Skip to main content
The context file is the single source of truth for all GTM skills. It captures everything about your company, ICP, voice, win cases, and campaign learnings in one Markdown file.

​
File Location

claude-code-gtm/context/{company}_context.md
One file per company. All skills reference this path. Not per-campaign—global and continuously updated.
Example: If your company is Extruct, the file is claude-code-gtm/context/extruct_context.md

​
File Structure

The context file has 8 main sections:
Product description, value proposition, key numbers, and terminology.Fields:
  • Product: One-liner description
  • Value prop: Core value proposition (internal version)
  • Email-safe value prop: Outreach-friendly version without banned words
  • Key lingo: Internal terms and definitions
  • Key numbers: Quantifiable claims (database size, speed, coverage)
Usage:
  • Read by email-prompt-building for P2 value angles
  • Email-safe version gets baked into prompt templates
  • Key numbers used in emails for credibility
Sender identity, tone rules, and hard constraints for all outreach.Fields:
  • Sender: Name and company (who emails come from)
  • Tone: 1-sentence description (e.g., “Calm, analytical, builder-to-builder”)
  • Language level: Complexity guide (e.g., “B2 English: simple, clear sentences”)
  • Hard constraints: Numbered list of rules (e.g., “No dashes, no exclamation marks”)
  • Banned words: Words that must never appear in outreach
  • Scope boundaries: What the product IS and ISN’T
Usage:
  • Read by email-prompt-building to set voice rules in templates
  • Read by email-response-simulation to constrain rewrites
  • Enforced during email-generation
Target customer profiles and anti-patterns.Structure:
### Primary profiles
| Profile | Company size | Roles | Geographies | Why they buy |

### Anti-patterns (who is NOT a fit)
- [Description of lookalikes that aren't real fits]
Usage:
  • Read by hypothesis-building to shape pain hypotheses
  • Read by list-building for query design and filters
  • Read by market-research to scope research queries
Past customers, what resonated, concrete outcomes.Structure:
| Customer | Profile | What worked | Result | Date |

### Quotes / signals from wins
- "[Direct quote]" — [context]
Usage:
  • Read by hypothesis-building to extract transferable patterns
  • Read by list-building for seed companies in lookalike mode
  • Source for Proof Library entries (every proof must trace to a win)
Pre-written proof point sentences for use in P4 of emails.Structure:
| Proof point | Best for audience | Best for hypothesis | Source win case |
| "PS. [Full sentence ready to paste]" | [audience] | [hypothesis] | [win case] |
Rules:
  • Every proof point must trace back to a real win case
  • Write as full sentence including “PS.”
  • Map to audience type and hypothesis for conditional selection
Usage:
  • Read by email-prompt-building to select proof points
  • Conditionally inserted in emails during email-generation
  • 3-dimensional selection: peer relevance + hypothesis alignment + non-redundancy
Past campaigns with metrics and learnings.Structure:
| Campaign | Vertical | List size | Reply rate | Top hypothesis | Key learning | Date |
Usage:
  • Updated via context-building feedback loop mode
  • Informs future hypothesis prioritization
  • Tracks which verticals and angles work
Pain hypotheses organized by validation status.Structure:
### Validated (reply rate > X%)
1. **[Name]** — [description with data]. Best fit: [company type]

### Testing  
1. **[Name]** — [description]. Best fit: [company type]

### Retired
1. **[Name]** — retired because [reason]. Last tested: [date]
Usage:
  • Read by email-prompt-building for P1 angle selection
  • Updated based on campaign results (promote/demote/retire)
  • Target 5-7 active hypotheses
Domains to exclude from all lists.Structure:
| Domain | Reason | Added |
| example.com | competitor | 2026-03-01 |
Reasons:
  • Competitor
  • Partner
  • Existing customer
  • Requested opt-out
Usage:
  • Read by list-building to filter search results
  • Read by campaign-sending before upload to sequencer
  • Can auto-populate by running competitor search

​
Creating a Context File

Use the context-building skill: 
Build a company context file for www.example.com
The skill will:
  1. Check if context file exists
  2. If not, read your website to extract product info
  3. Walk you through each section interactively
  4. Write the file using the standard schema
You can also provide a win case to seed ICP and proof points:
Build context for www.example.com.
One customer is www.acme.com—they use us for supplier scoring.

​
Updating the Context File

The context file evolves as you learn. Four update modes:

​
Mode 1: Manual Update

Add new sections or entries: 
Update context file:
- Add a new win case: BigCorp, enterprise logistics, 40% faster routing
- Add to DNC list: competitor1.com, competitor2.com

​
Mode 2: Call Recording Capture

Extract signals from sales calls: 
Update context from this call transcript [attach file]
Extracted signals:
  • ICP signals (role, company size, pain points)
  • Win case data (what resonated)
  • Proof point candidates (specific results)
  • Hypothesis validation (confirmed/refuted)
  • Voice feedback (reaction to tone/positioning)

​
Mode 3: Campaign Feedback Loop

Import campaign results: 
Update context with campaign results from Instantly [attach export]
Updates:
  • Adds row to Campaign History table
  • Promotes/demotes hypotheses based on reply rates
  • Adds new proof points from positive replies
  • Updates Proof Library with performance notes

​
Mode 4: Hypothesis Refinement

Move hypotheses between Validated/Testing/Retired based on data: 
Move hypothesis #3 to Validated—12% reply rate in last campaign
Never overwrite existing entries. Always append new rows to tables, new bullets to lists. The context file is append-only.

