Skip to main content
GTM Skills automatically create a claude-code-gtm/ directory to store all campaign artifacts. This page explains the structure, what each file contains, and when it’s created.

​
Directory Structure

After running a full campaign, you’ll have: 
claude-code-gtm/
├── context/
│   ├── {company}_context.md          ← Global context file
│   └── {vertical-slug}/              ← Per-vertical research
│       ├── hypothesis_set.md
│       └── sourcing_research.md      (optional, from market-research)
├── prompts/
│   └── {vertical-slug}/              ← Email prompt templates
│       ├── en_first_email.md
│       └── en_follow_up_email.md     (optional)
└── csv/
    ├── input/{campaign-slug}/        ← Segmented lists, people, contacts
    │   ├── companies.csv
    │   ├── companies_segmented.csv
    │   ├── people.csv
    │   └── contacts.csv
    └── output/{campaign-slug}/       ← Generated emails
        └── emails.csv

​
Top-Level Directories

​
context/

Global company knowledge and per-vertical research. Root level:
  • {company}_context.md — The single source of truth for all skills
    • Created by: context-building
    • Updated by: context-building (all modes)
    • Read by: All skills
    • Lifecycle: Permanent (one per company, continuously updated)
Per-vertical subdirectories:
  • {vertical-slug}/hypothesis_set.md — Pain hypotheses for this vertical
    • Created by: hypothesis-building OR market-research
    • Read by: list-building, enrichment-design, list-segmentation, email-prompt-building
    • Lifecycle: Reusable (update when entering the vertical again)
  • {vertical-slug}/sourcing_research.md — Deep research findings (optional)
    • Created by: market-research
    • Read by: email-prompt-building
    • Lifecycle: Reusable (reference for future campaigns in same vertical)
Vertical slug examples:
  • enterprise-saas
  • logistics-platforms
  • fintech-b2b
Use lowercase with hyphens, no spaces.

​
prompts/

Self-contained email prompt templates. Structure: 
prompts/
└── {vertical-slug}/
    ├── en_first_email.md
    └── en_follow_up_email.md
  • Created by: email-prompt-building
  • Read by: email-generation
  • Lifecycle: Per-campaign (update if voice/research changes)
What’s inside:
  • Voice rules (from context file)
  • Hypothesis-based P1 angles (from research)
  • P2 value angles (from context → What We Do)
  • P4 proof points (from context → Proof Library)
  • Structural variants (by role/seniority)
  • Banned phrasing
Prompt templates are self-contained. Email generation only reads this file + contact CSV—it doesn’t access the context file or research.

​
csv/

Input lists and output emails. Structure: 
csv/
├── input/{campaign-slug}/
│   ├── companies.csv
│   ├── companies_segmented.csv
│   ├── people.csv
│   └── contacts.csv
└── output/{campaign-slug}/
    └── emails.csv
Campaign slug examples:
  • enterprise-saas-q1
  • logistics-platforms-2026-03
  • fintech-expansion
Use descriptive names with dates/quarters.

​
Input Files (csv/input/{campaign-slug}/)

​
companies.csv

Raw company list from list building. Created by: list-building
Columns:
  • name — Company name
  • domain — Website domain
  • short_description — 1-2 sentence description
  • founding_year — Year founded
  • employee_count — Headcount
  • hq_country — Headquarters country
  • hq_city — Headquarters city
  • relevance_score — 0-100 (from Extruct API)
When created:
After running Search, Discovery, or Lookalike via list-building. This is the starting list before enrichment.

​
companies_segmented.csv

Companies with tier assignments and enrichment data. Created by: list-segmentation
Columns:
  • All columns from companies.csv
  • Enrichment columns (added by list-enrichment)
    • Custom research fields (e.g., data_infrastructure_maturity, recent_product_launch)
  • tier — 1, 2, or 3 (fit score)
  • hypothesis_number — Primary hypothesis match (e.g., #1, #2)
  • hypothesis_name — Hypothesis short name
  • segmentation_reasoning — Why this tier was assigned
When created:
After enrichment is complete and tiering logic is applied. Usage:
  • Filter to Tier 1 for high-touch emails
  • Filter to Tier 2 for hypothesis-based templates
  • Exclude Tier 3 or route back to re-enrichment

​
people.csv

Decision makers at target companies. Created by: people-search
Columns:
  • company_domain — Matches domain from companies.csv
  • full_name — Person’s name
  • first_name — Extracted first name
  • last_name — Extracted last name
  • job_title — Current role
  • linkedin_url — LinkedIn profile
  • seniority — Junior, Mid, Senior, Executive
When created:
After running LinkedIn search via Extruct API.

​
contacts.csv

People with verified contact information. Created by: email-search
Columns:
  • All columns from people.csv
  • All enrichment columns from companies_segmented.csv (joined on domain)
  • email — Verified email
  • phone — Phone number (if found)
  • email_status — verified, risky, invalid
  • confidence_score — 0-100 (from contact provider)
When created:
After contact enrichment via Prospeo, FullEnrich, or similar. This is the input to email generation.

