v0.34.1

Fixed

• AI agent attribution no longer falls back to ucp_unknown for Product/Version-form UCP-Agent headers (#588).
  • An agent that identified itself with a Product/Version header carrying a trailing comment — e.g. UCP-Agent: Claude/4.6 (Anthropic) — was attributed as utm_source=ucp_unknown instead of its brand. The parser rejected the parenthesized comment outright, and even a clean Claude/4.6 had no product-token mapping. The order was still correctly counted as an AI order (via utm_id=woo_ucp); only the which agent was lost.
  • The header parser now tolerates and discards a single trailing ( … ) comment (RFC 7231 §5.5.3), taking only the leading product token. The comment body is not scanned for a nested product, so a browser-style User-Agent still buckets as "Other AI" rather than letting a bot name inside the comment masquerade as a known agent.
  • Added product-token mappings for the known answer agents (ChatGPT, Claude, Gemini, Perplexity, Copilot) plus their crawler user-agent tokens (GPTBot, ClaudeBot, Perplexity-User, …), so a client that sends its User-Agent value in the UCP-Agent header resolves to the correct brand and hostname instead of fragmenting into "Other AI".
  • Because the same header parser feeds the UCP access gate, a merchant who has disabled a brand's crawlers now blocks that brand's Product/Version-form requests too — consistent with how the profile-URL form already behaved.