feat(graph): accept the caller's portfolio as run input (#1304, #1166)

- PortfolioContext input, rendered once into state and read by the trader, risk and portfolio agents
- --portfolio JSON file on the CLI
- a run without it is never treated as a flat book
- the checkpoint signature keys on the portfolio

tests/test_cli_decision_log.py

@@ -69,7 +69,7 @@ class _FakeGraph:
         self.graph = self
         self.propagator = self
 
-    def create_run_state(self, ticker, trade_date, asset_type="stock"):
+    def create_run_state(self, ticker, trade_date, asset_type="stock", portfolio=None):
         self.calls.append(("create_run_state", ticker, trade_date))
         return {"messages": [], "company_of_interest": ticker}
 

tests/test_portfolio_context.py

@@ -0,0 +1,240 @@
+"""Portfolio context: what the caller holds, threaded into the decision agents.
+
+Decisions were made with no knowledge of the current book, so "add to a full
+position" and "open a new one" read alike. The context is optional and carries
+three distinct states: a position, a flat book, and no context at all. Nothing
+may present the third as the second. The research team stays blind so the bull
+and bear cases are not anchored by the caller's position.
+"""
+
+from __future__ import annotations
+
+import json
+
+import pytest
+
+from tradingagents.agents.utils.agent_utils import get_portfolio_context_from_state
+from tradingagents.portfolio import PortfolioContext, load_portfolio
+
+HOLDING = {
+    "cash": 25000.0,
+    "currency": "USD",
+    "positions": [
+        {"ticker": "AAPL", "quantity": 120, "average_price": 150.0},
+        {"ticker": "MSFT", "quantity": 10},
+    ],
+}
+
+
+@pytest.mark.unit
+def test_position_in_the_analyzed_instrument_leads_the_render():
+    text = PortfolioContext.model_validate(HOLDING).render("AAPL")
+    assert "120" in text and "150" in text
+    assert "MSFT" in text and "25,000" in text and "USD" in text
+
+
+@pytest.mark.unit
+def test_flat_book_says_no_position_rather_than_omitting_it():
+    text = PortfolioContext.model_validate({"cash": 1000.0, "positions": []}).render("AAPL")
+    assert "No current position in AAPL" in text
+
+
+@pytest.mark.unit
+def test_a_ticker_held_under_another_spelling_is_matched():
+    text = PortfolioContext.model_validate({"positions": [{"ticker": "aapl", "quantity": 5}]}).render("AAPL")
+    assert "No current position" not in text and "5" in text
+
+
+@pytest.mark.unit
+def test_absent_context_is_reported_as_not_provided():
+    notice = get_portfolio_context_from_state({"company_of_interest": "AAPL"})
+    assert "not provided" in notice.lower()
+    assert "no position" not in notice.lower()  # missing must not read as flat
+
+
+@pytest.mark.unit
+def test_rendered_context_reaches_the_agents_from_state():
+    block = get_portfolio_context_from_state({"portfolio_context": "Portfolio: flat", "company_of_interest": "AAPL"})
+    assert block == "Portfolio: flat"
+
+
+@pytest.mark.unit
+def test_load_rejects_a_malformed_file_with_a_clear_error(tmp_path):
+    bad = tmp_path / "p.json"
+    bad.write_text(json.dumps({"positions": [{"quantity": 5}]}))
+    with pytest.raises(ValueError, match="portfolio"):
+        load_portfolio(bad)
+
+
+@pytest.mark.unit
+def test_load_reads_a_valid_file(tmp_path):
+    good = tmp_path / "p.json"
+    good.write_text(json.dumps(HOLDING))
+    assert load_portfolio(good).positions[0].ticker == "AAPL"
+
+
+# --- threading through the graph --------------------------------------------
+
+def _bare_graph(tmp_path):
+    from tradingagents.agents.utils.memory import TradingMemoryLog
+    from tradingagents.graph.propagation import Propagator
+    from tradingagents.graph.trading_graph import TradingAgentsGraph
+
+    graph = object.__new__(TradingAgentsGraph)
+    graph.config = {"memory_log_path": str(tmp_path / "m.md"), "max_debate_rounds": 1,
+                    "max_risk_discuss_rounds": 1}
+    graph.memory_log = TradingMemoryLog(graph.config)
+    graph.propagator = Propagator()
+    graph.selected_analysts = ["market"]
+    graph._resolve_pending_entries = lambda t: None
+    graph.resolve_instrument_context = lambda t, a="stock": ""
+    graph._memory_as_of = lambda d: None
+    return graph
+
+
+@pytest.mark.unit
+def test_create_run_state_renders_the_portfolio_once(tmp_path):
+    graph = _bare_graph(tmp_path)
+    state = graph.create_run_state("AAPL", "2026-08-14", portfolio=PortfolioContext.model_validate(HOLDING))
+    assert "120" in state["portfolio_context"]
+    assert graph.create_run_state("AAPL", "2026-08-14")["portfolio_context"] == ""
+
+
+@pytest.mark.unit
+def test_checkpoint_signature_changes_with_the_portfolio(tmp_path):
+    graph = _bare_graph(tmp_path)
+    none = graph._run_signature("stock")
+    flat = graph._run_signature("stock", PortfolioContext())
+    held = graph._run_signature("stock", PortfolioContext.model_validate(HOLDING))
+    assert len({none, flat, held}) == 3
+
+
+@pytest.mark.unit
+@pytest.mark.parametrize("module, factory", [
+    ("tradingagents.agents.trader.trader", "create_trader"),
+    ("tradingagents.agents.managers.portfolio_manager", "create_portfolio_manager"),
+    ("tradingagents.agents.risk_mgmt.aggressive_debator", "create_aggressive_debator"),
