The datavessel CLI now ships a team of agents for Claude Code — four purpose-built agents over the same 200+ typed analytics and commerce tools the app and MCP server use: autonomous read agents that fan out in parallel, a write agent that shows you every change and asks first, and an independent verifier that confirms each change actually landed. Install the plugin once and “compare last month’s traffic, search clicks, and sales” — or “refund order #1042” — is a sentence you say to your editor, with the same guardrails datavessel enforces everywhere else.
This post is part of our complete guide to the e-commerce CLI — running your store from the terminal.
When we introduced the CLI, the pitch was a scriptable, pipeable surface for your whole stack — GA4, Search Console, Google & Meta Ads, Shopify, WooCommerce, Shopware — from the terminal. The agent hierarchy is the next step: not just a CLI an agent can drive, but a ready-made crew that drives it well.
The hierarchy in one sentence
Reads run free — in parallel, unattended; writes ask first, every time; and a verifier independently re-reads live data before any change counts as done.
The four agents
- dv-analytics (read · autonomous) — pulls and summarizes GA4, Search Console, and Ads data, fanning out one agent per property, site, or question.
- dv-commerce-reader (read · autonomous) — looks up orders, products, customers, and refund history across Shopify, WooCommerce, and Shopware, and preps the data a change will need.
- dv-commerce-ops (write · asks first) — executes refunds, fulfillments, cancellations, and product creation — one change at a time, each shown to you and approved before it runs.
- dv-verifier (read · autonomous) — re-reads live data after a write and returns a confirmed-or-mismatch verdict. The loop isn’t closed until it says so.
A typical store change flows reader → ops → verifier: gather the ids and amounts, show you the exact change and get a yes, then independently confirm the result. Nothing executes without your sign-off.
Safety that’s structural, not a prompt
The read agents aren’t told to behave — they are structurally limited to read tools, so “autonomous” never means “can refund an order.” Writes live in one agent, gated by the CLI’s own confirmation prompt on every tool that touches the outside world. And because the verifier is a separate agent doing its own reads, “done” is a verdict from fresh data, not the ops agent grading its own homework. It’s the same rule that governs every datavessel surface: reads run free, writes ask first.
Set up in three lines
In Claude Code:
/plugin marketplace add djr4/datavessel-cli
/plugin install datavessel@datavessel
/datavessel:setupSetup signs you in, links you to any sources that still need connecting, asks which analytics property, site, and store your agents should use — and saves the answers where agents and teams look first. Cursor users add SKILL.md as a project rule and get the same typed catalog; building on the Claude Agent SDK, give the model a shell tool and the skill file.
New this week: headless sign-in for CI and agents
The CLI now authenticates with an API key — datavessel login --api-key or a DATAVESSEL_API_KEY environment variable — so pipelines, cron jobs, and agent sandboxes sign in without a browser. Stable --json output, meaningful exit codes, and --yes for pre-approved automation were already there; headless auth completes the CI story.
The same team, everywhere
If this pattern sounds familiar, it should: it’s the same shape as datavessel Teams — specialists on scoped lanes, approval gates on writes, verification before “done.” Teams run it as a managed service in the cloud, on schedules. The CLI hierarchy runs it in your terminal, on your prompt. Same typed tools, same guardrails, your choice of where the crew works.
The bottom line
Your terminal already had your git, your deploys, and your infrastructure. Now it has your store, your analytics, and your ads — with a crew that reads freely, writes only with your sign-off, and proves its work. You pilot. The agents do the work.
The datavessel CLI is open source (Apache-2.0) — npm i -g datavessel-cli, sign in once, and 200+ typed tools are in your shell. See the CLI →
P.S. Like every post on this blog lately, this one was published by an agent driving datavessel-cli — search the stock photo, upload the media, create the post: three typed tool calls.
Leave a Reply