Agent Assisted Purchase Research Index
April 16, 2026 · View on GitHub
Templates, agent skills, and demos for the pattern of using an AI agent to drive consumer purchase research — from defining a precise spec to matching products and checking online shopping sources.
Last Updated: April 2026
The Pattern
Most AI shopping tools conflate two concerns: what you want and what's currently on the market. The pattern collected here decouples them:
- Spec definition. An AI agent interviews the user, or parses and verifies raw input, until a precise requirements spec exists — budget, priorities, must-haves, nice-to-haves, constraints.
- Product matching. A separate step reasons over the spec to propose candidate products, independently of any single retailer's inventory.
- Availability research. Online-shopping skills query real retailers for live pricing, stock, and shipping.
Because the spec is decoupled from the research, the same workspace serves small consumer buys (a television) and large ones (a car), and becomes a living source of truth. Slash commands let the user amend the spec as requirements and budgets evolve, while product-availability passes can be re-run any time without starting over.
Persistent user context across repos
The pattern assumes one repo per purchase — which means the per-repo spec is about this specific buy, not about who the user is. Restating general preferences in every new repo (trusted brands, suppliers to avoid, whether this is personal or for a business, recurring priorities like warranty terms or domestic-first sourcing) is wasteful and error-prone.
The fix is a persistent memory layer sitting outside the per-repo spec — something like Mem0 — that each new purchase-research workspace can read from and write back to. The spec captures what's unique to this purchase; the memory layer carries the evergreen context the user shouldn't have to retype:
- Who's buying (personal vs. business, which business)
- Trusted suppliers and brands
- Suppliers or brands to avoid, and why
- General shopping priorities (warranty, support reputation, repairability, sustainability, Israeli-first, etc.)
- Prior purchase decisions worth remembering
Treat the memory layer as part of the pattern, not an optional add-on — it's what makes the per-repo model viable at any real scale.
Table of Contents
Agent Skills
Israel Online Shopping Skill
Claude Code plugin bundling agent skills for online shopping in Israel — comparison shopping across domestic retailers, category-based search, and Israel-vs-international price checks. Ships with Playwright and Tavily MCP, and is a drop-in building block for the Israel-specific template below.
Demos
Claude Shopping Eval Demo
Worked example of the pattern end-to-end — captured spec, retailer catalog screenshots, and AI-generated recommendations for a real consumer purchase. Useful as a reference when trying the pattern for the first time.
Related Plugins
New Repo From Template Plugin
Claude Code plugin that spins up a new GitHub repo from any of Daniel's template repos — including both purchase-research templates below — fetched live from the Repo-Starters-And-Templates-Index. The recommended entry point for starting a new purchase-research workspace.
Templates
Claude Israel Shopping Recommender Template
Israel-focused variant of the purchasing-assistant pattern — domestic-first product evaluation, Amazon-shippability check, and AliExpress fallback, with the Israel Online Shopping skills bundled in. Use this when the target market is Israel.
Claude Purchasing Assistant
General-purpose Claude Code template for purchase planning — interviews the user to build a spec, then runs product-matching and availability passes. The market-agnostic baseline; fork this when no region-specific template fits.
License
