Revert "feat: add replay server and web client"

This reverts commit 3c027eae0b.

.gitignore

@@ -32,5 +32,12 @@ htmlcov/
 # IDE and editor files
 .idea/
 .vscode/
+.codex/
 *.swp
 *.swo
+
+# debug resources
+debug/
+
+# Node dependencies for browser automation tooling
+node_modules/

AGENTS.md

@@ -0,0 +1,32 @@
+# Repository Guidelines
+
+## Project Structure & Module Organization
+
+Core poker service code lives in `texas_holdem/`. Important modules include `engine.py` for Texas Hold'em rules, `service.py` for game management, `server.py` for the HTTP API, `agents.py` for local/HTTP agents, and `ai_client.py` / `human_client.py` for standalone agents. Prompt templates live in `texas_holdem/prompts/`.
+
+Replay UI code lives in `texas_holdem_replay/`, with static browser assets under `texas_holdem_replay/static/`. Tests are in `tests/`, named by feature area such as `test_engine.py`, `test_service.py`, and `test_replay_server.py`. Design notes belong in `docs/`.
+
+## Build, Test, and Development Commands
+
+- `python -m unittest discover -v` runs the full test suite.
+- `python -m compileall texas_holdem texas_holdem_replay tests` checks import and syntax validity.
+- `python -m texas_holdem.server --host 127.0.0.1 --port 8000` starts the game service.
+- `python -m texas_holdem.human_client --port 9001 --keep-history` starts an interactive human HTTP agent.
+- `python -m texas_holdem.ai_client --port 9101 --api-key "$OPENAI_API_KEY" --model gpt-4o-mini` starts an OpenAI-compatible AI agent.
+- `python -m texas_holdem_replay.server --port 8088` starts the replay viewer.
+
+## Coding Style & Naming Conventions
+
+Use Python 3.11+ standard-library APIs unless a dependency is intentionally added to `pyproject.toml`. Keep modules focused and prefer explicit dataclasses for wire/state models. Use 4-space indentation, type hints, `snake_case` for functions and variables, `PascalCase` for classes, and concise comments only where logic is non-obvious.
+
+## Testing Guidelines
+
+Use `unittest`. Add tests near the behavior changed: engine rules in `tests/test_engine.py`, HTTP/service behavior in `tests/test_service.py`, agent transport in `tests/test_agents.py`, and replay UI server helpers in `tests/test_replay_server.py`. New bug fixes should include a regression test. Avoid tests that require external network access or real LLM calls.
+
+## Commit & Pull Request Guidelines
+
+History uses short Conventional Commit-style subjects, for example `feat: add replay server and web client` and `fix: game service api block when a game is running`. Keep commits scoped to one behavior change. Pull requests should include a short summary, test commands run, linked issue or motivation, and screenshots only for replay UI or visible terminal-output changes.
+
+## Security & Configuration Tips
+
+Do not commit API keys. Pass LLM credentials through `OPENAI_API_KEY` or CLI flags in local shells only. HTTP Agent endpoints are exclusive per active game; preserve this invariant when changing service concurrency.

README.md

@@ -11,6 +11,7 @@
 - 支持盲注、四条街下注、弃牌、过牌、跟注、下注、加注、全下、边池和摊牌结算。
 - 支持本地 Agent 和 HTTP Agent。
 - 支持 Human Agent 和 OpenAI-compatible AI Agent 的终端过程输出。
+- 游戏运行中可以并发查询状态；查询返回上一手完成后的稳定快照。
 
 ## 运行服务
 
@@ -51,6 +52,12 @@ curl -X POST http://127.0.0.1:8000/games/demo/hands/run \
 curl http://127.0.0.1:8000/games/demo
 ```
 
+也可以使用单数别名：
+
+```bash
+curl http://127.0.0.1:8000/game/demo
+```
+
 ## HTTP Agent 协议
 
 玩家配置可以使用远程 HTTP Agent：
@@ -62,12 +69,19 @@ curl http://127.0.0.1:8000/games/demo
   "agent": {
     "type": "http",
     "endpoint": "http://127.0.0.1:9101",
-    "timeout_seconds": 10
+    "timeout_seconds": 10,
+    "game_update_timeout_seconds": 3,
+    "retries": 2,
+    "retry_backoff_seconds": 0.25
   }
 }
 ```
 
