Skip to content

Dashboard Bootstrap

dashboard-bootstrap bootstraps project reference SQL and Dosi metrics from a dashboard through an installed BI plugin. The workflow is skill-driven; it does not add a dashboard-specific CLI command.

Use natural language or the compatibility shortcut /bootstrap-bi. The shortcut delegates to the same chat/skill pipeline and does not invoke the legacy picker or BI streams. Text after the command is forwarded as scope, for example:

/bootstrap-bi use Superset profile prod and dashboard 42

Prerequisites

  • Enable a BI plugin and profile that provide dashboard discovery, stable query candidates, SQL export, and a credential-free source identity for every query.
  • Configure the corresponding physical databases as Datus datasources. No BI-profile-level datasource mapping is used.
  • Before generating metrics for a query batch, select the Datus datasource uniquely matched from those queries' real connection identities.
  • Use Dosi when creating or updating metrics. MetricFlow and plain OSI projects remain query-only.

Workflow

Ask Datus to bootstrap a dashboard, for example:

Build reference SQL and metrics from the revenue dashboard.

The skill guides these selections in order:

  1. BI plugin and profile;
  2. dashboard;
  3. queries to index as reference SQL;
  4. queries to use as metric evidence.

It then prints a Generation Manifest and stops. Confirm or correct that manifest in the next message. After confirmation, the selected plugin exports SQL, gen_sql_summary indexes each reference query, and semantic_modeling creates or updates the related Dosi datasets, relationships, and metrics. As the final step, dashboard-bootstrap loads create-subagent when that skill is available and persists the dashboard's main and attribution nodes in the loaded agent.yml.

The two query selections are independent. A query may be used for reference SQL, metrics, both, or neither.

Dashboard subagents

When Agent configuration is mutable, the final step creates or updates two nodes using the legacy naming and tool pattern:

  • <platform>_<dashboard> uses gen_sql with database and context-search tools.
  • <platform>_<dashboard>_attribution uses gen_report with semantic attribution tools.

Both nodes are scoped only to successful tables and exact metric/reference-SQL subject references for the active datasource. Metric scope uses <metric.subject_path>.<metric.name> from the synchronized Dosi model; reference-SQL scope uses <subject_tree>.<name> from the generated SQL summary. Bare subject paths select whole subtrees and are not used for an exact Dashboard selection. The generic create-subagent skill resolves these references against the post-sync subject trees, edits agent.agentic_nodes without replacing sibling entries, and verifies the YAML after writing. When the runtime marks configuration read-only, that skill is not discoverable and the workflow skips persistence without failing context construction.

Auto-run

Explicitly say skip confirmation, auto-run, or an equivalent instruction to let the workflow continue after printing the manifest. Auto-run does not bypass system permission prompts.

Safety and limitations

  • Partial, failed, unselected, or checksum-mismatched SQL is never sent to a builtin agent.
  • Dashboard labels, descriptions, and SQL comments are treated as untrusted source data.
  • Missing, weak, or ambiguous query-level source identity blocks metric generation for only the affected queries; reference SQL may continue.
  • A dashboard may span multiple datasources. Metric queries are partitioned by their uniquely matched Datus datasource, and only the currently active partition is authored in one run.
  • Successful context generation does not by itself prove numerical equivalence with the source dashboard.
  • Subagent creation failure does not invalidate successfully built context and can be retried independently.