Step types (overview)
Core execution step kinds:
- start: Entry point. Routes to the first executable step via
next. - end: Terminal step.
- http: HTTP requests (endpoint, method, headers, data, timeout, assert, save, optional retry)
- python: Inline code (code with main(...), data args, assert, save, optional retry)
- iterator: Loop controller (collection, element, task, mode, concurrency, where, order_by, limit, chunk)
- duckdb: DuckDB SQL (commands/sql, unified auth mapping, extensions, attach, save, optional retry)
- postgres: PostgreSQL SQL (command/sql, auth, idempotent DDL, upserts, save, optional retry)
- snowflake: Snowflake SQL (DDL/DML/query, VARIANT/JSON handling, setup/verify around transfers, optional retry)
- workbook: Invoke a named task defined in the
workbookblock. - playbook: Compose and call another playbook (
path, optionalreturn_step), enabling modular pipelines. - sink: Not a top-level type; used inside steps to persist results (event log, postgres, duckdb, http, python custom code).
Cross-cutting capability:
- retry: Inline policy block available on action steps (http, python, postgres, duckdb, workbook task, inner iterator task) controlling bounded re-attempt logic (see
retry.md).
Auth references:
- See
../credentials.mdfor credential JSON structure and multi-credential mapping patterns.
See individual pages in this folder for capabilities, required/optional keys, context rules, and usage patterns.