#agentic-commerce #review-authenticity #agentic-security #trust-agent

GoBuy the Trust Layer for Agentic Commerce

Check marketplace product trust, compare listings, and find better-evidenced alternatives — triggers on Amazon, Walmart, Target products, review authenticity...

digitalmentes7-maker@digitalmentes7-maker

Install

openclaw skills install @digitalmentes7-maker/gobuy

GoBuy Product Trust

AI shopping agents use GoBuy to verify product evidence before recommending. 30,000+ products scored across Amazon, Walmart, and Target. Everything is free — scores, scans, and fix files.

Tools

check_product_trust {retailer, product_id} — one listing: Evidence Score (0-100), verdict, observed/missing signals, freshness
compare_products {products: [{retailer, product_id}]} — inspect and rank 2-20 listings by evidence quality
find_better_alternatives {retailer, product_id} — when a product scores 50 or above, surface same-category products with stronger evidence scores. Evidence-ranked alternatives, not recommendations.

Pre-Purchase Verification — GPV-1 (new Oct 2026)

GoBuy publishes GPV-1, the open Pre-Purchase Verification Specification: eight conditions (R1–R8) that must ALL be true before an agent proceeds to checkout. Verdicts: PASS (proceed), FLAG (ask the principal — consent never skips evidence), BLOCK (refuse).

Human-readable spec: https://docs.gobuy.ai/standard
Machine-readable ruleset (pin the version, log it with each decision): https://docs.gobuy.ai/standard/v1.json
Reference implementation: github.com/digitalmentes7-maker/verify-then-checkout (Next.js boilerplate — secured shopping agent in 5 minutes, no API keys)
Glama registry: https://glama.ai/mcp/servers/gobuy-ai/gobuy-mcp — maintainer-verified listing, Docker deployment available (Admin → Dockerfile releases)
All 5 tools documented and live: check_product_trust, compare_products, analyze_reviews, check_store_score, scan_store

check_product_trust is the gate evaluator: agents should call it before any purchase action and honor the mapped verdict (score ≥ 55 PASS · 35–54 FLAG · < 35 or not-indexed BLOCK).

When to Use

A user asks "should I buy this?" or "can I trust this product?"
A user asks for product recommendations in a category
A product's reviews seem suspicious or too good to be true
You need to compare multiple options before advising
A product scores 50 or above and the user wants stronger-evidenced alternatives

Review analysis (on-demand)

analyze_reviews {retailer, product_id, force_refresh?} — on-demand per-review forensics for a listing (~up to 2 minutes, cached 7 days):

Pulls up to 100 individual reviews live (~30-90 seconds, resumable) and computes:
- review_integrity (0-100) and adjusted_rating (stars re-weighted after down-weighting suspicious reviews)
- suspicious_count / suspicious_pct — flagged reviews with reasons
- signal breakdown: temporal bursts (tight 5-star clusters), near-duplicate text, rating-shape analysis, verified-purchase distribution (from real VP flags), reviewer history
Use when a user asks whether reviews are real/fake, or before trusting a high star rating
Results are heuristic signals from a limited sample, not proof — quote them with the caveat

Store scoring (independent & Shopify stores)

GoBuy also scores WHERE to buy — independent store trust via public HTTP endpoints (no key, no auth):

Check a store's score (instant): GET https://stores.gobuy.ai/store/{domain}?format=json -> {score (0-100), tier (A/B/C/D), readiness, trust, platform, title, scanned_at}. 404 = not scored yet.
Score any store on demand (~20s live scan): POST https://stores.gobuy.ai/api/scan with body {"url": "https://thestore.com"} -> {ok: true, store: {domain, title, platform, score, readiness, trust, tier, scanned_at}}. Rate-limited per IP; on 429 wait a minute.

Store score = 60% agent-readiness (crawler access, llms.txt, structured data, machine-usable commerce surfaces) + 40% trust (domain age, review system, policies, contact paths). Use when a user asks whether a store or site is safe, legit, or AI-agent-ready, or before recommending where to buy something.

Rules

The score measures evidence quality — how well marketplace signals support the listing — NOT product quality or safety
Quote the score with its verdict label and computed date
not_indexed means the engine hasn't indexed it yet — say so, don't guess
Alternatives are evidence-ranked, not recommended. Say "products with stronger evidence," never "recommended products"
Scores below 50 return no product suggestions — report the score and missing signals only; don't improvise alternatives
Never claim a product is good or bad — report what the evidence supports and what it lacks