viant/endly/endly-authoring

Work from the workflow directory. Inspect endly -r=run -t='?' for top-level tasks, endly -s='*' for services, and endly -s=dsunit -a=prepare for a live action contract. A pipeline maps names to tasks in declaration order. action: service:action selects an action; action: run with request: '@relative/workflow' invokes another workflow. Use init to publish inputs, post to publish outputs, when for conditions, defer/onError according to the actual workflow schema, and $var/${var} for state.

endly -r=run runs the whole workflow. -t=init,build,test chooses tasks in the order supplied. Legacy selection recursively resolves bare task names and flattens groups. Opt in with -selector-mode=path -t=group.child,other for exact dotted paths, retained parent init/post/conditions, and execution in selector order. Each selector is a separate execution, including repeated parents; account for repeated setup. A parent path selects its entire subtree. Paths address tasks in the current workflow, not the contents of a separate request: '@...' workflow.

-i is a comma-separated list of exact generated TagIDs, not a task path or a glob. Capture IDs from workflow output. It filters matching tagged actions while ordinary setup runs. Unknown IDs can leave a task unfiltered; never report a focused rerun without checking the emitted case IDs and executed assertions. Keep leading zeros.

For database fixtures, register the datastore, query dsunit:sequence for the fixture tables, publish Sequences, then call $AsTableRecords('data.dbsetup') and feed its table-to-record map to dsunit:prepare. AsTableRecords is a UDF, not dsunit:AsTableRecords or AsTableRecors. Sequence allocation resolves symbolic IDs and cross-table references in two passes. Use $Sequences.TABLE/${tag}.name for allocation and the same symbol for foreign keys; a suffix like data.catalog_dbsetup/catalog publishes rows under catalog.TABLE.<tag>.name. Do not reorder sequence lookup after conversion or convert the same mutable fixture repeatedly to allocate IDs. Read service/testing/dsunit/sequences/e2e/run.yaml and service/testing/dsunit/udf.go in the Endly checkout for implementation details.

Use the service-specific skill for each action. For testing methodology retrieve endly-testing; retrieve the local project skill when one is explicitly configured.

In path mode, tag filtering excludes unmatched template instances across groups while retaining untagged setup. Discover IDs in the actual root/nested run context with endly_listInstances; standalone loading can change the workflow name prefix.