diff --git a/src/activity_core/rules/executor.py b/src/activity_core/rules/executor.py
index 7c6894c..a9ecf0a 100644
--- a/src/activity_core/rules/executor.py
+++ b/src/activity_core/rules/executor.py
@@ -180,7 +180,10 @@ def _llm_run_config(instr: Any) -> dict[str, Any]:
         value = getattr(instr, field, None)
         if value is not None:
             config[field] = value
-    model_params = getattr(instr, "model_params", None)
+    model_params = dict(getattr(instr, "model_params", None) or {})
+    schema = _load_output_schema(getattr(instr, "output_schema", ""))
+    if schema is not None:
+        model_params.setdefault("json_schema", schema)
     if model_params:
         config["model_params"] = model_params
     return config
@@ -263,16 +266,33 @@ def _validate_against_schema(data: Any, schema_path: str) -> str | None:
     if not schema_path:
         return None
 
+    try:
+        schema = _load_output_schema(schema_path)
+    except (OSError, json.JSONDecodeError, TypeError) as exc:
+        return f"could not read output schema: {exc}"
+    if schema is None:
+        return None
+
+    return _validate_schema_node(data, schema, "$")
+
+
+def _load_output_schema(schema_path: str) -> dict[str, Any] | None:
+    """Load a JSON schema file when present.
+
+    Missing schema files are intentionally tolerated for backward
+    compatibility with existing tests and definitions.
+    """
+    if not schema_path:
+        return None
+
     path = Path(schema_path)
     if not path.exists():
         return None
 
-    try:
-        schema = json.loads(path.read_text(encoding="utf-8"))
-    except (OSError, json.JSONDecodeError) as exc:
-        return f"could not read output schema: {exc}"
-
-    return _validate_schema_node(data, schema, "$")
+    schema = json.loads(path.read_text(encoding="utf-8"))
+    if not isinstance(schema, dict):
+        raise TypeError("output schema must be a JSON object")
+    return schema
 
 
 def _validate_schema_node(data: Any, schema: dict[str, Any], path: str) -> str | None:
diff --git a/tests/rules/test_executor.py b/tests/rules/test_executor.py
index 558a68a..aa73b7d 100644
--- a/tests/rules/test_executor.py
+++ b/tests/rules/test_executor.py
@@ -275,6 +275,46 @@ def test_execute_instruction_forwards_llm_connect_run_config():
     ]
 
 
+def test_execute_instruction_forwards_output_schema_to_llm_connect(tmp_path, monkeypatch):
+    schema_dir = tmp_path / "schemas"
+    schema_dir.mkdir()
+    schema_path = schema_dir / "daily-triage-report.json"
+    schema = {
+        "type": "object",
+        "required": ["summary", "recommendations"],
+        "properties": {
+            "summary": {"type": "string"},
+            "recommendations": {"type": "array", "items": {"type": "object"}},
+        },
+    }
+    schema_path.write_text(json.dumps(schema), encoding="utf-8")
+    monkeypatch.chdir(tmp_path)
+
+    llm = _CountingLLM([
+        json.dumps({"summary": "Review open work.", "recommendations": []})
+    ])
+    instr = _instr(
+        id="daily-triage-report",
+        prompt="Report.",
+        trusted_fields=[],
+        output_schema="schemas/daily-triage-report.json",
+        model_params={"reasoning_effort": "medium"},
+    )
+
+    result = execute_instruction_with_audit(instr, _Event(), {}, llm)
+
+    assert result.output_validated is True
+    assert llm.calls == [
+        {
+            "model_name": "claude-sonnet-4-6",
+            "model_params": {
+                "reasoning_effort": "medium",
+                "json_schema": schema,
+            },
+        }
+    ]
+
+
 def test_execute_instruction_with_audit_accepts_report_payload():
     report_data = {
         "summary": "State Hub has loose ends.",
diff --git a/tests/test_instruction_evaluation.py b/tests/test_instruction_evaluation.py
index 41c3e9e..a32cf52 100644
--- a/tests/test_instruction_evaluation.py
+++ b/tests/test_instruction_evaluation.py
@@ -157,5 +157,18 @@ async def test_evaluate_instructions_forwards_llm_connect_depth_config(monkeypat
         "temperature": 0.2,
         "max_tokens": 1200,
         "max_depth": 2,
-        "model_params": {"reasoning_effort": "medium"},
+        "model_params": {
+            "reasoning_effort": "medium",
+            "json_schema": {
+                "type": "object",
+                "required": ["summary", "recommendations"],
+                "properties": {
+                    "summary": {"type": "string"},
+                    "recommendations": {
+                        "type": "array",
+                        "items": {"type": "object"},
+                    },
+                },
+            },
+        },
     }