linking-to-the-dashboard

Linking to the dashboard

Every run, prompt, and trace turn has a canonical dashboard page. When you cite one in an answer, give the full clickable URL — a bare slug or uuid forces the reader to hunt. The base URL is https://2027.dev/evals (BASE_URL in lib/url.ts).

Link formats

With {orgDomain} = the org's domain (e.g. mastra.ai):

• Prompt detail: https://2027.dev/evals/{orgDomain}/prompts/{promptId} (promptId is a uuid)
• Report overview: https://2027.dev/evals/{orgDomain}/reports/{slug}
• Full report: https://2027.dev/evals/{orgDomain}/reports/{slug}/report
• Trace with turn anchor: https://2027.dev/evals/{orgDomain}/reports/{slug}/trace#turn-{N}
• Run status (in-flight): https://2027.dev/evals/{orgDomain}/runs/{runId}
• Org home: https://2027.dev/evals/{orgDomain}

#turn-N anchors

N is the turn's 1-based `index` field returned by get_trace — NOT the 0-based offset parameter. Don't confuse them.

Screenshot-merge caveat: the dashboard trace page always interleaves browser screenshots into its turn numbering, but get_trace only does that with includeBrowser: true. So for a report that has screenshots, build #turn-N anchors from a trace fetched with includeBrowser: true, or your turn numbers will point at the wrong turns. For runs without screenshots the indices match either way.

Where the ids come from

• get_prompt_benchmarks returns prompt.{ id, orgDomain } — prompt links are fully constructible.
• list_prompts / get_prompt return id but NOT orgDomain — use the org domain you're operating for (your conversation's org context).
• list_runs / get_run already return absolute runUrl, statusUrl, and (when a report exists) report.url (the dashboard report page) and report.apiUrl (the REST resource) — prefer those returned URLs over hand-constructing.
• list_reports / get_report / get_trace return slug but NOT orgDomain — use your conversation's org context for the link.

The org segment must be the canonical org domain that owns the prompt/report — the pages redirect if it's wrong, so use the orgDomain a tool returned (or the returned absolute URL), not a guess.

Practical loop

1. Whenever you cite a run, prompt, or trace turn, construct the full URL from the formats above.
2. Get orgDomain from the tool response when available, otherwise from your conversation's org context.
3. Citing evidence from a specific trace turn? Link .../trace#turn-N (with the screenshot-merge caveat above) so the human lands on the exact moment — this is also how you let them see browser screenshots you can only read as text (see the reading-browser-screenshots skill).