Context Building

Company Context Builder

One global context file per company. Every other GTM skill reads from this file for voice, value prop, ICP, win cases, proof points, and campaign learnings.

Context File Location

claude-code-gtm/context/{company}_context.md

Single file per company, not per-campaign. All skills reference this path.

Modes

Mode 1: Create

Use when no context file exists yet. Walk the user through each section.

Step 1: Check if claude-code-gtm/context/{company}_context.md exists.

Step 2: If not, ask the user for each section (one at a time or in bulk):

Step 3: Write the file using the schema from references/context-schema.md.

Key sections to get right:

What We Do — must include:

• Product one-liner
• Core value prop (internal version, can use any language)
• Email-safe value prop (outreach-friendly version of the value prop)
• Key numbers (quantifiable claims — database size, speed benchmarks, coverage stats)
• Key lingo (internal terms and definitions)

Proof Library — must include:

• Full PS sentences ready to paste into emails
• Each mapped to: best audience, best hypothesis, source win case
• Every proof point must trace back to a real win case
• Write the sentence as it would appear in the email (including "PS.")

Mode 2: Update

Use when context file exists and user wants to add or modify a section.

Step 1: Read existing context file.

Step 2: Ask what to update. Common updates:

• Add a new win case
• Add a campaign result
• Update ICP based on new learnings
• Add domains to DNC
• Revise or add hypotheses
• Add or update proof points in the Proof Library
• Update voice rules
• Update key numbers (e.g., database size grew)

Step 3: Append to the relevant section. Never overwrite existing entries — add new rows to tables, new bullets to lists.

Mode 3: Call Recording Capture

Use when the user pastes a call transcript or meeting notes.

Step 1: Read the transcript.

Step 2: Extract and categorize signals:

• ICP signals — who was on the call, their role, company size, what they care about
• Win case data — what resonated, what they said about their current workflow, pain points confirmed
• Proof point candidates — specific results or quotes that could become Proof Library entries
• DNC signals — any companies or domains mentioned as off-limits
• Hypothesis validation — which existing hypotheses were confirmed or refuted
• Voice feedback — any reaction to tone, language, or positioning that should update Voice rules

Step 3: Present extracted signals to the user for confirmation.

Step 4: Update the context file with confirmed signals.

Mode 4: Feedback Loop

Use when importing campaign results from your email sequencer (e.g. Instantly) or manual tracking.

Step 1: Read campaign results (CSV, pasted data, or email sequencer export e.g. Instantly).

Step 2: Extract metrics:

• Campaign name, vertical, list size
• Open rate, reply rate, positive reply rate
• Top-performing hypotheses (which P1 angles got replies)
• Patterns in positive vs negative replies

Step 3: Add a new row to the ## Campaign History table.

Step 4: Update ## Active Hypotheses based on results:

• Promote hypotheses with high reply rates to Validated
• Demote hypotheses that didn't resonate to Retired
• Note any new hypotheses suggested by reply patterns

Step 5: Update ## Proof Library if campaign results surfaced new proof points:

• New win cases → write new PS sentences
• Existing proof points that didn't resonate → add notes or remove

Cross-Skill References

This context file is consumed by:

• hypothesis-building — reads ICP, Win Cases, and product value prop to generate pain hypotheses
• email-prompt-building — reads Voice, What We Do, Proof Library, and Active Hypotheses to build prompt templates
• email-generation — reads the prompt template (which was built from this file)
• list-building — reads ICP and Win Cases for seed companies
• market-research — reads ICP and hypotheses for research scope
• enrichment-design — reads hypotheses for segmentation column design
• list-segmentation — reads hypotheses for tiering logic
• email-response-simulation — reads Voice rules to constrain rewrites
• campaign-sending — reads DNC list for exclusions

Reference

See references/context-schema.md for the full file schema with all sections and field definitions.

Context File Schema

Template for claude-code-gtm/context/{company}_context.md. Copy and fill in.

