From c9774e541e0e2b40f41c86bd2dcacc26fe3c0b61 Mon Sep 17 00:00:00 2001
From: Y1fe1Zh0u <zhouyifei210@gmail.com>
Date: Fri, 14 Aug 2026 20:21:10 +0800
Subject: [PATCH] End unrecoverable tool repair loops visibly

A depleted Tool repair budget previously parked the Run in waiting_user, leaving users with an apparently active task and only a plain assistant message. Treat the exhausted budget as a structured terminal failure so the existing run_failed delivery and frontend error paths are used.

Constraint: Historical waiting_user checkpoints for repair exhaustion must remain resumable.

Rejected: Add a special frontend waiting card | the exhausted repair budget is not recoverable without starting a new turn and should release the Run lane.

Confidence: high

Scope-risk: narrow

Directive: Keep repair-budget exhaustion terminal unless a bounded automatic recovery mechanism is introduced.

Tested: 62 Runtime node, checkpoint, chat stream, and repair budget tests; scoped Ruff; git diff --check.

Not-tested: Live WebSocket rendering against the deployed frontend.
---
 .../services/agent_runtime/node_executor.py   | 24 ++++++----------
 .../tests/test_agent_runtime_node_executor.py | 28 ++++++-------------
 2 files changed, 17 insertions(+), 35 deletions(-)

diff --git a/backend/app/services/agent_runtime/node_executor.py b/backend/app/services/agent_runtime/node_executor.py
index bb9d380e7..0414faa11 100644
--- a/backend/app/services/agent_runtime/node_executor.py
+++ b/backend/app/services/agent_runtime/node_executor.py
@@ -1051,24 +1051,16 @@ async def _tool(
             lifecycle.pop("step_tool_context", None)
             lifecycle.update(
                 {
-                    "status": "waiting_user",
-                    "next_route": "wait",
+                    "status": "failed",
+                    "next_route": "terminal",
                     "reason": repair_pause_reason,
                     "pending_tool_calls": [],
-                    "waiting_request": {
-                        "waiting_type": "user",
-                        "correlation_id": _runtime_message_id(
-                            context,
-                            f"tool-repair:{repair_pause_reason}:{repair_pause_tool}",
-                        ),
-                        "reason": repair_pause_reason,
-                        "question": (
-                            f"Tool {repair_pause_tool or 'unknown'} reached its "
-                            "repair safety limit. Provide corrected requirements "
-                            "or arguments to continue."
-                        ),
-                    },
-                    "error": None,
+                    "waiting_request": None,
+                    "error": _error(
+                        repair_pause_reason,
+                        f"Tool {repair_pause_tool or 'unknown'} reached its "
+                        "repair safety limit.",
+                    ),
                 }
             )
         else:
diff --git a/backend/tests/test_agent_runtime_node_executor.py b/backend/tests/test_agent_runtime_node_executor.py
index f47c6ae9f..4d07cdb2c 100644
--- a/backend/tests/test_agent_runtime_node_executor.py
+++ b/backend/tests/test_agent_runtime_node_executor.py
@@ -784,7 +784,7 @@ async def no_sleep(_seconds: float) -> None:
 
 
 @pytest.mark.asyncio
-async def test_tenth_same_tool_failure_pauses_before_next_model_call() -> None:
+async def test_tenth_same_tool_failure_fails_run_before_next_model_call() -> None:
     run_id = uuid.uuid4()
     proposals = tuple(
         ModelStepResult(
@@ -822,10 +822,17 @@ async def test_tenth_same_tool_failure_pauses_before_next_model_call() -> None:
     result = await _invoke(run_id, executor)
 
     lifecycle = result["lifecycle"]
-    assert lifecycle["status"] == "waiting_user"
+    assert lifecycle["status"] == "failed"
+    assert lifecycle["next_route"] == "terminal"
     assert lifecycle["reason"] == (
         "tool_repair_same_fingerprint_limit_reached"
     )
+    assert lifecycle["error"] == {
+        "code": "tool_repair_same_fingerprint_limit_reached",
+        "message": "Tool read_file reached its repair safety limit.",
+    }
+    assert lifecycle["pending_tool_calls"] == []
+    assert lifecycle.get("waiting_request") is None
     assert lifecycle["model_step_count"] == 10
     repair_episode = lifecycle["tool_repair_episodes"]["by_tool"]["read_file"]
     assert repair_episode["total_failures"] == 10
@@ -833,23 +840,6 @@ async def test_tenth_same_tool_failure_pauses_before_next_model_call() -> None:
     assert model.calls == 10
     assert len(tools.calls) == 10
 
-    resumed = await executor.execute(
-        "wait",
-        cast(RuntimeGraphState, result),
-        _context(run_id, executor, "command-user-correction"),
-        resume_value={
-            "resume_type": "user_input",
-            "payload": {"content": "Use README.md as the path."},
-        },
-    )
-    assert resumed["lifecycle"]["tool_repair_episodes"] == {
-        "version": 1,
-        "by_tool": {},
-    }
-    assert resumed["lifecycle"]["tool_repair_reset"]["reason"] == (
-        "explicit_user_correction"
-    )
-
 
 @pytest.mark.asyncio
 async def test_duplicate_tool_call_ids_fail_before_any_provider_execution() -> None: