fix(engine): restore skip.columns edges in topologically_sort_column_configs

The sync builder executes generators in compile-time sort order (from
_column_configs, populated via this function), not ExecutionGraph order.
Dropping skip.columns edges caused evaluate_skip_when to hit UndefinedError
when the referenced column hadn't been generated yet, silently skipping rows.

Also:
- Refactor edge building into _add_edge() helper with a label parameter to
  distinguish "required" from "skip.when" edges in debug output
- Rename test_dag.py -> test_topological_sort.py to match the new module location
- Add from __future__ import annotations (required by AGENTS.md)
- Add test_side_effect_column_ordering covering the side_effect_map.get() path
- Add test_skip_when_column_ordering covering the skip.columns edge path

Author: przemekboruta

packages/data-designer-engine/src/data_designer/engine/dataset_builders/utils/execution_graph.py

@@ -363,19 +363,21 @@ def resolve(col_name: str) -> str | None:
     upstream: dict[str, set[str]] = {name: set() for name in dag_col_dict}
     downstream: dict[str, set[str]] = {name: set() for name in dag_col_dict}
 
-    # Only required_columns edges are added here. skip.columns edges are intentionally
-    # omitted: ExecutionGraph.create handles them in its own two-pass build, which is
-    # the authoritative execution-order graph. This function only needs to produce a
-    # valid topological ordering of ColumnConfigT objects for config compilation.
+    def _add_edge(name: str, dep: str, label: str) -> None:
+        resolved = resolve(dep)
+        if resolved is None:
+            return
+        logger.debug(f"{LOG_INDENT}🔗 `{name}` depends on `{resolved}` [{label}]")
+        upstream[name].add(resolved)
+        downstream[resolved].add(name)
+
     logger.info("⛓️ Sorting column configs into a Directed Acyclic Graph")
     for name, col in dag_col_dict.items():
         for req in col.required_columns:
-            resolved = resolve(req)
-            if resolved is None or resolved == name:
-                continue
-            logger.debug(f"{LOG_INDENT}🔗 `{name}` depends on `{resolved}`")
-            upstream[name].add(resolved)
-            downstream[resolved].add(name)
+            _add_edge(name, req, "required")
+        if col.skip is not None:
+            for skip_col in col.skip.columns:
+                _add_edge(name, skip_col, "skip.when")
 
     in_degree = {name: len(ups) for name, ups in upstream.items()}
     queue: deque[str] = deque(name for name, deg in in_degree.items() if deg == 0)

…ngine/dataset_builders/utils/test_dag.py …_builders/utils/test_topological_sort.pypackages/data-designer-engine/tests/engine/dataset_builders/utils/test_dag.py renamed to packages/data-designer-engine/tests/engine/dataset_builders/utils/test_topological_sort.py packages/data-designer-engine/tests/engine/dataset_builders/utils/test_dag.py renamed to packages/data-designer-engine/tests/engine/dataset_builders/utils/test_topological_sort.py

@@ -1,6 +1,8 @@
 # SPDX-FileCopyrightText: Copyright (c) 2025-2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 # SPDX-License-Identifier: Apache-2.0
 
+from __future__ import annotations
+
 from typing import Any
 
 import pytest
@@ -135,3 +137,39 @@ def gen_b(row: dict[str, Any]) -> dict[str, Any]:
     ]
     with pytest.raises(ConfigCompilationError, match="already produced by"):
         topologically_sort_column_configs(column_configs)
+
+
+def test_side_effect_column_ordering() -> None:
+    """A column that depends on a side-effect column is sorted after its producer."""
+
+    @custom_column_generator(required_columns=["seed"], side_effect_columns=["seed_trace"])
+    def gen_with_trace(row: dict[str, Any]) -> dict[str, Any]:
+        return row
+
+    column_configs = [
+        LLMTextColumnConfig(name="seed", prompt="generate seed", model_alias=MODEL_ALIAS),
+        ExpressionColumnConfig(name="consumer", expr="{{ seed_trace }}"),
+        CustomColumnConfig(name="producer", generator_function=gen_with_trace),
+    ]
+    sorted_configs = topologically_sort_column_configs(column_configs)
+    names = [c.name for c in sorted_configs]
+    assert names.index("producer") < names.index("consumer")
+
+
+def test_skip_when_column_ordering() -> None:
+    """A column with skip.when referencing another DAG column is sorted after that column."""
+    from data_designer.config.base import SkipConfig
+
+    column_configs = [
+        LLMTextColumnConfig(name="seed", prompt="generate seed", model_alias=MODEL_ALIAS),
+        LLMTextColumnConfig(
+            name="gated",
+            prompt="generate gated",
+            model_alias=MODEL_ALIAS,
+            skip=SkipConfig(when="{{ seed == 'bad' }}"),
+        ),
+    ]
+    # gated has no required_columns referencing seed, only a skip.when dependency
+    sorted_configs = topologically_sort_column_configs(column_configs)
+    names = [c.name for c in sorted_configs]
+    assert names.index("seed") < names.index("gated")