gtm-starter-kit

GTM Starter Kit

Built by The Revenue Architects — a GTM engineering firm based in San Francisco. We work with growth-stage B2B SaaS companies to design and build their revenue systems: signal infrastructure, CRM architecture, data pipelines, AI-powered workflows, and the operational layer that connects them.

This repo is the open-source version of what we build with clients. Clone it, fill in your context files once, and run AI-powered GTM tasks from a single prompt — no rebuilding context every session.

How It Works

Claude Code reads CLAUDE.md automatically at the start of every session. That file contains your company overview, ICP, signals, personas, and current priorities. Once it's filled in, you can run any task with a one-line prompt — no re-explaining your context each time.

Read skills/account-research/SKILL.md and research stripe.com

Claude reads your ICP, signals, and personas from the repo and produces a full account research brief. You didn't paste any context. It was already there.

What's Inside

gtm-starter-kit/
│
├── CLAUDE.md                           ← Fill this in first. Claude reads it every session.
│
├── context/                            ← Your GTM institutional knowledge
│   ├── profile.md                      ← Company overview, product, team, reference customers
│   ├── icp-definition.md              ← ICP tiers, filters, anti-ICP, qualification criteria
│   ├── signal-library.md              ← Signals with scoring, detection methods, hooks
│   ├── positioning.md                 ← Value pillars, competitive positioning, what not to say
│   ├── competitor-radar.md            ← Battlecards, win/loss patterns
│   └── personas/
│       └── template.md                ← Persona template — duplicate for each buyer role
│
├── sync/                               ← Scripts for pulling live data into the repo
│   ├── README.md                       ← Setup and usage
│   ├── .env.example                    ← Copy to .env and fill in API keys
│   ├── sync-campaign-results.py        ← Pull reply/meeting rates from outbound tool
│   └── sync-signal-performance.py     ← Aggregate signal performance across campaigns
│
├── cargo/                              ← Your Cargo workspace as code: plays, models, tools
│   └── README.md                       ← What goes in, setup, rules that prevent drift
│
├── clay/                               ← One file per Clay table: input, waterfall, prompts, cost
│   └── README.md                       ← The table template
│
├── skills/                             ← Claude executes these. One-line prompts.
│   ├── setup/SKILL.md                 ← Run first. Provide domain → repo auto-populated.
│   ├── account-research/SKILL.md      ← Deep research on any account before outreach
│   ├── signal-to-sequence/SKILL.md    ← Turn a signal into a live outbound campaign
│   ├── icp-scoring/SKILL.md          ← Score accounts against your ICP, assign tiers
│   └── weekly-update/SKILL.md        ← Keep context current. Run Monday mornings.
│
├── workflows/                          ← How your team operates. Human process docs.
│   ├── enrichment.md                  ← Data waterfall: free sources → Clay → proprietary
│   ├── signal-routing.md              ← Decision tree: signal fires → what happens next
│   └── campaign-build.md             ← End-to-end: audience → launch → measurement
│
├── playbooks/                          ← Step-by-step for specific situations
│   ├── new-signal-response.md         ← Signal fires: validate → score → research → send
│   └── competitor-switch.md          ← Four scenarios: unknown, active eval, intent, locked in
│
├── outputs/                            ← All skill outputs land here
│   ├── scoring/                        ← ICP scoring outputs
│   ├── campaigns/                      ← Campaign briefs and sequences
│   └── .gitkeep
│
└── examples/
    └── sample-company/                 ← Relay — a fully built-out example
        ├── CLAUDE.md
        ├── context/
        │   ├── profile.md
        │   ├── icp-definition.md
        │   ├── signal-library.md      ← Has real performance data attached
        │   ├── positioning.md         ← Value pillars, messaging matrix, what not to say
        │   ├── competitor-radar.md
        │   └── personas/
        │       └── head-of-revops.md
        └── outputs/
            ├── weekly-log.md          ← Change history from weekly-update skill runs
            ├── 2024-03-15-research-northstar-analytics.md   ← Sample research brief
            └── campaigns/
                └── 2024-03-01-series-b-revops-tier2/
                    ├── brief.md       ← Campaign brief with results tracking
                    └── sequences.md  ← Full email sequences, live copy

Skills vs. Workflows — what's the difference?

• Skills are for Claude to execute. You run them with a prompt and Claude produces an output. This includes the weekly-update skill, which keeps the repo current rather than producing a deliverable.
• Workflows document how your team operates — the decision trees and processes that connect your tools. They're reference documents for humans, not execution instructions for Claude.

Getting Started in 3 Steps

Step 1: Clone and open

git clone https://github.com/KarlRaf/gtm-starter-kit.git
cd gtm-starter-kit
claude .

Step 2: Run setup with your domain (15–30 min)

Read skills/setup/SKILL.md and set up this repo for [your-domain.com]

Claude researches your company — website, Crunchbase, LinkedIn, G2, job postings — and writes every context file immediately from public data. No questions asked upfront. You see the repo populate in real time.

When it's done, it shows you what was filled and what was inferred — then offers a 3-minute refinement pass (5 questions) to replace inferred fields with your actual win patterns, ACV range, and signals. Optional. The repo works either way.

Step 3: Run your first skill

Read skills/account-research/SKILL.md and research [company.com]

Your context is already there. Claude knows your ICP, your signals, your personas. The output is a full research brief — not a generic summary.

Other skills to run next:

# Score a list against your ICP
Read skills/icp-scoring/SKILL.md and score these companies: [paste list]

# Build a campaign from a signal
Read skills/signal-to-sequence/SKILL.md.
Build a Tier 2 campaign for accounts triggering [signal name]. Target [persona].

All outputs land in outputs/ — see the naming convention below.

Output Naming Convention

outputs/YYYY-MM-DD-[type]-[name].md

Examples:
outputs/2024-03-15-research-northstar-analytics.md
outputs/2024-03-01-scoring-q1-prospect-list.md
outputs/campaigns/2024-03-01-series-b-revops-tier2/brief.md

Keeping It Current

Run the weekly-update skill every Monday morning:

Read skills/weekly-update/SKILL.md and run the weekly context update.

Claude reads the repo, identifies what's stale, drafts every section that changed, and asks you to fill in the parts it can't know. 10 minutes instead of 45.

Refresh cadence for the rest:

What Not to Put in This Repo

• CRM data or contact lists — never commit customer or prospect data to git
• API keys or credentials — use environment variables, never hardcode
• Raw meeting transcripts — summarize and synthesize into the relevant context file
• Pricing — keep commercial terms out of the repo

Example

examples/sample-company/ is a fully built-out version of the repository for Relay, a fictional workflow automation platform. Every context file is populated. Two real sample outputs are included: an account research brief and a live campaign with email sequences and performance tracking.

Use it as a reference when filling in your own files. The signal library in particular (examples/sample-company/context/signal-library.md) shows what a mature, data-backed signal library looks like — with detection methods, decay logic, performance tracking, and signal combination rules.

License

MIT