From 6e700be30a343e5600d28cc8952b06d445b3a313 Mon Sep 17 00:00:00 2001
From: Wangmerlyn <wsy0227@sjtu.edu.cn>
Date: Thu, 2 Jul 2026 08:57:52 +0000
Subject: [PATCH] fix(cli): include custom endpoint in start hints

---
 AGENTS.md                          |  2 ++
 docs/guides/cli.md                 |  3 +++
 src/keep_gpu/cli.py                | 22 +++++++++++++++++++---
 tests/test_cli_service_commands.py | 20 ++++++++++++++++++++
 4 files changed, 44 insertions(+), 3 deletions(-)

diff --git a/AGENTS.md b/AGENTS.md
index 9ec3363..0ee8ea1 100644
--- a/AGENTS.md
+++ b/AGENTS.md
@@ -256,6 +256,8 @@ This file defines how coding agents should work in this repository.
   before service RPC, daemon auto-start, stop-all fallback, or daemon ownership
   operations. JSON-output commands must return structured `{"error": "..."}`
   objects for invalid endpoints, not tracebacks.
+- `keep-gpu start` follow-up status/stop/service-stop hints must target the
+  selected service endpoint while keeping default-endpoint hints concise.
 - Shared public endpoint validation belongs in
   `src/keep_gpu/utilities/endpoint_validation.py`; CLI and MCP entry points
   should only translate those `ValueError`s into their interface-specific error
diff --git a/docs/guides/cli.md b/docs/guides/cli.md
index 878d446..495bbfc 100644
--- a/docs/guides/cli.md
+++ b/docs/guides/cli.md
@@ -28,6 +28,9 @@ keep-gpu start --gpu-ids 0 --vram 1GiB --interval 60 --busy-threshold 25
 - follow-up status/stop command hints,
 - daemon shutdown hint (`keep-gpu service-stop`).
 
+When `--host` or `--port` selects a non-default endpoint, the printed
+status/stop/service-stop hints include the same endpoint flags.
+
 If this invocation auto-starts the service but session creation fails with an
 expected startup-unavailable error before a session is created, `start`
 best-effort stops the just-created daemon so it does not remain idle.
diff --git a/src/keep_gpu/cli.py b/src/keep_gpu/cli.py
index 11a6806..cb0e07a 100644
--- a/src/keep_gpu/cli.py
+++ b/src/keep_gpu/cli.py
@@ -135,6 +135,17 @@ def _service_command(host: str, port: int) -> List[str]:
     ]
 
 
+def _service_hint_args(host: str, port: int) -> List[str]:
+    if host == DEFAULT_SERVICE_HOST and port == DEFAULT_SERVICE_PORT:
+        return []
+    return ["--host", host, "--port", str(port)]
+
+
+def _keep_gpu_hint_command(command: str, host: str, port: int, *args: str) -> str:
+    parts = ["keep-gpu", command, *_service_hint_args(host, port), *args]
+    return " ".join(shlex.quote(part) for part in parts)
+
+
 def _print_machine_json(data: Dict[str, Any]) -> None:
     console.file.write(f"{json.dumps(data, indent=2, allow_nan=False)}\n")
     console.file.flush()
@@ -1295,11 +1306,16 @@ def start(
             f"[bold green]Started keep session[/bold green] job_id={result_job_id}"
         )
         console.print(f"[cyan]Dashboard:[/cyan] http://{host}:{port}/")
-        console.print(
-            f"[dim]Next: keep-gpu status --job-id {result_job_id} | keep-gpu stop --job-id {result_job_id}[/dim]"
+        status_command = _keep_gpu_hint_command(
+            "status", host, port, "--job-id", result_job_id
+        )
+        stop_command = _keep_gpu_hint_command(
+            "stop", host, port, "--job-id", result_job_id
         )
+        service_stop_command = _keep_gpu_hint_command("service-stop", host, port)
+        console.print(f"[dim]Next: {status_command} | {stop_command}[/dim]")
         console.print(
-            "[dim]When all sessions are done, stop daemon with: keep-gpu service-stop[/dim]"
+            f"[dim]When all sessions are done, stop daemon with: {service_stop_command}[/dim]"
         )
     except ServiceRPCError as exc:
         _rollback_auto_started_service_on_startup_unavailable(
diff --git a/tests/test_cli_service_commands.py b/tests/test_cli_service_commands.py
index 0e276e2..30bf831 100644
--- a/tests/test_cli_service_commands.py
+++ b/tests/test_cli_service_commands.py
@@ -2388,6 +2388,26 @@ def fake_rpc(method, params, host, port):
     assert "keep-gpu service-stop" in result.output
 
 
+def test_start_prints_custom_endpoint_in_follow_up_hints(monkeypatch):
+    monkeypatch.setattr(cli, "_ensure_service_running", lambda *args, **kwargs: True)
+    monkeypatch.setattr(cli, "_rpc_call", lambda *args, **kwargs: {"job_id": "job-abc"})
+
+    result = runner.invoke(cli.app, ["start", "--host", "localhost", "--port", "9999"])
+
+    normalized_output = " ".join(result.output.split())
+    assert result.exit_code == 0
+    assert "Dashboard: http://localhost:9999/" in normalized_output
+    assert (
+        "keep-gpu status --host localhost --port 9999 --job-id job-abc"
+        in normalized_output
+    )
+    assert (
+        "keep-gpu stop --host localhost --port 9999 --job-id job-abc"
+        in normalized_output
+    )
+    assert "keep-gpu service-stop --host localhost --port 9999" in normalized_output
+
+
 def test_start_rolls_back_auto_started_service_on_startup_unavailable(monkeypatch):
     stopped = []