Claude to propose weekly groceries for approval
Wmoto combined four years of household purchase history with a voice-driven shopping list. Alexa adds requested items to Todoist, while a weekly GitHub Actions job uses simple frequency rules to suggest regular purchases that the household might have forgotten.
zenn.dev
Original prompt
Shared by the source author. See source notes for context.
/apply
Reach people exploring AI agents
See placements →Evidence at a glance
Sourced account- Source author
- 生ビール (@wmoto_ai)
- Original account
- Read the source ↗
- Published
- Source link checked
- Reported result
- The weekly routine produces a reviewed shopping plan and then updates the supermarket basket; purchase completion is not reported.
- Human involvement
- The household reviewed product images and prices in GitHub and used approval or revision comments before basket changes.
A linked account supports attribution. It does not mean we independently reproduced its result.
Report an issue with this case
Open correction emailOpens your email app with the case link included. Nothing is sent automatically.
How the agent did it
- Combine Todoist requests with weekly purchase-frequency suggestions.
- Have Claude search familiar supermarket products and publish a GitHub approval issue.
- Approve, revise or cancel the plan before the separate basket-update step.
Wmoto combined four years of household purchase history with a voice-driven shopping list. Alexa adds requested items to Todoist, while a weekly GitHub Actions job uses simple frequency rules to suggest regular purchases that the household might have forgotten.
Claude Code handles product selection after that statistical preparation. It reads the shopping brief, searches the online supermarket through a dedicated command-line tool and chooses familiar products, such as the usual brand and size of milk. It writes candidate and purchase-plan files, which become a GitHub issue showing product images and prices.
The household reviews that issue on a phone. A comment can approve the plan, request a substitution or cancel it. Only after approval does a separate Playwright step put the validated products into the supermarket basket. The account does not describe this as an unattended paid order.
Several controls make the workflow specific. Hooks restrict the commands Claude can run and check that chosen product identifiers actually came from search results. Authentication files are kept outside the agent's permitted access. The author retained simple frequency calculations because recurring purchases were explainable and easier to debug that way. Claude's role was resolving the shopping brief into real products, while the household retained the final choice.
Prompt
Original prompt
Source: zenn.dev · 生ビール (@wmoto_ai)
Frequently asked questions
Why was replenishment prediction not left to the model?
Simple purchase-frequency rules were sufficient, explainable and easier for the author to debug.
Could Claude directly use supermarket credentials?
The described controls denied access to authentication files and limited its work to the dedicated search tool and plan files.
What did the approval command do?
The published /apply command approved a validated plan for the separate Playwright basket step.
More details
About this account
Claude selected supermarket products through a restricted CLI and produced validated candidates and a purchase plan.
Firsthand source opened and read. Prompt is a verbatim excerpt of the published original instruction.
Credit: 生ビール (@wmoto_ai)
Source published
Added to Agents In Use
Source checked
Tools and topics
Claude Code · Todoist · Alexa · GitHub Actions · Playwright
Grocery replenishment · GitHub approval · Purchase patterns
Editorial completeness: 100/100. Measures documented fields, not agent performance or verified outcomes.