AgentShelf: OpenAI Product Feed Validator
Under maintenancePricing
Pay per usage
Offline validation of OpenAI agentic-commerce product feeds with per-row rejection reports and optional normalized export. Independent tool: never uploads feeds or calls an OpenAI API.
The feed itself, as CSV, TSV, JSON or NDJSON text. Use this for feeds small enough to paste; use feedKey for anything larger.
Name of a record in this run's own default key-value store holding the feed. A plain record name, not a path and not an address; nothing outside this run's store can be read.
Force a parser instead of sniffing the content.
Which field-name set and rule set to apply.
Your registered display name, used to check the seller_name rule. Supplied by you and taken on trust: AgentShelf has no merchant account and cannot verify it. Not a credential.
Apply the additional rules that only bind when the feed is used for Ads.
Also report fields that are recommended rather than required. These are warnings and never block ingest on their own.
Map a Google-dialect feed into OpenAI native field names and include the mapped rows. Required for the accepted-rows CSV. This is the higher-priced pay-per-event outcome.
Write rejections.csv, and accepted-rows.csv when normalizing, as key-value store records. The CSV bodies are never inlined into the dataset item.
Which finding severity makes the run exit 1. never still fails a run that broke; it only suppresses findings.
Refuse a feed larger than this rather than validating a truncated prefix of it.
