-sending
  3. All emails → Review sample and refine prompt if needed

Build docs developers (and LLMs) love

Get started for freeTalk to us