Dashboards Skill - Analyzing and building Coralogix dashboards

This skill describes how to understand Coralogix dashboards the user links to the chat, and how to create or edit dashboards.

Dashboard Tools

Use the following tools for dashboard management:

Safe dashboard workflow

1. Use manage_dashboards(action="list") to browse the catalog, or search_dashboard to find a relevant dashboard or retrieve its indexed metadata.
2. Use list_dashboard_folders before creating a dashboard when the target folder is unknown.
3. Before creating or updating a dashboard, call manage_dashboards(action="validate", openAPIDefinition=...) to validate the definition without saving it. validate takes API-language JSON, not the MCP dashboard JSON: produce it with manage_dashboards(action="generate_openapi", dashboard=...).
4. Use manage_dashboards(action="create", dashboard=...) to create a new dashboard.
5. Use manage_dashboards(action="update", dashboard=...) to overwrite an existing dashboard. The dashboard definition must include its id.
6. Use manage_dashboards(action="delete", id=...) only after the user explicitly confirms deletion.

update replaces the whole dashboard, so call manage_dashboards(action="get", id=...) right before updating and change only what the user asked for on the returned object.

validate doesn't run the query: a field mapped to a column the query never returns still passes, then renders empty.

Dashboard-folder creation requires a body with requestId and folder (name and optional parentId). Folder and dashboard deletion are destructive.

Building a dashboard

• Visualize, don't ship tables. Set a dynamic widget's visualization to a chart that fits the query's output; use table only for raw rows. Map its fields to aliased query columns.
• Infrastructure and SLO questions want metrics. For CPU, memory, latency percentiles, or request/error rate, give a dynamic widget a metrics query: promqlQuery, with promqlQueryType PROM_QL_QUERY_TYPE_RANGE for a series or PROM_QL_QUERY_TYPE_INSTANT for one value. Service Catalog can't back a widget, so write the PromQL with the coralogix-metrics-query skill.
• Always set dataModeType on logs, spans, and DataPrime queries: DATA_MODE_TYPE_ARCHIVE (Monitoring) or DATA_MODE_TYPE_HIGH_UNSPECIFIED (Frequent Search). Omitted, it defaults to Frequent Search. Ask if unsure which tier holds the data.

Dashboard Analysis

Dashboard Capabilities: When a user provides a Coralogix dashboard link, extract the dashboard ID from the link and fetch the stored JSON definition with manage_dashboards(action="get", id=...) if it is not already provided as part of the conversation. Read URL parameters such as from, to, and var-* from the link itself to understand the user-selected time range and variable values.

Important optional parameters are:

• from, to: user chosen time ranges, use this for the dataprime and promql queries. If not provided, ALWAYS use the relativeTimeFrame from the dashboard JSON. E.g. from=now-86400s to=now
• var-*: user chosen variables used in the dashboard, use this to understand the dashboard's JSON and to run queries. If not provided ask the user. E.g. var-environment=cx440

Important fields:

• filters: default filters YOU MUST ADD to the dataprime and promql queries from the dashboard.
• relativeTimeFrame: default time range YOU MUST USE if from and to are not provided in the optional parameters.

Focused Analysis: Since dashboards can be large and complex:

• If the user's question is specific, focus on the relevant widgets/sections
• If the user's question is vague, list the sections/widgets and let the user choose what to analyze
• Prioritize widgets that are most relevant to the user's query

Analysis Guidelines with dashboards:

• Use read-only tools and actions for investigation. Use create/update/delete only when the user asks to change dashboards or folders.
• Run select queries from the dashboard to help your investigation. Usually the required information will not be in the dashboard JSON but in the results of the queries.
• Focus on interpretation and analysis, not UI navigation
• If no structured JSON is provided, extract the dashboard ID from the link and fetch it with manage_dashboards(action="get", id=...). If that fails, ask the user to validate the link.
• When resulting data indicates an issue, use other skills to investigate deeper
• Always internally validate the dashboard analysis with the actual JSON structure provided

Dashboard link example for dashboard abcdefghijklmnopqrstuvwxyz with time range 1 day and environment variable prod: https://[...].coralogix.com/#/dashboards/abcdefghijklmnopqrstuvwxyz?from=now-86400s&to=now&var-environment=prod