datasqrl/sqrl/init-sqrl

Follow these steps to initialize a DataSQRL project:

1. When an available examples directory contains a useful starting point, review its README.md and relevant projects. Treat its contents as optional reference material, not required templates.
2. Adapt relevant example files into the working directory only when they fit the task. Don't overwrite existing files. If no example fits, scaffold the project with /opt/agent/cmd.sh init <type> <name> instead (it creates the <name>/ folder; see CLI_REFERENCE.md). Pick the type by the pipeline's output:

• api: data served through a GraphQL API (Flink, PostgreSQL, Kafka, Vert.x). The production source connectors/sources-prod.sqrl is an internal Kafka table (/*+ engine(kafka) */), exposed as a mutation.
• stream: results written to Kafka topics (Flink, Kafka). The production source is an internal Kafka table, and the script exports its result with EXPORT HelloWorld TO log.HelloWorld;. Without an API, only exported tables are written anywhere: export each result the user needs, and the script must have at least one EXPORT to compile.
• dataset: results written to Iceberg and queried through DuckDB (Flink, Iceberg, DuckDB). The production source is an existing Iceberg table ('connector' = 'iceberg'): set its catalog options (catalog-type, catalog-name, warehouse, catalog-table) to the user's lakehouse.
• The test package of every type reads the sample data in connectors/test-data/ and compiles and tests out of the box. Replace the Messages sources and the sample data with the real sources (step 4).

1. Review provided source/sink table definitions in sub-folders to identify the sqrl files that define the sources and sinks required for the implementation. If the requirements reference a provided data catalog, include that catalog and delete copied connectors unless the requirements specifically ask to create additional sources/sinks outside the catalog.
2. Update copied files to match user requirements by:

• ALWAYS invoke the relevant skill before updating or implementing anything
• Use user-provided source/sink definitions from sub-folders or an included data catalog if they exist (include/import them, reusing what exists), otherwise update the template connectors. Invoke the /manage-connector skill (Connector Organization) before writing or editing any source/sink definition.
• Updating the SQRL files to use those sources & sinks and adjust implementation by matching the code to the updated table schemas and removing features that are not required. Invoke the /implement-sqrl skill before writing or significantly modifying any SQRL logic during this step.
• When renaming the main .sqrl script, update script.main key in the (sub-)project's base config so every package.json reference matches the actual filename before the first compile. When creating or updating any configuration, invoke /configure-sqrl skill.
• Update the project configuration if necessary.
• Update the GraphQL API definition and defined operations to match the new schema.
• Create or update the configuration files: a *-shared-package.json base plus *-<env>-package.json thin per-environment overlays. Invoke the /configure-sqrl skill (Base Config and Environment Overlays).
• Add /*+test */ annotated SQL test queries at the end of the main SQRL script(s) for each exposed table. These test queries should SELECT from the table with a WHERE filter on a known test data value and produce deterministic results. Invoke the /test-sqrl skill (Writing Tests) for the test and test-data rules, then delete the template's existing snapshots and run the tests to generate fresh ones (Running Tests, Snapshot Lifecycle).
• Create or update the test runner script run-tests.sh which is the entry point for the project's tests. Invoke the /test-sqrl skill for details.
• Update README.md with user requirements and project description. Add a list of features that are required by the user but not part of the template and need to be implemented