← ALL PLAYBOOKS
—— THE AI-NATIVE GTM SERIES · PART 2 OF 4

Build your GTM context brain this week

Episode 2, made runnable. Context is the product, but bad context in is bad output out. Define your ICP, positioning, and voice first, then build the repo any agent can run on.

Install the whole playbook ↓BUILD BETTER · SERIES 1 Watch the episode →
START FROM THE KIT

Clone the Built GTM starter kit.

Do not start from an empty folder. Clone the Built GTM starter kit: a five-layer context repo with the CLAUDE.md brain, every context template, the loop and context skills, workflows, and a fully worked example. Substance-first, runnable on clone.

git clone https://github.com/Built-GTM/builtgtm-starter-kit.git
Built-GTM/builtgtm-starter-kit View on GitHub →

New to GitHub? Run the github-for-gtm skill first.

POWERS UP WITH YOUR STACK
HOW TO CONNECT →
ClaudeChatGPTGeminiCopilotNotionGoogle DriveConfluenceGitHub
1Solve the problem

Most teams rush to build the repo before they have defined their ICP, positioning, or voice. Bad context in is bad output out. The tool was never the problem, and neither was the folder structure. The missing, undefined context was.

RUN THIS AND YOU CAN
  • ✓Define your ICP, positioning, and voice before anything reads them
  • ✓Write a CLAUDE.md brain any session, teammate, or agent warm-starts from
  • ✓Build a greppable context repo an agent can actually follow
  • ✓Keep it current with a decision and ICP-evolution log
2See it in production

Before you run it: I ran it for real.

This workflow is genericized from a real build. Here is the number it moved, what I actually did, and a way into the whole story.

THE BUILD · BUILD RELIABILITY
The Governance File
7 gates
no score writes to the CRM until all seven pass
6 accounts
the last run mis-scored, caught and corrected before the next one
468 fields
the write guard re-checks every run, so a bad field never fails
0 errors
the plays check must pass on all three jobs before publish
WHAT I DID ›
01
Turn each failure into a rule. Every way the pipeline broke became a written rule: a minimum brief size, a link-integrity check, an index-count guard. Not documentation, enforced gates.
02
Lock the design. I put the rules in a governance file the pipeline reads before it ships anything. If a run would violate a past-failure rule, it stops instead of shipping broken output confidently.
03
Make the QA part of the run, not a person. The checks run every time, automatically. The system polices itself against its own history, so a fixed bug stays fixed.

When you automate, your failures go silent. Write them down as enforced gates, not documentation. A governance file that reads your own failure history is how a pipeline stops repeating its worst week.

Dig into the full build →
SEE IT RUN

The plain questions, answered in order.

Each step is the plain question you are already asking. The skills answer them in order, one handing its receipt to the next.