​
Section Rules and Best Practices

​
What We Do

  • Keep under 100 words
  • Update when positioning changes
  • Always maintain email-safe version (no banned words)
  • Include key numbers for credibility

​
Voice

  • Define sender identity clearly (name + company)
  • Keep tone to 1 sentence
  • Hard constraints should be enforceable (“no dashes” not “be authentic”)
  • Update banned words when simulation flags issues

​
ICP

  • Max 5 primary profiles
  • Always include anti-patterns (prevents wasted outreach)
  • Be specific on company size ranges (“50-200” not “mid-market”)

​
Win Cases

  • Add every closed deal (anonymous is fine)
  • Include concrete outcomes with metrics
  • Note what pain triggered the purchase

​
Proof Library

  • Every proof point must map to a real win case
  • Write full PS sentences (“PS. Acme reduced…” not “reduced”)
  • Map to audience and hypothesis for conditional selection
  • Test proof points in campaigns before adding

​
Campaign History

  • One row per campaign
  • Update reply rate when final numbers are in
  • Extract 1-sentence learning per campaign

​
Active Hypotheses

  • Target 5-7 active hypotheses (3-4 Validated, 2-3 Testing)
  • Move to Retired after 2-3 failed campaigns
  • Promote to Validated after 10%+ reply rate

​
Do Not Contact

  • Check before every list build
  • Include competitors, partners, existing customers
  • Consider running Extruct search for competitors to auto-populate

​
Cross-Skill References

The context file is consumed by:

hypothesis-building

Reads: ICP, Win Cases, product value propGenerates pain hypotheses from patterns

list-building

Reads: ICP, Win Cases, DNC listUses for query design and seed companies

market-research

Reads: ICP, hypothesesScopes research queries

enrichment-design

Reads: HypothesesDesigns segmentation columns

list-segmentation

Reads: HypothesesMatches companies to hypotheses for tiering

email-prompt-building

Reads: Voice, What We Do, Proof Library, HypothesesBuilds self-contained prompt templates

email-response-simulation

Reads: Voice rulesConstrains rewrites

campaign-sending

Reads: DNC listFilters before upload

​
Full Schema Template

See the complete template in the source: 
~/.claude/skills/gtm-skills/skills/context-building/references/context-schema.md
Or view in the context-building skill reference.

​
Example: Extruct Context File

# Company Context

## What We Do

**Product:** API-first company search and lookalikes engine for GTM teams.

**Value prop:** Extruct finds companies that match your ICP, even the ones invisible to standard databases. We combine deep web research with AI-powered similarity to surface prospects others miss.

**Email-safe value prop:** Extruct is a company search API that finds prospects matching your best customers, including companies invisible to standard B2B databases.

**Key lingo:**
- **Lookalike Search**: Find companies similar to a seed company using AI embeddings
- **Deep Search**: Discovery mode with criteria-based auto-grading
- **AI Tables**: Research any data point per company with agent columns

**Key numbers:** 
- 50M+ companies indexed globally
- 80-90% of lookalike results not in Apollo/ZoomInfo
- Research agents run in 2-5 minutes per company

---

## Voice

**Sender:** Danny from Extruct

**Tone:** Calm, analytical, builder-to-builder. No hype.

**Language level:** B2 English: simple, clear sentences. Polite but not over-polite.

**Hard constraints:**
1. No dashes, no exclamation marks, no emojis
2. No buzzwords: "AI-powered" (unless describing a specific feature), "revolutionary", "game-changer"
3. Sentence case only (no title case in subject lines)
4. No questions in P1 (state observations only)
5. No flattery ("I was impressed by...")

**Banned words:** agents (use "research" or "API"), try (use "test" or "run"), just (filler word)

**Scope boundaries:** Extruct is company-level intelligence, not people data or panel surveys. We find and research companies, not contacts.

---

## ICP

### Primary profiles

| Profile | Company size | Roles | Geographies | Why they buy |
|---------|-------------|-------|-------------|-------------|
| Sales tools / CRMs | 20-200 | Founder, Head of Growth | North America, Europe | Need broader company coverage than Apollo/ZoomInfo |
| Vertical SaaS platforms | 50-500 | VP Product, Head of Data | North America | Want to enrich their user base with company intelligence |
| GTM agencies | 10-100 | Founder, Partner | North America | Build prospect lists for clients across verticals |

### Anti-patterns (who is NOT a fit)

- Enterprise sales teams (not technical enough to use API)
- Recruiters (we don't have people data)
- Individual SDRs (need managed service, not API)

...

​
Next Steps

context-building Skill

Learn how to create and update context files

Campaign Artifacts

See where context files fit in directory structure

End-to-End Workflow

See how context flows through a campaign

hypothesis-building

Generate pain hypotheses from context

Build docs developers (and LLMs) love

Get started for freeTalk to us