arize-link

Arize-link generates deep links to the Arize UI for traces, spans, sessions, datasets, labeling queues, evaluators, and annotation configs, producing clickable URLs that can be shared with team members. It solves the practical problem of turning IDs found in exported data or logs into working links back to the corresponding resources in the Arize application, so users can quickly open or view a trace, span, session, dataset, evaluator, or annotation config.

The skill collects required inputs and substitutes them into URL templates. Every link requires org_id and space_id (both base64-encoded), plus resource-specific IDs: project_id and trace_id (optionally span_id) for traces and spans, project_id and session_id for sessions, dataset_id for datasets, queue_id for a specific queue, and evaluator_id (optionally version) for evaluators. A critical rule is that all path IDs must be base64-encoded, because a raw numeric ID produces a valid-looking URL that 404s; if a user provides a raw internal ID such as Organization:1:abC1 it must be base64-encoded first. For trace, span, and session links the startA and endA epoch-millisecond time range is required, with a priority order of reusing a user-provided URL's values, padding a span's start_time by plus or minus one day or hour, or falling back to the last 90 days. The base URL is https://app.arize.com and can be overridden for on-prem, and evaluator version values must be URL-encoded (for example a trailing = becomes %3D).

The skill targets Arize users and teammates who work with exported trace data or parsed URLs and need to link back to the UI. Its four-step process gathers IDs, verifies base64 encoding, determines the time window, and presents a clickable markdown link. A troubleshooting table covers empty views (widen the time window), 404s (check base64 encoding of IDs), unhighlighted spans (the span_id may belong to another trace), and an unknown org_id (the ax CLI does not expose it, so copy it from the browser URL). The related arize-trace skill can export spans to obtain trace_id, span_id, and start_time.