-服务会向 `endpoint` 发送当前行动玩家的观察 JSON。Agent 返回：
+服务会向 `endpoint + /game` 推送每手开始时的游戏快照，向 `endpoint + /act` 发送当前行动玩家的观察 JSON。`endpoint` 也可以传入历史形式的 `/act` 或 `/game` 后缀，服务会归一化为 base URL。
+
+同一个 HTTP Agent endpoint 不能同时被不同游戏占用；后创建的游戏会返回错误。服务会给 HTTP Agent 请求自动重试，`/act` 重试仍失败时，规则引擎会按 `check > call > fold` 选择默认动作，避免整桌卡死。
+
+Agent 返回：
 
 ```json
 {"action": "call"}

tests/test_agents.py

@@ -0,0 +1,60 @@
+import json
+import unittest
+from unittest.mock import patch
+from urllib.error import URLError
+
+from texas_holdem.agents import HttpAgent, normalise_http_agent_endpoint
+
+
+class FakeResponse:
+    def __init__(self, payload: dict[str, object]) -> None:
+        self.payload = payload
+
+    def read(self) -> bytes:
+        return json.dumps(self.payload).encode("utf-8")
+
+    def __enter__(self) -> "FakeResponse":
+        return self
+
+    def __exit__(self, *args: object) -> None:
+        return None
+
+
+class AgentTests(unittest.TestCase):
+    def test_normalise_http_agent_endpoint_accepts_action_or_game_paths(self) -> None:
+        self.assertEqual(
+            normalise_http_agent_endpoint("http://127.0.0.1:9101/act"),
+            "http://127.0.0.1:9101",
+        )
+        self.assertEqual(
+            normalise_http_agent_endpoint("http://127.0.0.1:9101/game/"),
+            "http://127.0.0.1:9101",
+        )
+
+    def test_http_agent_post_retries_and_sets_player_header(self) -> None:
+        calls = []
+
+        def fake_urlopen(request, timeout):  # type: ignore[no-untyped-def]
+            calls.append((request, timeout))
+            if len(calls) == 1:
+                raise URLError("temporary")
+            return FakeResponse({"ok": True})
+
+        agent = HttpAgent(
+            "http://agent.test/act",
+            player_id="p1",
+            retries=1,
+            retry_backoff_seconds=0,
+        )
+
+        with patch("texas_holdem.agents.urlopen", fake_urlopen):
+            payload = agent._post_json("/game", {"game_id": "g1"}, timeout_seconds=2)
+
+        self.assertEqual(payload, {"ok": True})
+        self.assertEqual(len(calls), 2)
+        self.assertEqual(calls[1][0].headers["X-player-id"], "p1")
+        self.assertEqual(calls[1][1], 2)
+
+
+if __name__ == "__main__":
+    unittest.main()

tests/test_service.py

@@ -1,8 +1,26 @@
 import unittest
+from threading import Event, Thread
 
+from texas_holdem.agents import PokerAgent
+from texas_holdem.models import Observation, PlayerAction
 from texas_holdem.service import GameManager
 
 
+class BlockingAgent(PokerAgent):
+    def __init__(self, entered: Event, release: Event) -> None:
+        self.entered = entered
+        self.release = release
+
+    def decide(self, observation: Observation) -> PlayerAction:
+        self.entered.set()
+        if not self.release.wait(timeout=5):
+            raise RuntimeError("test timed out waiting to release blocking agent")
+        for action in observation.legal_actions:
+            if action["action"] == "check":
+                return PlayerAction("check")
+        return PlayerAction("call")
+
+
 class ServiceTests(unittest.TestCase):
     def test_create_and_run_game(self) -> None:
         manager = GameManager()
@@ -23,7 +41,62 @@ class ServiceTests(unittest.TestCase):
         hands = manager.run_hands(game.game_id, count=1)
 
         self.assertEqual(len(hands), 1)
-        self.assertEqual(manager.get_game("demo").to_dict()["hand_number"], 1)
+        self.assertEqual(manager.get_game_state("demo")["hand_number"], 1)
+
+    def test_get_game_state_does_not_block_during_run(self) -> None:
+        manager = GameManager()
+        entered = Event()
+        release = Event()
+        game = manager.create_game(
+            {
+                "game_id": "blocking",