# Company Context

## What We Do

**Product:** [One-liner description]

**Value prop:** [Core value proposition in 1-2 sentences]

**Email-safe value prop:** [Same value prop rewritten without any banned words from voice rules. This version gets baked into prompt templates.]

**Key lingo:**
- [Term 1]: [definition — how we use it internally]
- [Term 2]: [definition]

**Key numbers:** [quantifiable claims about the product — e.g., database size, speed benchmarks, coverage stats. These get used in P2 of emails.]

---

## Voice

**Sender:** [Name and company — who the emails come from]

**Tone:** [e.g., "Calm, analytical, builder-to-builder." Keep to 1 sentence.]

**Language level:** [e.g., "B2 English: simple, clear sentences. Polite but not over-polite."]

**Hard constraints:**
- [Rule 1 — e.g., "No dashes, no exclamation marks, no emojis."]
- [Rule 2 — e.g., "No buzzwords, no flattery, no hype."]
- [Rule 3 — e.g., "Sentence case only."]
- [Add as many as needed]

**Banned words:** [Words that must never appear in outreach — e.g., "agents", "try"]

**Scope boundaries:** [What the product IS and ISN'T — e.g., "Company-level intelligence, not people/panel data."]

---

## ICP

### Primary profiles

| Profile | Company size | Roles | Geographies | Why they buy |
|---------|-------------|-------|-------------|--------------|
| [Profile 1] | [range] | [titles] | [regions] | [reason] |
| [Profile 2] | [range] | [titles] | [regions] | [reason] |

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

- [Description of companies that look like ICP but aren't]

---

## Win Cases

| Customer | Profile | What worked | Result | Date |
|----------|---------|------------|--------|------|
| [Name/anon] | [profile type] | [what resonated] | [concrete outcome] | [YYYY-MM] |

### Quotes / signals from wins

- "[Direct quote or paraphrase from customer]" — [context]

---

## Proof Library

Pre-written proof point sentences for use in P4 of emails. Each entry has the sentence, the audience it works for, and the hypothesis it validates.

| Proof point | Best for audience | Best for hypothesis | Source win case |
|-------------|-------------------|--------------------|----|
| "[Full PS sentence ready to paste into an email]" | [audience type] | [hypothesis name or "general"] | [win case reference] |
| "[Another proof point]" | [audience type] | [hypothesis name] | [win case reference] |

Rules:
- Every proof point must trace back to a real win case above.
- Write the full sentence as it would appear in the email (including "PS.").
- When building a campaign prompt, pick 2-3 proof points that match the campaign audience and bake them into the prompt template with conditions for when to use each one.

---

## Campaign History

| Campaign | Vertical | List size | Reply rate | Top hypothesis | Key learning | Date |
|----------|----------|-----------|------------|---------------|--------------|------|
| [name] | [vertical] | [N] | [X%] | [#N name] | [1-sentence takeaway] | [YYYY-MM] |

---

## Active Hypotheses

### Validated (reply rate > X%)

1. **[Name]** — [2-3 sentence description with data points]. Best fit: [company type]

### Testing

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

### Retired

1. **[Name]** — retired because [reason]. Last tested: [date]

---

## Do Not Contact

| Domain | Reason | Added |
|--------|--------|-------|
| [domain.com] | [competitor/partner/requested/other] | [YYYY-MM-DD] |

Section Rules

• What We Do: Keep under 100 words. Update when positioning changes. Include email-safe version and key numbers.
• Voice: Defines the sender identity and tone for all outreach. Skills read this section to set the voice in prompt templates.
• ICP: Max 5 primary profiles. Anti-patterns prevent wasted outreach.
• Win Cases: Add every closed deal. Anonymous is fine ("a mid-market company in [industry]").
• Proof Library: Derived from win cases. Every proof point must map to a real win. Skills read this section when building P4 of prompt templates.
• Campaign History: One row per campaign. Update reply rate when final numbers are in.
• Active Hypotheses: Move between Validated/Testing/Retired based on campaign results. Target 5-7 active.
• Do Not Contact: Check before every list build and Instantly upload.