feat(p1): persist workflows and semantic learning artifacts

agent_v0/server_v1/workflow_replay.py

@@ -126,6 +126,25 @@ def build_workflow_replay(
                             "x_relative": "",
                         },
                     }
+                    _merge_semantic_target_fields(
+                        step_action["target_spec"],
+                        target,
+                        params,
+                        step,
+                    )
+                    target_label = _first_non_empty_text(
+                        step_action["target_spec"].get("by_text"),
+                        step_action["target_spec"].get("target_text"),
+                        step_action["target_spec"].get("description"),
+                        step_action["target_spec"].get("ocr_description"),
+                        step_action["target_spec"].get("vlm_description"),
+                    )
+                    if target_label:
+                        step_action.setdefault(
+                            "target_text",
+                            step_action["target_spec"].get("target_text") or target_label,
+                        )
+                        step_action.setdefault("target_description", target_label)
                     # Ajouter le crop anchor si disponible
                     _attach_anchor(step_action, step, session_dir)
 
@@ -171,6 +190,58 @@ def _map_action_type(step_type: str) -> str:
     return mapping.get(step_type, step_type)
 
 
+_TARGET_SEMANTIC_KEYS = (
+    "by_text",
+    "by_role",
+    "anchor_id",
+    "target_text",
+    "ocr_description",
+    "description",
+    "vlm_description",
+    "by_text_source",
+    "anchor_bbox",
+    "original_size",
+)
+
+
+def _first_non_empty_text(*values: Any) -> str:
+    for value in values:
+        text = str(value or "").strip()
+        if text and text.casefold() not in {"none", "null"}:
+            return text
+    return ""
+
+
+def _merge_semantic_target_fields(
+    target_spec: Dict[str, Any],
+    *sources: Dict[str, Any],
+) -> None:
+    for source in sources:
+        if not isinstance(source, dict):
+            continue
+        visual_anchor = source.get("visual_anchor") or {}
+        if isinstance(visual_anchor, dict):
+            _merge_semantic_target_fields(target_spec, visual_anchor)
+        for key in _TARGET_SEMANTIC_KEYS:
+            value = source.get(key)
+            if value and not target_spec.get(key):
+                target_spec[key] = value
+
+    if not target_spec.get("by_text"):
+        target_text = _first_non_empty_text(target_spec.get("target_text"))
+        if target_text:
+            target_spec["by_text"] = target_text
+            target_spec.setdefault("by_text_source", "visual_anchor")
+
+    if not target_spec.get("vlm_description"):
+        description = _first_non_empty_text(
+            target_spec.get("description"),
+            target_spec.get("ocr_description"),
+        )
+        if description:
+            target_spec["vlm_description"] = description
+
+
 def _attach_anchor(action: dict, step: dict, session_dir: str) -> None:
     """Attacher le crop anchor au target_spec si disponible."""
     import base64