Get the substance right
“Who exactly are we for?”
OPEN THE SKILL →
Get the substance right
Say what you do and why you win, the same way every time.
OPEN THE SKILL →
Get the substance right
“How should we sound?”
OPEN THE SKILL →
Build the repo
“What is GitHub and why do I need it?”
OPEN THE SKILL →
Build the repo
Write CLAUDE.md, the brain Claude reads every session.
OPEN THE SKILL →
Build the repo
Build the context files, greppable and durable.
OPEN THE SKILL →
Build the repo
Organize the repo so an agent can actually follow it.
OPEN THE SKILL →
Keep it current
Log the calls and ICP changes so you stop re-litigating them.
OPEN THE SKILL →
EXAMPLE CHATNadia, Head of GTM, running the loop in one sitting
N
I want to build one of those Claude Code context repos so my AI stops starting from scratch every session. Where do I begin?
G
Get the substance right· Define your ICP
Before we touch a repo: bad context in is bad output out. First we get the substance right, starting with your ICP. Not Series B SaaS, that is a category. Real tiers with criteria, an anti-ICP, and must-haves kept separate from red flags.
Tiers3, with real criteria
Anti-ICPwho to exclude on purpose
Receipt
A real ICP definition with tiers, an anti-ICP, and an evolution log, saved to context/icp-definition.md.
N
That is the sharpest our ICP has ever been. What next?
G
Get the substance right· Positioning
Now positioning, so the repo says one thing about what you do and why you win. One statement, value pillars with proof, and a what-not-to-say list so the AI never oversells.
Receipt
A positioning statement plus a message house, saved to context/positioning.md.
N
Good. Honestly I worry the AI will still make everything sound generic.
G
Get the substance right· Brand voice
That is the third piece of substance: your voice. Voice pillars, tone by context, a do and do-not word list, and sample lines the model can copy. Now every output sounds like you, not like AI.
Receipt
A voice guide with pillars, a do and do-not list, and sample lines, saved to your context.
N
Now can we finally build the repo? I have honestly never used GitHub.
B
Build the repo· GitHub for GTM
No terminal needed. A repo is a folder that keeps its history, and GitHub stores it in the cloud. Install GitHub Desktop, make a private repo, and remember the one rule: never commit CRM data, keys, transcripts, or pricing.
Receipt
A private repo created in GitHub Desktop, with a .gitignore keeping your keys out.
N
Repo is created. How do I make Claude actually use all this?
B
Build the repo· Context pack
Write CLAUDE.md at the root, the brain Claude reads every session: your ICP, positioning, voice, and this week's priorities, kept scannable in two minutes.
Receipt
A CLAUDE.md brain committed to the repo, so every session starts warm.
N
And the rest of the context, the signals, competitors, personas?
B
Build the repo· Knowledge base
Built out the context files in plain, searchable markdown: signals, competitors, personas. Any answer is one search away, for you and for the agent.
Receipt
A searchable context folder: signals, competitors, and personas.
N
It is getting big. How do I keep it from turning into a mess?
B
Build the repo· Workspace organizer
Organized the repo so a human, and an agent, can follow it: clear folders, a README front door, plain file names. Nothing moved until you approved the plan.
Receipt
A clean structure and a README, so the repo stays findable as it grows.
N
Last thing. How do I keep it current so it does not rot?
K
Keep it current· Decision log
Start a decision and ICP-evolution log: each entry captures the call, the why, and when to revisit. Update the CLAUDE.md priorities weekly, five minutes. The repo now compounds instead of going stale.
Receipt, loop closed
Your ICP, positioning, and voice defined, then a context brain any teammate or agent runs on, kept current with a weekly rhythm.
RUN IT YOURSELF

Two steps to run this yourself.

1

Install the skills. Grab the one-command install below. Once installed you never call a skill by name, you just describe your situation and the right skill fires on its own.

2

Paste this to kick off. Swap in your own situation and the skills run in order.

“Here is my situation: [describe your workflow or account in a sentence or two]. Walk me through Build your GTM context brain this week step by step, and run each skill as we go.”
3The ideal run, step by step

The workflow that solves it, one step at a time. Each step hands off to the next. Tab through the run.

Turn a category into a real ICP definition.

Get the substance right
Turn a category into a real ICP definition.
WHY THIS STEP
Define who you are for, in tiers with real criteria. Bad ICP in is bad everything out.
WHAT IT DOES
Names your tiers with explicit criteria, the fit signals, the anti-ICP you exclude on purpose, and a qualification framework, plus an evolution log. A category is not a definition.
YOU GETThe tier criteria, the anti-ICP, the must-have vs red-flag framework, and the evolution log.
POWER IT UPClaudeChatGPTGeminiCopilotNotionGoogle Drive
OPEN THE SKILL →
WHERE YOU LAND

Your ICP, positioning, and voice defined, then structured into a CLAUDE.md brain and a context repo any teammate or agent runs on, kept current.

INSTALL THE WHOLE PLAYBOOK

One plugin. 8 skills, bundled.

The whole playbook installs as a single plugin in Claude Code, no copy-pasting 8 times. Or grab any single step from the run above. It runs on what you paste; connect your stack to go live.

THE AI-NATIVE GTM SERIES · PART 2 OF 4
← PART 1
Run the GTM loop on one workflow
PART 3 →
Build your signal library and wire it to motion
── GET THE NEXT PLAYBOOK

One runnable workflow a week. Free.