+                "seed": 13,
+                "starting_stack": 200,
+                "small_blind": 5,
+                "big_blind": 10,
+                "players": [
+                    {"id": "a", "type": "calling"},
+                    {"id": "b", "type": "calling"},
+                ],
+            }
+        )
+        manager.run_hands("blocking", count=1)
+        game.agents["a"] = BlockingAgent(entered, release)
+
+        thread = Thread(target=lambda: manager.run_hands("blocking", count=1))
+        thread.start()
+        self.assertTrue(entered.wait(timeout=2))
+
+        state = manager.get_game_state("blocking")
+
+        release.set()
+        thread.join(timeout=2)
+        self.assertFalse(thread.is_alive())
+        self.assertEqual(state["hand_number"], 1)
+        self.assertEqual(len(state["hands"]), 1)
+
+    def test_duplicate_http_agent_endpoint_is_rejected_across_active_games(self) -> None:
+        manager = GameManager()
+        payload = {
+            "starting_stack": 200,
+            "small_blind": 5,
+            "big_blind": 10,
+            "players": [
+                {
+                    "id": "ai",
+                    "agent": {
+                        "type": "http",
+                        "endpoint": "http://127.0.0.1:9101/act",
+                    },
+                },
+                {"id": "b", "type": "calling"},
+            ],
+        }
+
+        manager.create_game({"game_id": "g1", **payload})
+
+        with self.assertRaisesRegex(ValueError, "already belongs to game g1"):
+            manager.create_game({"game_id": "g2", **payload})
 
 
 if __name__ == "__main__":

texas_holdem/agents.py

@@ -2,10 +2,11 @@ from __future__ import annotations
 
 import json
 import sys
+import time
 from abc import ABC, abstractmethod
 from random import Random
 from typing import IO, Any
-from urllib.error import URLError
+from urllib.error import HTTPError, URLError
 from urllib.request import Request, urlopen
 
 from texas_holdem.human_io import clear_screen, prompt_action, render_observation
@@ -54,6 +55,27 @@ class CallingStationAgent(PokerAgent):
         return PlayerAction("fold")
 
 
+def normalise_http_agent_endpoint(raw: str) -> str:
+    """Return the canonical base URL for an HTTP agent endpoint."""
+    url = raw.rstrip("/")
+    if url.endswith("/act"):
+        url = url[: -len("/act")]
+    if url.endswith("/game"):
+        url = url[: -len("/game")]
+    return url
+
+
+def http_agent_endpoint_from_spec(spec: dict[str, Any]) -> str | None:
+    """Extract the canonical HTTP endpoint from an agent spec, if present."""
+    agent_type = str(spec.get("type", "calling")).lower()
+    if agent_type != "http":
+        return None
+    endpoint = spec.get("endpoint")
+    if not endpoint:
+        raise ValueError("http agent requires an endpoint")
+    return normalise_http_agent_endpoint(str(endpoint))
+
+
 class HttpAgent(PokerAgent):
     """Remote agent that talks to a base URL exposing ``/act`` and ``/game``.
 
@@ -66,28 +88,36 @@ class HttpAgent(PokerAgent):
     ACT_PATH = "/act"
     GAME_PATH = "/game"
 
-    def __init__(self, endpoint: str, timeout_seconds: float = 10.0) -> None:
-        self.base_url = self._normalise_base_url(endpoint)
+    def __init__(
+        self,
+        endpoint: str,
+        timeout_seconds: float = 10.0,
+        player_id: str | None = None,
+        game_update_timeout_seconds: float | None = None,
+        retries: int = 2,
+        retry_backoff_seconds: float = 0.25,
+    ) -> None:
+        self.base_url = normalise_http_agent_endpoint(endpoint)
         self.timeout_seconds = timeout_seconds
-
-    @staticmethod
-    def _normalise_base_url(raw: str) -> str:
-        """Strip a trailing slash so URL joins do not produce double slashes.
-
-        Centralising this also tolerates the legacy "endpoint already points
-        at /act" mistake by chopping off a redundant ``/act`` suffix.
-        """
-        url = raw.rstrip("/")
-        if url.endswith("/act"):
-            url = url[: -len("/act")]
-        return url
+        self.player_id = player_id
+        self.game_update_timeout_seconds = (
+            float(game_update_timeout_seconds)
+            if game_update_timeout_seconds is not None
+            else min(timeout_seconds, 3.0)
+        )
+        self.retries = max(0, retries)
+        self.retry_backoff_seconds = max(0.0, retry_backoff_seconds)
 
     def _url(self, path: str) -> str:
         """Compose a full URL by joining the base with a path component."""
         return f"{self.base_url}{path}"
 
     def decide(self, observation: Observation) -> PlayerAction:
-        payload = self._post_json(self.ACT_PATH, observation.to_dict())
+        payload = self._post_json(
+            self.ACT_PATH,
+            observation.to_dict(),
+            timeout_seconds=self.timeout_seconds,
+        )
         if not isinstance(payload, dict):
             raise RuntimeError("agent endpoint must return a JSON object")
         return PlayerAction.from_dict(payload)
@@ -100,30 +130,54 @@ class HttpAgent(PokerAgent):
         only by way of the raised exception bubbling to the engine guard.
         """
         try:
-            self._post_json(self.GAME_PATH, game_state)
+            self._post_json(
+                self.GAME_PATH,
+                game_state,
+                timeout_seconds=self.game_update_timeout_seconds,
+            )
         except RuntimeError:
             # ``/game`` is informational; treat any HTTP error as a benign
             # drop rather than reraising and aborting the hand loop.
             return None
 
-    def _post_json(self, path: str, payload: dict[str, Any]) -> Any:
+    def _post_json(
+        self,
+        path: str,
+        payload: dict[str, Any],
+        timeout_seconds: float,
+    ) -> Any:
         """POST ``payload`` as JSON to ``base_url + path`` and return parsed body.
 
         Extracted as a tiny helper so ``decide`` and ``on_game_update`` share
         identical transport semantics (timeout, error wrapping, content-type).
         """
         body = json.dumps(payload).encode("utf-8")
-        request = Request(
-            self._url(path),
-            data=body,
-            headers={"Content-Type": "application/json"},
-            method="POST",
-        )
-        try:
-            with urlopen(request, timeout=self.timeout_seconds) as response:
-                raw = response.read().decode("utf-8")
-        except (OSError, URLError) as exc:
-            raise RuntimeError(f"agent endpoint failed: {self._url(path)}") from exc
+        last_error: BaseException | None = None
+        raw = ""
+        for attempt in range(self.retries + 1):
+            request = Request(
+                self._url(path),
+                data=body,
+                headers=self._headers(),
+                method="POST",
+            )
+            try:
+                with urlopen(request, timeout=timeout_seconds) as response:
+                    raw = response.read().decode("utf-8")
+                break
+            except HTTPError as exc:
+                detail = exc.read().decode("utf-8", errors="replace")
+                last_error = RuntimeError(
+                    f"agent endpoint failed with HTTP {exc.code}: "
+                    f"{self._url(path)} {detail}"
+                )
+            except (OSError, URLError) as exc:
+                last_error = exc
