fix(cli): finish the run surface

- backtest takes --run-id, so an interrupted sweep continues
- an inverted date range or an empty ticker list is an error, not a clean zero-cell run
- a setup failure in a sweep reports one line instead of a traceback
- the remembered endpoint URL is offered back
- a round count set in the environment says it overrode the chosen research depth
- the run directory validates the ticker, and the report saves under results_dir
- the run says whether it resumed a saved run or started fresh

tests/test_cli_commands.py

@@ -89,3 +89,61 @@ def test_every_command_is_registered_when_run_as_a_module():
                          capture_output=True, text=True, timeout=120)
     assert out.returncode == 0, out.stderr[-400:]
     assert "--start" in out.stdout
+
+
+@pytest.mark.unit
+def test_the_cli_says_whether_a_run_resumed(monkeypatch):
+    """The README promises the user can tell a resumed run from a fresh one.
+    The graph logs it, but nothing configures logging, so it was never shown."""
+    import cli.main as m
+
+    messages = []
+    monkeypatch.setattr(m.message_buffer, "add_message",
+                        lambda kind, text: messages.append(text), raising=False)
+
+    m._announce_checkpoint_state(type("G", (), {"_resuming": True})(), "NVDA", "2026-01-10")
+    m._announce_checkpoint_state(type("G", (), {"_resuming": False})(), "NVDA", "2026-01-10")
+
+    assert any("resum" in text.lower() for text in messages)
+    assert any("fresh" in text.lower() for text in messages)
+
+
+@pytest.mark.unit
+def test_backtest_can_continue_an_interrupted_sweep(runner, monkeypatch, tmp_path):
+    """Resuming is what makes a long sweep practical, and the Python API has it."""
+    swept = []
+    monkeypatch.setattr(m, "run_backtest", lambda *a, **kw: swept.append(kw) or _Result(tmp_path))
+    monkeypatch.setattr(m, "summarize", lambda log: _Summary())
+
+    result = runner.invoke(m.app, ["backtest", "NVDA", "--start", "2026-06-01",
+                                   "--end", "2026-06-08", "--run-id", "20260617_120000"])
+
+    assert result.exit_code == 0, result.output
+    assert swept[0]["run_id"] == "20260617_120000"
+
+
+@pytest.mark.unit
+@pytest.mark.parametrize("args, expected", [
+    (["backtest", "NVDA", "--start", "2026-08-01", "--end", "2026-06-08"], "before"),
+    (["backtest", ",,", "--start", "2026-06-01", "--end", "2026-06-08"], "ticker"),
+])
+def test_backtest_rejects_input_that_would_sweep_nothing(runner, args, expected):
+    """An inverted range or an empty ticker list reported a clean zero-cell run,
+    which reads as 'nothing to find' rather than 'you asked for nothing'."""
+    result = runner.invoke(m.app, args)
+    assert result.exit_code == 1
+    assert expected in result.output.lower()
+
+
+@pytest.mark.unit
+def test_backtest_reports_a_setup_failure_in_one_line(runner, monkeypatch):
+    """A missing key or a bad analyst name produced a raw traceback."""
+    def _explode(*a, **kw):
+        raise ValueError("API key for provider 'openai' is not set")
+
+    monkeypatch.setattr(m, "run_backtest", _explode)
+    result = runner.invoke(m.app, ["backtest", "NVDA", "--start", "2026-06-01", "--end", "2026-06-08"])
+
+    assert result.exit_code == 1
+    assert "API key" in result.output
+    assert "Traceback" not in result.output