​
Output Files (csv/output/{campaign-slug}/)

​
emails.csv

Generated emails ready for sending. Created by: email-generation
Columns:
  • All columns from contacts.csv (for reference)
  • subject_line — Email subject
  • email_body — Full email text
  • p1 — Paragraph 1 (pain angle)
  • p2 — Paragraph 2 (value prop)
  • p3 — Paragraph 3 (CTA)
  • p4 — Paragraph 4 (proof point, optional)
  • hypothesis_used — Which hypothesis was applied
  • structural_variant — A, B, C, or D (role-based structure)
  • word_count — Total words
  • flagged_issues — Any voice violations detected
When created:
After running the prompt template against contacts.csv. Usage:
  • Import to email sequencer (Instantly, Lemlist, etc.)
  • Spot-check for quality before sending
  • Route Tier 1 emails through email-response-simulation first

​
File Lifecycle

1

Foundation (Once per Company)

Created:
  • context/{company}_context.md
By: context-buildingLifecycle: Permanent, continuously updated
2

Research (Once per Vertical)

Created:
  • context/{vertical-slug}/hypothesis_set.md
  • context/{vertical-slug}/sourcing_research.md (optional)
By: hypothesis-building or market-researchLifecycle: Reusable across campaigns in same vertical
3

List Building (Per Campaign)

Created:
  • csv/input/{campaign-slug}/companies.csv
By: list-buildingLifecycle: Campaign-specific
4

Enrichment (Per Campaign)

Created:
  • Enrichment columns added to Extruct table
  • Exported as updated companies.csv or merged into segmented CSV
By: list-enrichmentLifecycle: Campaign-specific
5

Segmentation (Per Campaign)

Created:
  • csv/input/{campaign-slug}/companies_segmented.csv
By: list-segmentationLifecycle: Campaign-specific
6

People & Contact Search (Per Campaign)

Created:
  • csv/input/{campaign-slug}/people.csv
  • csv/input/{campaign-slug}/contacts.csv
By: people-search → email-searchLifecycle: Campaign-specific
7

Email Prompt (Per Vertical or Campaign)

Created:
  • prompts/{vertical-slug}/en_first_email.md
By: email-prompt-buildingLifecycle: Reusable if voice/research unchanged, otherwise update per campaign
8

Email Generation (Per Campaign)

Created:
  • csv/output/{campaign-slug}/emails.csv
By: email-generationLifecycle: Campaign-specific, final output

​
Naming Conventions

​
Company Name

Use the company’s primary domain without TLD: 
extruct_context.md
salesforce_context.md  
acme_context.md

​
Vertical Slug

Lowercase, hyphen-separated, descriptive: 
enterprise-saas
logistics-platforms
fintech-b2b-payments  
marketplace-platforms

​
Campaign Slug

Descriptive + date or identifier: 
enterprise-saas-q1-2026
logistics-expansion-march
fintech-tier1-test

​
File Retention

Keep forever:
  • context/{company}_context.md — continuously updated
  • context/{vertical-slug}/ — reusable research
Keep per campaign:
  • csv/input/{campaign-slug}/ — for reference and re-runs
  • csv/output/{campaign-slug}/ — for results tracking
Update as needed:
  • prompts/{vertical-slug}/ — update when voice or research changes
After a campaign completes, import results back to the context file using context-building feedback loop mode. This closes the learning loop.

​
Example: Full Campaign File Tree

Running this prompt: 
I'm building www.extruct.ai.
Find 200 enterprise SaaS platforms similar to salesforce.com,
enrich with relevant data, and generate emails to VPs of Product.
Creates: 
claude-code-gtm/
├── context/
│   ├── extruct_context.md
│   └── enterprise-saas/
│       └── hypothesis_set.md
├── prompts/
│   └── enterprise-saas/
│       └── en_first_email.md
└── csv/
    ├── input/enterprise-saas-q1/
    │   ├── companies.csv                    (200 rows)
    │   ├── companies_segmented.csv          (200 rows + tiers)
    │   ├── people.csv                       (~150 VPs found)
    │   └── contacts.csv                     (~120 with emails)
    └── output/enterprise-saas-q1/
        └── emails.csv                       (120 generated emails)

​
Extruct Table Integration

Some artifacts live in Extruct tables instead of local CSVs: What stays in Extruct:
  • Company lists uploaded by list-building
  • Enrichment columns added by list-enrichment
  • People search results from people-search
What’s exported to CSV:
  • Final segmented company list (companies_segmented.csv)
  • Contacts with emails (contacts.csv)
  • Generated emails (emails.csv)
Why hybrid?
  • Extruct tables are great for research and enrichment (parallel agents, web UI)
  • CSVs are great for email generation and sequencer imports (local processing)
You can access Extruct tables anytime at https://app.extruct.ai/tables/{table_id}

​
Next Steps

End-to-End Workflow

See how these files are created in a full campaign

Context Files

Deep dive on the global context file structure

list-building

Learn how companies.csv is created

email-generation

Learn how emails.csv is created

Build docs developers (and LLMs) love

Get started for freeTalk to us