A plug-and-play go-to-market system run by Claude Code agents. Clone it, open it in Claude Code and say "set me up". Claude interviews you, writes your playbook and pricing into the repo, and walks you through connecting your CRM and call recorder one step at a time. Then you have:
- Call prep before every meeting, built from your playbook and the deal's history.
- Post-call automation: the call ends, a transcript is scored 0 to 100 against your playbook, the CRM deal is updated, follow-up tasks are created, a follow-up email is drafted, and a card lands in your sales Slack channel. Nobody types anything.
- Calls graded on MEDDPICC. Every call shows which letters it advanced and which are still open, so coaching is specific and immediate.
- A weekly pipeline digest and deal-by-deal risk analysis that names each deal's real blocker by its MEDDPICC letter.
- Inbound that announces itself: a demo booking or form fill is enriched through Clay, scored for ICP fit, and posted to Slack in seconds with where they came from (INBOUND.md).
- Outbound: Clay target lists into lemlist on warm and cold rails, a click that queues a LinkedIn connect, and a daily call list ranked by engagement (OUTBOUND.md).
- The close: terms sheet and order form generated from the deal and your pricing, a Gmail draft for you to send, a DocuSign envelope that moves the deal to Closed Won when it's signed, and a draft invoice from there (CLOSE.md).
- Every gotcha we paid for, with the fix (TRICKS.md).
- A canon that stays true: one file per fact, checks that catch drift, and an agent that keeps it clean.
It was extracted from a working B2B sales system that ran an enterprise motion, then stripped of everything company-specific. What is left is the structure and the rules that kept it honest.
KNOWLEDGE LAYER (.claude/skills/, docs/)
what is true and what to do: pricing, product, playbook, rubric
│ read by
┌───────────────┬───────────┴───────┬──────────────────┐
▼ ▼ ▼ ▼
OPERATOR POSTCALL DIGEST KEEPER
(you + Claude (hourly routine: (Friday routine: (daily routine:
Code, live) score + write CRM) pipeline roll-up) hygiene, tasks)
│ │ │ │
└───────────────┴─────── git (the bus) ────────────────┘
│
CRM ── Slack ── call recorder ── email drafts ── enrichment
Two layers. Skills hold knowledge; agents do work and read the skills. An agent never carries its own copy of a fact. When a fact changes you change one file and every agent picks it up on its next run.
Git is the bus. Agents do not message each other. They write files on their own branches, and the files are the messages. Every change is a commit, so you can always answer "what did we believe on the 3rd, and who changed it."
- Clone and open in Claude Code.
git clone … && cd gtm-engine && claude. Accept the "trust this folder" prompt. Claude will ask before editing files under.claude/skills/; approve those (or pick "always allow"), because that is setup writing your answers down. - Say "set me up". Claude walks you through setup one step at a time. It interviews you,
fills
company-context,product-contextandcommercial.md, tells you exactly where to click for the steps it can't do, and checks each step before moving on. Stop whenever you like and say "continue setup" next time; progress is kept inSETUP-PROGRESS.md. - Connect the stack. At minimum: a CRM, a call recorder with a Claude connector, Slack and Gmail. See STACK.md.
- Score a call by hand. Paste a transcript and say "process this call". That runs
sales-intake, which scores, writes the deal, and drafts the follow-up. Confirm each write. - Turn on the routines once the interactive run is right:
agents/postcall, thenagents/digest, thenagents/keeper.
Full walkthrough: SETUP.md.
| Path | What |
|---|---|
| CLAUDE.md | Instructions every agent loads. Your non-negotiables go here (derived copy) |
| SETUP.md | First-day setup, in order |
| STACK.md | Recommended tools per layer, alternatives, and the gotchas we paid for |
.claude/skills/ |
Knowledge layer. One folder per skill. Versioned like code |
agents/ |
Agent layer. One runbook per scheduled agent, plus its state |
| docs/architecture.md | How the pieces fit, and the six loops |
| docs/non-negotiables.md | The rules that cause real damage when broken. Single home |
| docs/system-hygiene.md | One source of truth, retire don't duplicate, worktrees |
| docs/bus-contract.md | How agents share work through git |
| docs/design-system.md | Who reads this, and are we asking them for something? |
clients/ |
Client evidence. One folder per account. Permanent records |
client-facing/ |
The only folder cleared to send to a client as is |
tasks/ |
Work queue, one file per task, with an owner and a done-when |
NEEDS.md |
Open decisions and gaps. The only open-items list |
ops/ |
Checks (check-corpus.py, check-client-copy.py) and webhook receivers |
| ops/receivers/INBOUND.md | Book a demo → Clay enrichment → fit score → Slack card → deal |
site/attribution.js |
Website snippet: first-touch UTMs, Calendly handoff, hidden form fields |
| outbound/OUTBOUND.md | Clay target list → lemlist (warm/cold rails, UTMs) → click → LinkedIn connect → daily call list |
outbound/ |
Outbound scripts: load-lemlist.py (CRM-checked list load), daily-call-list.py (engagement ranking, phones, CALL tasks) |
| TRICKS.md | Every gotcha we paid for, with the fix |
| close/CLOSE.md | Close: deal → Summary of Terms + Order Form from commercial.md → Gmail draft → filed and tagged; DocuSign signature → Closed Won → draft HubSpot invoice |
sessions/ |
Hooks that give every new Claude session a handoff from the last one |
archive/ |
Retired documents with tombstones. Never loaded as truth |
| Skill | Job |
|---|---|
setup |
The guided first run: walks SETUP.md one step at a time and tracks progress |
company-context |
ICP, sale motion, buying roles, positioning, proof. Every sales skill reads it |
product-context |
What the product does and what state each part is in |
sales-playbook |
The stages, the calls, objections, and commercial.md, the one home for prices |
crm-records |
How contacts, companies and deals are created, and the stage ladder |
meeting-prep |
A plan to win the next meeting, per attendee |
sales-call-scorer |
Scores a call 0 to 100 against RUBRIC.md and PLAYBOOK.md. Evaluates only |
sales-intake |
The post-call entry point: match, score, write the deal, tier, tasks, follow-up draft |
call-debrief |
The leadership readout of one call |
pipeline-analysis |
Every open deal: health, the real blocker, one move |
weekly-exec-update |
The Friday cross-deal digest |
humanize-copy |
Strips AI tells from outgoing prose |
MIT. Use it, fork it, sell with it.