Six skills and four commands for building a shopping agent or a merchant agent on the packages
in this repo (../../README.md), or bringing one that exists onto them. Skills
and commands alike load when a conversation matches their description, and a command can also be
typed; the commands read the reference repo (this checkout, a local clone, or a fresh one) while
they run. The plugin runs no code of its own.
claude plugin marketplace add anthropics/commerce-agents # or the path of a local clone
claude plugin install commerce-builder@claude-commerce-agents| Command | Does |
|---|---|
/scaffold-commerce-agent |
Interviews you about your stack, plays the plan back, and scaffolds a shopping agent, a merchant agent, or both on the reference packages |
/add-commerce-flow <flow> |
Adds one shopping or merchant flow to an existing agent: copies its skill, wires the tools it calls, authors its first eval cases |
/author-commerce-evals |
Builds the eval suite: a runner, the first ten cases against your own catalog, and a replay gate for CI |
/review-commerce-agent |
Maps an agent you already run, compares it with the reference row by row, and converts the rows you pick |
The scaffold and the review write what they learned to your project's CLAUDE.md under
## Commerce agent decision record; the other two commands read that section instead of asking again.
| Skill | Rules for |
|---|---|
commerce-architecture |
One model per conversation, the backend interface, and which layer (tool description, prompt, skill) a rule belongs in |
commerce-prompt-caching |
Keeping the static prompt and tool list byte-stable, and checking that requests hit the cache |
commerce-ui-tools |
Presentation tools, server-side enrichment, the event stream, and adding a component |
commerce-trust-safety |
Fencing, provenance and caps, identity, gate state, grounding, memory writes, adversarial cases |
commerce-evals |
The case shape, the scorers, hostile fixtures, and the run pattern |
commerce-merchant-operations |
Staged changes, guardrails, host approval, metrics grounding, and marketplace scoping |
/scaffold-commerce-agent, then wire the backend methods it stubbed; with an agent already running,/review-commerce-agentand the rows it converts./add-commerce-flowfor each flow in your v1 index./author-commerce-evals, and re-record after every prompt, skill, or runtime change.- Deploy per
../../docs/deployment.md.
Authentication, and the memory controls a deployment owes its users, are wired by the deployment;
../../docs/safety.md lists them. Where this plugin's text and the
reference code disagree, the code is right.