Add tool failure recovery decisions

tests/test_tools_flow.py

@@ -1,7 +1,7 @@
 import json
 from pathlib import Path
 
-from app.core.contracts import UserTask
+from app.core.contracts import ExecutionDirective, UserTask
 from app.runtime.runtime_controller import RuntimeController
 
 
@@ -112,21 +112,37 @@ def test_shell_exec_allows_safe_command(tmp_path: Path) -> None:
     assert str(tmp_path) in result["result"]["output"]
 
 
-def test_shell_exec_treats_grep_no_matches_as_information(tmp_path: Path) -> None:
+class _RecoveryCritic:
+    async def generate(self, prompt: str, max_tokens: int | None = None) -> str:
+        return '{"action":"continue","reason":"No matches is acceptable information for this exploratory check."}'
+
+
+def test_failed_shell_step_can_recover_and_continue(tmp_path: Path) -> None:
     _write_config_tree(tmp_path)
     controller = RuntimeController(base_dir=tmp_path)
-    result = controller.handle_task(
+    controller.execution_engine.set_critic(_RecoveryCritic())
+    controller.execution_engine._recovery_limit = 1
+    result = controller.execution_engine.execute(
         UserTask(
-            input="run grep with no matches",
-            context={
-                "requested_tool": "shell_exec",
-                "tool_args": {"command": "printf 'abc\\n' | grep definitely_missing"},
+            input="run grep with no matches and recover",
+        ),
+        ExecutionDirective(
+            type="plan",
+            payload={
+                "steps": [
+                    {
+                        "id": "1",
+                        "tool": "shell_exec",
+                        "args": {"command": "printf 'abc\\n' | grep definitely_missing"},
+                        "depends_on": [],
+                    }
+                ]
             },
-        )
+        ),
     )
     assert result["status"] == "completed"
-    assert result["result"]["metadata"]["exit_code"] == 1
-    assert result["result"]["metadata"]["no_matches"] is True
+    failed_result = result["result"]["step_results"][0]["result"]["result"]
+    assert failed_result["metadata"]["exit_code"] == 1
 
 
 def test_permission_resolution_can_resume_task(tmp_path: Path) -> None: