product-searchlisted

# Product Search — Verified Supplement Sourcing ## Why this skill exists This skill exists because of a specific, repeatable failure mode: recommending supplements based on **brand name as a proxy for clean formulation** or **Amazon search titles as if they were ingredient facts**. The typical failure looks like: "Nootropics Depot PEA is a clean brand" → recommend URL → user opens the actual product page → the supplement facts panel lists **rice** as an excipient → rice is on the user's hard-no list → recommendation was worthless. The fix is not "try harder." The fix is **structural discipline**: every recommendation must survive a real fetch of the actual product page, a full extraction of every ingredient (active AND excipient), and a cross-check against the user's current documented constraints. No shortcuts. No assumptions from brand reputation. No guessing when a page fails to load. This skill also builds **persistent institutional memory** — a brand database and a verified-products list — so that 3 months from now when the user asks about the same brand or SKU, the work already done is reused rather than redone. ## Core principle Three words the skill operates under: **Fetch. Cross-check. Persist.** - **Fetch**: every candidate product gets its actual page fetched and parsed — no shortcuts from search results. - **Cross-check**: every ingredient is tagged against the user's current constraints, which are pulled fresh from their memory files at the time of invocat