Manifest

Scaffold a GTM repo readable by humans and coding agents alike

Manifest is the repo framework for GTM engineering: one conventional, version-controlled repository holding everything your go-to-market runs on, readable by humans and coding agents alike. cargo-ai project init scaffolds that repo.

A Manifest repo has six layers:

Layer	Directory	One sentence
Context	context/	What the company knows: the GTM brain every agent reads before acting
Skills	.claude/skills/	What agents know how to do in this specific repo
Infra	infra/	What runs in production, declared as a project
Scripts	scripts/	What you run by hand, for surfaces the project cannot declare yet
Evals	evals/	What keeps agents honest: prompt regression tests in CI
Outputs	outputs/	What happened: the append-only archive agents read as memory

The layer that executes (infra/) needs a Cargo workspace. The rest (context, skills, evals, outputs, the context linter) work in any repo with no account at all.

project init

# scaffold a fresh repo
cargo-ai project init acme-gtm

# or omit the directory and be asked — name, cookbook, install, and
# whether to hand the repo straight to your coding agent
cargo-ai project init

Options:

• --name <name> — repo name written into package.json (default: the directory name)
• --cookbook <name> — cookbook to install into the new repo, skipping the “Start from” question (cargo-ai project cookbook list names them)
• --force — scaffold into a non-empty directory (existing files are still kept, never overwritten)

Every repo is scaffolded from the same base, getcargohq/cargo-manifest. What varies is what you layer on top of it — a cookbook, either with --cookbook here or cargo-ai project add later.

Everything is one command group now: cargo-ai manifest was removed in @cargo-ai/cli 1.0.70, because it scaffolded the same repo from the same base as cargo-ai project init and the split only made you guess which one to run. The standalone cargo-cdk bin takes the same subcommands.

Adding a cookbook

A cookbook is a worked example — resources, the procedure that operates them, and the inputs it will ask you for. add installs one into a repo that already exists, splitting it the way the repo is laid out: infra/ to the project, the rest to .claude/skills/.

cargo-ai project cookbook list          # what is available
cargo-ai project cookbook view cookbook/tam-building
cargo-ai project add cookbook/tam-building

add never deploys. It reports every file it would overwrite and stops unless you pass --overwrite, then prints what the cookbook still needs from you. Omit the address to pick from the catalog.

Growing the repo

Once the repo exists, add to it the way you add any other code:

• Skills — npx skills add getcargohq/gtm-skills installs one-job GTM skills your agent can run; npx skills add getcargohq/cargo-skills teaches it this CLI.
• Resources — declare models, plays, tools and agents in infra/ as a project, then cargo-ai project plan and cargo-ai project deploy.
• A cookbook — cargo-ai project add cookbook/<name> installs a worked example into the repo you already have.