Workflow batches and dispatch reliability
Workflow runs now validate artifact inputs, expose full dispatch schemas, and batch template and relationship operations with clearer all-or-none behavior.
Workflow execution got stricter, faster, and easier for the agent to call correctly. This entry combines the May workflow reliability wave that started with batch operations and finished with full dispatch declarations.
Full declarations for dispatch plans
Plan steps with may_dispatch_to now emit full workflow_<id>() declarations with parameter schemas and dispatch labels. Function artifacts required by the plan get the same full-schema treatment, with GCS fallback when the index row is incomplete.
That means guided and autonomous runs can see the actual function and workflow contract before dispatching, instead of relying on a slim or incomplete tool declaration.
Batch template creation
Template-instance creation now has a dedicated batch path. Multi-instance requests validate required fields and enum values up front; if one instance is invalid, the whole batch is rejected before anything is written.
Agents can create structured sets of instances in one pass without leaving half-written template outputs behind.
Batch relationship cleanup
Relationships now support operation="batch_archive" for bulk cleanup in one pass. System-managed relationship types are skipped with explicit reasons, and graph/cache invalidation runs after archived edges are written.
The relationship and canvas-editor guides now document the accepted batch shapes so agents can use the operations directly.
Reliability: workflow artifact inputs accept both single IDs and validated ID lists, required-input validation closes placeholder and junk-value bypasses, and autonomous workflow runs can be cancelled cleanly with parent notification wiring.