+    ("tradingagents.agents.risk_mgmt.conservative_debator", "create_conservative_debator"),
+    ("tradingagents.agents.risk_mgmt.neutral_debator", "create_neutral_debator"),
+])
+def test_decision_agents_see_the_portfolio(module, factory, monkeypatch):
+    """The prompt each decision agent sends carries the portfolio block."""
+    import importlib
+
+    mod = importlib.import_module(module)
+    seen = []
+
+    class _LLM:
+        def invoke(self, prompt, *a, **k):
+            seen.append(prompt if isinstance(prompt, str) else json.dumps(str(prompt)))
+            from langchain_core.messages import AIMessage
+            return AIMessage("Rating: Hold\n\nnothing to do")
+
+        def with_structured_output(self, *a, **k):
+            raise NotImplementedError  # force the free-text path
+
+    state = {
+        "company_of_interest": "AAPL", "trade_date": "2026-08-14", "asset_type": "stock",
+        "instrument_context": "", "market_report": "M", "sentiment_report": "S",
+        "news_report": "N", "fundamentals_report": "F", "investment_plan": "P",
+        "trader_investment_plan": "T", "past_context": "",
+        "portfolio_context": "PORTFOLIO_BLOCK_MARKER",
+        "investment_debate_state": {"history": "", "judge_decision": "", "count": 0},
+        "risk_debate_state": {"history": "", "latest_speaker": "", "count": 0,
+                              "aggressive_history": "", "conservative_history": "", "neutral_history": "",
+                              "current_aggressive_response": "", "current_conservative_response": "",
+                              "current_neutral_response": "", "judge_decision": ""},
+    }
+    node = getattr(mod, factory)(_LLM())
+    node(state)
+    assert any("PORTFOLIO_BLOCK_MARKER" in p for p in seen), f"{factory} prompt lacks the portfolio block"
+
+
+@pytest.mark.unit
+def test_research_team_stays_blind_to_the_portfolio():
+    import inspect
+
+    from tradingagents.agents.researchers import bear_researcher, bull_researcher
+    for mod in (bull_researcher, bear_researcher):
+        assert "portfolio_context" not in inspect.getsource(mod)
+
+
+@pytest.mark.unit
+def test_completed_run_clears_the_checkpoint_it_wrote(tmp_path, monkeypatch):
+    """The clear must key on the same portfolio the run was checkpointed under.
+
+    Keyed on a different one it deletes nothing, and the next identical call
+    resumes the finished thread and returns the old decision without running.
+    """
+    import tradingagents.graph.trading_graph as tg
+
+    graph = _bare_graph(tmp_path)
+    graph.config.update({"checkpoint_enabled": True, "data_cache_dir": str(tmp_path),
+                         "results_dir": str(tmp_path)})
+    graph.debug = False
+    graph._resuming = False
+    graph.propagator.get_graph_args = lambda callbacks=None: {}
+    graph.process_signal = lambda d: "Hold"
+    graph._log_state = lambda *a, **k: None
+    graph.graph = type("G", (), {"invoke": lambda self, i, **k: {"final_trade_decision": "Rating: Hold\n\nx"}})()
+    book = PortfolioContext.model_validate(HOLDING)
+
+    written = graph._run_signature("stock", book)  # what begin_checkpoint keys on
+    cleared = []
+    monkeypatch.setattr(tg, "clear_checkpoint", lambda d, t, dt, signature: cleared.append(signature))
+
+    graph._run_graph("AAPL", "2026-08-14", "stock", checkpoint_thread_id=None, portfolio=book)
+
+    assert cleared == [written]
+
+
+@pytest.mark.unit
+def test_research_layer_sizes_against_a_standard_allocation():
+    """The research team is blind to the book, so its plan cannot promise
+    position-relative sizing: it sizes against a standard allocation instead."""
+    from tradingagents.agents.schemas import ResearchPlan
+
+    description = ResearchPlan.model_fields["strategic_actions"].description
+    assert "standard allocation" in description
+    assert "does not see the caller's holdings" in description
+
+
+@pytest.mark.unit
+def test_partial_portfolio_states_only_what_it_was_given():
+    """Cash omitted is not cash zero; the line is absent rather than invented."""
+    text = PortfolioContext.model_validate({"positions": [{"ticker": "AAPL", "quantity": 5}]}).render("AAPL")
+    assert "Cash" not in text
+    assert "5" in text
+
+
+@pytest.mark.unit
+def test_cli_rejects_an_unusable_portfolio_file_before_running(tmp_path, monkeypatch):
+    from typer.testing import CliRunner
+
+    import cli.main as m
+
+    bad = tmp_path / "bad.json"
+    bad.write_text('{"positions": [{"quantity": 5}]}')
+    ran = []
+    monkeypatch.setattr(m, "run_analysis", lambda **k: ran.append(k))
+
+    result = CliRunner().invoke(m.app, ["--portfolio", str(bad)])
+
+    assert result.exit_code == 1 and ran == []
+
+
+@pytest.mark.unit
+def test_cli_passes_a_valid_portfolio_into_the_run(tmp_path, monkeypatch):
+    from typer.testing import CliRunner
+
+    import cli.main as m
+
+    good = tmp_path / "good.json"
+    good.write_text(json.dumps(HOLDING))
+    ran = []
+    monkeypatch.setattr(m, "run_analysis", lambda **k: ran.append(k))
+
+    result = CliRunner().invoke(m.app, ["--portfolio", str(good)])
+
+    assert result.exit_code == 0
+    assert ran[0]["portfolio"].position_in("AAPL").quantity == 120