**Cargo Manifest** is a 📝Cargo product: an MIT-licensed repository template for running a 📝Go-To-Market (GTM) motion the way engineers run software, with everything the company knows, everything it is trying to do, and everything that runs in production in one repo an AI agent can operate. It is the structure Cargo's own GTM team runs on, extracted into a template, and it is what cargo-ai cdk init scaffolds; the infra/ directory is a 📝Cargo CDK project and every other layer works standalone without a Cargo workspace. It is aimed at founders doing founder-sales, heads of growth, and RevOps leads who never write code themselves and let 📝Claude Code or 📝Cursor read the operating manual and handle git, deploys, and file conventions.
Manifest describes a structure our Agentic Collaboration System already covers in MythOS: plan, cadence, initiatives, and outputs all have homes here. What I want from it is the infra/ layer and its guardrails, with defineContext pushing our doctrine into the workspace rather than a second context store in a repo.
Key Features
- Ten fixed layers —
plan/,context/,initiatives/,cadence/,.claude/skills/,infra/,scripts/,evals/,outputs/, andscratch/, with a strict dependency direction: infra reads context, skills read everything, outputs are memory only. - Confidence ladder —
proof/,signal/, andinsight/files carryhypothesis,validated, orproven; outreach motions may cite only validated or proven claims, and a claim earns validated on its second occurrence. - Hypothesis-first motions — an outreach motion declares segment, angle, trigger, and expected reply rate before any copy exists, and the
new-motionskill creates the context file, the infra, and the eval suite together. - CI-only deploys — CI runs
cargo-ai cdk planon every pull request and deploys on merge to main; local deploys are denied by convention and hard-blocked per agent in Claude Code, Cursor, and 📝Codex. - Append-only outputs — every list built, campaign shipped, or brief written lands in a dated
outputs/directory with anoutcome:field, so motions rank by results and agents read recent entries before related work.
Getting Started
- Clone the template:
git clone https://github.com/getcargohq/cargo-manifest acme-gtm. - Run
npm install, then open the folder in an agent that readsAGENTS.md. - Tell the agent to read
AGENTS.mdand seed the repo; it interviews you and fills the layers. - Replace every
ACME:marker in the knowledge layers with your own claims. - Wire the engine:
cd infra && npx cargo-ai cdk plan, adding cookbooks withcdk add.