From 2696824e09d372409bd169db04d13dbaa7f85def Mon Sep 17 00:00:00 2001
From: frankbria <frank.bria@gmail.com>
Date: Tue, 7 Jul 2026 15:42:06 -0700
Subject: [PATCH 1/2] test(tasks): cover server-side execute_agent (BUILD) path
 (#749)

Adds an API-level test that calls POST /tasks/{id}/start?execute=true with
MockProvider injected (CODEFRAME_LLM_PROVIDER=mock), polls the run to a
terminal status, and asserts the agent-dispatch branch (_run_agent ->
runtime.execute_agent) completes and produces output. Joins the endpoint's
fire-and-forget worker thread before teardown to avoid a DB-race on the
temp workspace.

Closes #749
---
 tests/ui/test_v2_routers_integration.py | 63 +++++++++++++++++++++++++
 1 file changed, 63 insertions(+)

diff --git a/tests/ui/test_v2_routers_integration.py b/tests/ui/test_v2_routers_integration.py
index 57458c80..3521b113 100644
--- a/tests/ui/test_v2_routers_integration.py
+++ b/tests/ui/test_v2_routers_integration.py
@@ -709,6 +709,69 @@ def test_start_task_not_found(self, test_client):
 
         assert response.status_code == 404
 
+    def test_start_task_execute_runs_agent_to_terminal(
+        self, test_client_with_task, monkeypatch
+    ):
+        """Start with execute=true dispatches the agent (BUILD path) to a terminal run.
+
+        Covers the agent-dispatch branch (`_run_agent` -> `runtime.execute_agent`),
+        which the other `/start` tests skip by omitting `execute=true` (#749).
+
+        MockProvider is injected via CODEFRAME_LLM_PROVIDER=mock: the router builds
+        it through `get_provider("mock")`, and its text-only response makes the
+        ReAct agent complete on the first iteration (react_agent.py:518).
+        """
+        import threading
+        import time
+
+        from codeframe.core import runtime, streaming
+
+        monkeypatch.setenv("CODEFRAME_LLM_PROVIDER", "mock")
+
+        task_id = test_client_with_task.task.id
+        workspace = test_client_with_task.workspace
+
+        # Snapshot threads so we can join the endpoint's fire-and-forget worker
+        # before the fixture deletes the temp workspace (avoids a teardown race
+        # where the still-draining thread touches a removed DB).
+        threads_before = set(threading.enumerate())
+
+        response = test_client_with_task.post(
+            f"/api/v2/tasks/{task_id}/start?execute=true"
+        )
+        assert response.status_code == 200
+        data = response.json()
+        assert data["success"] is True
+        assert data["status"] == "executing"
+        run_id = data["run_id"]
+
+        # execute_agent runs in a background daemon thread; poll the run until it
+        # leaves RUNNING (terminal = COMPLETED / FAILED / BLOCKED).
+        deadline = time.monotonic() + 30.0
+        status = "RUNNING"
+        while time.monotonic() < deadline:
+            run_resp = test_client_with_task.get(f"/api/v2/tasks/{task_id}/run")
+            assert run_resp.status_code == 200
+            status = run_resp.json()["status"]
+            if status != runtime.RunStatus.RUNNING.value:
+                break
+            time.sleep(0.2)
+
+        # Let the worker thread finish its tail work (logging/recording) before
+        # assertions and fixture teardown remove the workspace out from under it.
+        for thread in set(threading.enumerate()) - threads_before:
+            thread.join(timeout=10.0)
+
+        assert status != runtime.RunStatus.RUNNING.value, (
+            f"Run {run_id} never reached a terminal status (stuck at {status})"
+        )
+        assert status == runtime.RunStatus.COMPLETED.value, (
+            f"Expected COMPLETED for the mock-driven BUILD path, got {status}"
+        )
+
+        # The BUILD path produced output events (the whole point of dispatching).
+        assert streaming.run_output_exists(workspace, run_id) is True
+
     def test_get_task_run(self, test_client_with_task):
         """Get task run status after starting."""
         task_id = test_client_with_task.task.id

From 4ca329bee2bd93bb427afb6bb53d5dc510a70ef4 Mon Sep 17 00:00:00 2001
From: frankbria <frank.bria@gmail.com>
Date: Tue, 7 Jul 2026 15:49:53 -0700
Subject: [PATCH 2/2] test(tasks): assert output file has content, not just
 existence (#749)

Address claude-review: run_output_exists() is only path.exists(), and
RunOutputLogger creates the file at execute_agent start, so existence only
proved the run started. Assert st_size > 0 to prove the BUILD path actually
emitted output.
---
 tests/ui/test_v2_routers_integration.py | 8 ++++++--
 1 file changed, 6 insertions(+), 2 deletions(-)

diff --git a/tests/ui/test_v2_routers_integration.py b/tests/ui/test_v2_routers_integration.py
index 3521b113..28c25e3a 100644
--- a/tests/ui/test_v2_routers_integration.py
+++ b/tests/ui/test_v2_routers_integration.py
@@ -769,8 +769,12 @@ def test_start_task_execute_runs_agent_to_terminal(
             f"Expected COMPLETED for the mock-driven BUILD path, got {status}"
         )
 
-        # The BUILD path produced output events (the whole point of dispatching).
-        assert streaming.run_output_exists(workspace, run_id) is True
+        # The BUILD path emitted output events (the whole point of dispatching).
+        # Assert on file *content*, not mere existence: RunOutputLogger creates the
+        # output file at execute_agent start, so path.exists() only proves the run
+        # started — a non-empty file proves the agent actually wrote output.
+        output_path = streaming.get_run_output_path(workspace, run_id)
+        assert output_path.stat().st_size > 0, "BUILD path produced no output"
 
     def test_get_task_run(self, test_client_with_task):
         """Get task run status after starting."""