cheatsheets

Add debug logging nodes to track iteration progress, verify the convergence condition uses correct types (numeric comparison, not string), and ensure the cycle mapping uses the result.* prefix for PythonCodeNode outputs. Always call workflow.build() before creating the cycle.

Use the node-selection-guide cheatsheet, which provides a decision guide for choosing between nodes like PythonCodeNode for data transformation, HTTPRequest for API calls, and LLMNode for AI tasks. It is marked as a critical reference for node selection.

Cyclic workflows are a planned feature not yet implemented in SDK v0.9.31 according to the cycle-debugging cheatsheet. Current alternatives include Python loops, recursive workflows, or SwitchNode state machines.

Use external pool injection by creating one shared asyncpg pool at application startup and passing it to AsyncSQLDatabaseNode via the external_pool parameter. The SDK borrows the pool without closing it, preventing 240+ connections across Gunicorn workers.

The common-mistakes-catalog lists critical rules: never use raw SQL (use DataFlow instead), always call .build() before runtime.execute(), never use relative imports, and never mock in Tier 2-3 tests where real infrastructure is required.