dagster-workflows
Route Dagster intent to Seshat's governed medallion workflow or the official Dagster competence owner without bypassing readiness, approvals, or evidence.
- License 许可
- Apache-2.0
- 最近更新
- 2026-11-11
- Stars
- ★ 2
- Forks
- 0
01 前言 Front Matter
- name
- dagster-workflows
- description
- Route Dagster intent to Seshat's governed medallion workflow or the official Dagster competence owner without bypassing readiness, approvals, or evidence.
03 SKILL.md 正文
Governed Dagster workflows
Read ../../portable-operating-contract.md before acting. For a governed
Seshat medallion run, use only the installed seshat dagster helpers; never
invoke dagster directly against the orchestration project and never bypass
the gate readers, the closed-argv runner, or the redaction layer. Dagster RUNS
already-approved steps; the gate exit code and the named human decide whether
a stage passed.
Intent ownership
| User intent | Seshat pre-gate | Execution owner | Seshat afterward |
|---|---|---|---|
| Decide whether orchestration fits the BI workflow | Readiness state and orchestration assessment | Seshat selection policy | State one truthful next action |
| Run the governed medallion graph | Per-asset readiness, evidence, and named-human approval gates | Dagster through seshat dagster |
Record derived run evidence, blockers, and the earliest truthful next action |
| Author assets, jobs, resources, components, or project structure | Prove the official skill is activated and discoverable | Official dagster-expert skill |
If the project enters Seshat's governed flow, return through the Seshat gates above |
| Design schedules, sensors, or declarative automation | Prove the official skill is activated and discoverable | Official dagster-expert skill |
Keep Seshat automations stopped until a named human starts them |
| Generic Dagster CLI guidance, validation, or troubleshooting | Prove the official skill is activated and discoverable | Official dagster-expert skill |
Do not infer a readiness effect from native success |
| Publish a Power BI result | publish_ready: pass and the separate publish adapter gate |
Official Power BI executor when available | Dagster records the trigger result and never publishes itself |
Spec 148 provides the read-only harness proof for the official dagster-expert
skill. Before routing generic competence, run
seshat integrations setup --profile orchestration --harness <claude-code|codex>
and require the dagster-agent-skills result to be discoverable. Otherwise report
the exact upstream-integration blocker. Do not copy official guidance into this
router, claim an installed payload is usable, or invoke native Dagster commands
against the governed project as a fallback.
Fixed workflow
-
Preflight read-only, without a database query:
seshat dagster doctor --jsonReport blockers verbatim with their remediation hints. Database credentials are reported present/absent only -- never echo a connection value. An absent DSN is a deferred boundary, not a failure to hide.
-
Execute one governed job as a shell-free child process in the orchestration project's own environment:
seshat dagster run --job <full_sequence_job|through_gold_job> [--table <table>] [--source-mode <csv|existing-bronze>] [--json]--source-modeselects the Bronze origin (the gated tail is identical for both and is never bypassed):csv(default): a raw<table>.csvlands and the loader OWNS and reloadsbronze.<table>(drop-and-reload). Use it for a file-first workspace. This is the one destructive-by-design mode.existing-bronze: NON-DESTRUCTIVE DB-first. An already-loadedbronze.<table>is verified READ-ONLY (existence, row count, and that its columns cover the approved source-map) and used as the satisfied upstream. It issues ZERO Bronze DDL/DML. Use it when Bronze already lives in the warehouse (BCP, COPY, an external ETL, a prior tool); never export Bronze to a CSV just to feed the default path.raw_source_filerecords adeferredboundary (no landing file by design), not a failure. A missing, empty, or source-map-mismatched relation fails closed with a named blocker.
The chosen mode is explicit in the command result and in the run evidence (
bronze_table's measured facts). Any doctor blocker stops the run before anything executes. A failed or blocked gate halts every downstream asset fail-closed; report the recordedblocking_reasonand its named owner, then stop. -
List runs or render the committed derived record:
seshat dagster evidence [--run-id <id>] [--json]The committed record is written by that verb as
orchestration/dagster/run-evidence/<run-id>.md, rendered deterministically from the raw records. Never edit raw records or rendered evidence by hand.
Hard boundaries
- A green asset means "the command ran and returned this exit code" -- NEVER "the stage is now pass". Readiness passes come only from the gate exit code and a named human; this adapter records, it does not decide.
- The shipped schedule and sensor are STOPPED by default; starting either is a named-human action, never an agent default.
- Publishing is walled off: the graph only TRIGGERS the publish feature and fails closed while it is absent. Never publish or simulate publishing.
- Without database credentials, DB-touching assets record a deferred boundary and block fail-closed. Report that truthfully; never fabricate a run, an asset result, or run evidence.
- Never write a readiness pass, an approval, a confidence score, raw adapter output, a credential, a DSN, or an absolute local path.
Exit meanings
- exit 0: command completed; run evidence remains derived, not an approval.
- exit 1: usage error (unknown verb or flag).
- exit 2: preflight or gate refusal -- doctor blockers, a gate not CLEARED, a missing prerequisite, or evidence records refused schema validation (nothing executed).
- exit 3: the run failed or halted fail-closed on a gate (the CI signal); evidence is still rendered and must be reported.
- exit 4: unexpected internal error, already redacted; report it verbatim -- it is not a fixable evidence-record problem.
04 同仓库更多 Skills
fake-skill
A fixture skill with declaring frontmatter, used as positive shipped evidence.
fake-skill
A fixture skill with declaring frontmatter, used as positive shipped evidence.
bare-skill-no-frontmatter
暂无简介
fake-skill
A fixture skill with declaring frontmatter, used as positive shipped evidence.
retail-kpi-knowledge
Retail KPI / metric-contract reasoning layer for BI agents in the Seshat BI project. Use when an agent must reason about the BUSINESS MEANING of a retail KPI before any implementation — defining a KPI in business terms, classifying additivity (fully/semi/non-additive), identifying grain, listing required source fields, resolving ambiguity (gross vs net, VAT, returns, cost method, same-store), choosing an MVP KPI pack, reviewing a metric contract for completeness, or preparing a clean handoff to an implementation layer. Also use it to classify KPI sufficiency or prepare an owner policy decision packet. This is a meaning/definition + review layer, not an executor: it does not write DAX/SQL/Python, grant readiness, approve policy, or design dashboards. See INDEX.md for live/planned routes; registry.yaml is the authoritative machine-readable inventory.
bi-sql-knowledge
SQL reasoning, validation, reconciliation, transformation, diagnostics, and anti-pattern knowledge layer for Seshat BI agents. Use when a task involves reasoning about source SQL: source profiling, table grain, keys & uniqueness, joins & fan-out amplification, aggregation correctness, COUNT/NULL semantics, deduplication, validation queries, reconciliation queries, silver/gold transformation logic, window/date-time analytics, SQL anti-patterns, or basic SQL performance reasoning, or interpretation of a supplied sanitized PostgreSQL JSON execution plan. The SQL counterpart to bi-dax-knowledge. It is a thinking & validation layer, not a database executor, runtime validator, dbt/Dagster project, or automatic tuning service.