What problem does it solve? Running a multi-stage retail data medallion pipeline (bronze, silver, gold, semantic model, publish) by hand in CI is error-prone and risks skipping required human approvals. This Skill orchestrates the already-approved sequence unattended as a Dagster asset graph, enforcing every gate and recording run evidence without ever self-approving a stage. ## Core Features & Use Cases - Gated asset graph execution: Runs an 11-asset Dagster graph where each asset executes only after its gate (gate exit code or committed human approval) is satisfied, and fails closed otherwise. - Human-seam enforcement: Halts at mapping approval, judgment calls (grain, PII, rollup), and the publish wall, escalating to a named human instead of auto-resolving. - Derived run evidence: Writes per-run records with per-asset exit codes, outcomes, blocking reasons, and named owners, without ever marking a readiness stage as passed. - Use Case: A data team wants nightly CI runs of their retail warehouse pipeline. They use this Skill so Dagster executes only approved steps, stops at unapproved mapping gates, and produces auditable run evidence for each run. ## Quick Start Ask the agent to run the Dagster medallion pipeline for a table using the seshat dagster run command after confirming the mapping gate is cleared.
