Latitude Artifacts

An Artifact is a single HTML file your agent builds from Latitude data: a one-off report that answers a specific question, or a KPI dashboard the user reopens every morning. Latitude provides the data through the MCP, CLI, API and SDKs; the agent picks the queries, renders the page, and writes the findings. The user may never open the Latitude web app.

This skill covers the whole loop: figure out what the user wants, pull the right data, render it in the shared Latitude design language (light + dark), interpret it, and, when asked, make it refreshable.

Files in this skill:

• data-sources.md: which Latitude operation answers which question, queryAnalytics reference, units, filters, privacy rules. Read it before fetching anything.
• template.html: the starter artifact with the Latitude theme, building blocks and an example data blob. Every artifact starts from this file.
• refresh.md: generator-script recipes (Node, Python, shell + CLI) and scheduling for refreshable artifacts.

Entry points

• Invoked directly. The user asks for a report, dashboard or "how is X doing" page. Run the intake below, skipping any question the request already answers.
• After latitude-setup. Setup builds its own first Artifact (a single-session page from its bundled first-artifact.html) and does not call this skill. Use this entry when the user, having just finished setup, asks for more: a report over a window, a comparison, or a refreshable dashboard. Auth (LATITUDE_API_KEY, LATITUDE_PROJECT_SLUG in .env) is in place, so skip the preflight; run the intake with the project already known, and treat the first-artifact default (below) as the suggested answer to question 1. Remember the project is minutes old: say the window is small and the numbers are a baseline, not a trend.
• Refreshing an existing artifact. The user points at an artifacts/<slug>.html that has a refresh script. Run the script, reload the blob, rewrite the findings, done. No intake.

Preflight: how will you read Latitude?