tests/test_cli_config_precedence.py

@@ -83,3 +83,25 @@ def test_glm_resolves_to_the_endpoint_its_key_belongs_to():
     assert get_api_key_env("glm") == "ZHIPU_API_KEY"
     assert "z.ai" in OPENAI_COMPATIBLE_PROVIDERS["glm"].base_url
     assert "bigmodel.cn" in OPENAI_COMPATIBLE_PROVIDERS["glm-cn"].base_url
+
+
+@pytest.mark.unit
+def test_a_half_set_round_count_says_which_value_won(capsys, monkeypatch):
+    """With only one of the two round-count variables set, the depth prompt is
+    still shown but half the answer is discarded; the user was never told."""
+    import cli.main as m
+
+    monkeypatch.setenv("TRADINGAGENTS_MAX_DEBATE_ROUNDS", "1")
+    monkeypatch.delenv("TRADINGAGENTS_MAX_RISK_ROUNDS", raising=False)
+    printed = []
+    monkeypatch.setattr(m.console, "print", lambda *a, **k: printed.append(str(a[0]) if a else ""))
+
+    config = m._build_run_config({
+        "ticker": "NVDA", "analysis_date": "2026-09-01", "asset_type": "stock",
+        "analysts": [], "research_depth": 5, "llm_provider": "openai",
+        "quick_think_llm": "gpt-5.6-luna", "deep_think_llm": "gpt-5.6",
+        "backend_url": None, "output_language": "English",
+    }, None)
+
+    assert config["max_risk_discuss_rounds"] == 5
+    assert any("TRADINGAGENTS_MAX_DEBATE_ROUNDS" in line for line in printed), printed

tests/test_cli_prefs.py

@@ -141,3 +141,29 @@ def test_a_custom_language_is_remembered_without_breaking_the_next_run():
         select.return_value.ask.return_value = "English"
         ask_output_language(load_last_run()["output_language"])
     assert select.call_args.kwargs["default"] is None
+
+
+@pytest.mark.unit
+def test_a_remembered_endpoint_is_offered_back(monkeypatch):
+    """Users of a local or custom endpoint retyped the URL every run: it was
+    remembered and validated, then never read."""
+    import cli.main as m
+
+    save_last_run({"llm_provider": "openai_compatible", "backend_url": "http://localhost:1234/v1"})
+    offered = {}
+    monkeypatch.setattr(m, "select_llm_provider", lambda default=None: ("openai_compatible", None))
+    monkeypatch.setattr(m, "prompt_openai_compatible_url",
+                        lambda default=None: offered.setdefault("default", default) or "http://x/v1")
+    monkeypatch.setattr(m, "fetch_announcements", lambda: [])
+    monkeypatch.setattr(m, "display_announcements", lambda *a: None)
+    monkeypatch.setattr(m, "get_ticker", lambda: "NVDA")
+    monkeypatch.setattr(m, "get_analysis_date", lambda: "2026-09-01")
+    monkeypatch.setattr(m, "ask_output_language", lambda default=None: "English")
+    monkeypatch.setattr(m, "select_analysts", lambda asset_type, default=None: [AnalystType.MARKET])
+    monkeypatch.setattr(m, "select_research_depth", lambda default=None: 1)
+    monkeypatch.setattr(m, "select_shallow_thinking_agent", lambda p, default=None: "local-model")
+    monkeypatch.setattr(m, "select_deep_thinking_agent", lambda p, default=None: "local-model")
+
+    m.get_user_selections()
+
+    assert offered["default"] == "http://localhost:1234/v1"

tests/test_cli_symbol_handling.py

@@ -60,3 +60,16 @@ def test_cli_normalize_delegates_to_data_layer():
     # CLI must produce the same canonical symbol the data path will price.
     for raw in ("XAUUSD", "BTCUSD", "btc-usdt", "AAPL"):
         assert normalize_ticker_symbol(raw) == normalize_symbol(raw)
+
+
+@pytest.mark.unit
+def test_the_run_directory_cannot_escape_the_results_directory(tmp_path, monkeypatch):
+    """Every other path that interpolates a ticker validates it first; the CLI's
+    own results tree did not, so a ticker of '..' wrote a level up."""
+    import cli.main as m
+
+    with pytest.raises(ValueError):
+        m._run_directory({"results_dir": str(tmp_path)}, "..", "2026-09-01")
+
+    ok = m._run_directory({"results_dir": str(tmp_path)}, "NVDA", "2026-09-01")
+    assert str(ok).startswith(str(tmp_path))