Product Catalog

When to use

Call into this skill whenever the user asks about gear / products / stock / ratings / prices for things the app sells. Examples:

• "what's the highest-rated luggage under $200"
• "show me what's in stock right now"
• "compare the carry-on bags by rating"
• "any travel adapters under $50"

Do NOT use for questions about flights, hotels, weather, or itinerary — those have their own skills. Do NOT use for general retail questions ("where can I buy X") — that's the web-search skill.

How to use

Two-step pattern, ALWAYS:

1. Call describe_schema() first if you don't already know the table

layout this turn. Returns table + column shapes so you don't have to guess column names.

1. Call query_products({ sql, limit? }) with a precise SELECT. The

validator REJECTS anything other than SELECT (no INSERT/UPDATE/DELETE/ DROP/ALTER/CREATE/PRAGMA/ATTACH/REPLACE/TRUNCATE/VACUUM/REINDEX, no statement chaining). limit defaults to 50 (max 200); if you over- fetch, the response includes truncated: true.

Query craft

• Use JOIN reviews USING (product_id) and AVG(rating) for "highest

rated" questions.

• Use WHERE price <= ? for budget filters.
• Use WHERE in_stock = 1 for stock filters.
• Order by what the user implicitly wants (rating desc, price asc, etc).
• Limit to 3-5 unless the user explicitly asks for more.

Output narration

Quote the actual product names from the result rows. Round prices to the nearest dollar. Mention rating to one decimal place. Do NOT invent products or specs that aren't in the result.

If the validator returns { ok: false, reason: "only SELECT..." }, that means you tried to write — re-think the query as a SELECT.

Anti-patterns

• Do NOT skip describe_schema and guess column names.
• Do NOT issue multiple SELECT statements separated by semicolons (the

validator rejects chains).

• Do NOT invent product names that aren't in the SELECT result.
• Do NOT recommend a product based on your training data; this catalog

is the source of truth for what THIS app sells.

Source & license

This open-source skill is cataloged on AgentStack and links to its original source — we do not rehost the code.

• Author: hec-ovi
• Source: hec-ovi/agentickit
• License: MIT

Install and usage instructions live in the source repository linked above.