+            if attempt < self.retries and self.retry_backoff_seconds > 0:
+                time.sleep(self.retry_backoff_seconds * (2**attempt))
+        else:
+            raise RuntimeError(f"agent endpoint failed: {self._url(path)}") from last_error
+
         if not raw:
             return None
         try:
@@ -133,6 +187,12 @@ class HttpAgent(PokerAgent):
                 f"agent endpoint returned invalid JSON: {self._url(path)}"
             ) from exc
 
+    def _headers(self) -> dict[str, str]:
+        headers = {"Content-Type": "application/json", "Connection": "close"}
+        if self.player_id:
+            headers["X-Player-Id"] = self.player_id
+        return headers
+
 
 class HumanAgent(PokerAgent):
     """Interactive CLI agent for debugging and manual play.
@@ -189,7 +249,11 @@ class HumanAgent(PokerAgent):
         return line.rstrip("\n")
 
 
-def build_agent(spec: dict[str, Any], rng: Random | None = None) -> PokerAgent:
+def build_agent(
+    spec: dict[str, Any],
+    rng: Random | None = None,
+    player_id: str | None = None,
+) -> PokerAgent:
     agent_type = str(spec.get("type", "calling")).lower()
     if agent_type == "random":
         return RandomAgent(rng)
@@ -199,7 +263,18 @@ def build_agent(spec: dict[str, Any], rng: Random | None = None) -> PokerAgent:
         endpoint = spec.get("endpoint")
         if not endpoint:
             raise ValueError("http agent requires an endpoint")
-        return HttpAgent(str(endpoint), float(spec.get("timeout_seconds", 10.0)))
+        return HttpAgent(
+            str(endpoint),
+            timeout_seconds=float(spec.get("timeout_seconds", 10.0)),
+            player_id=player_id,
+            game_update_timeout_seconds=(
+                float(spec["game_update_timeout_seconds"])
+                if "game_update_timeout_seconds" in spec
+                else None
+            ),
+            retries=int(spec.get("retries", 2)),
+            retry_backoff_seconds=float(spec.get("retry_backoff_seconds", 0.25)),
+        )
     if agent_type in {"human", "cli", "interactive"}:
         return HumanAgent()
     raise ValueError(f"unknown agent type: {agent_type}")

texas_holdem/ai_client.py

@@ -69,6 +69,109 @@ ANSI_RESET = "\x1b[0m"
 # ---------------------------------------------------------------------------
 
 
+class _ThinkingIndicator:
+    """Animated "thinking..." marquee for the AI agent console.
+
+    Design rationale:
+    - Encapsulated as its own class so the animation lifecycle (timer
+      thread, frame state, screen erase sequence) does not pollute the
+      surrounding console class.
+    - Runs in a daemon background thread driven by ``threading.Event`` so
+      ``stop`` returns promptly even if the current frame is mid-sleep.
+    - Uses ANSI ``\\r`` plus a clearing escape sequence to overwrite the
+      previous frame in place, avoiding scrollback noise. The frames
+      cycle through 0/1/2/3 dots every 0.5s as requested.
+    - ``start``/``stop`` are idempotent so the higher-level console can
+      call ``stop`` defensively (e.g. on the fallback path) without
+      tracking whether a marquee is actually running.
+    """
+
+    # Frame interval in seconds; matches the user-visible cadence.
+    _FRAME_INTERVAL = 0.5
+    # 0..3 dots, looping.
+    _FRAMES = ("thinking", "thinking.", "thinking..", "thinking...")
+    # ANSI escape that clears from the cursor to the end of the line; we
+    # combine it with a leading carriage return to redraw the frame in
+    # place.
+    _ERASE_LINE = "\r\x1b[K"
+
+    def __init__(
+        self,
+        write_fn: Callable[[str], None],
+        gray_fn: Callable[[str], str],
+    ) -> None:
+        self._write = write_fn
+        self._gray = gray_fn
+        self._stop_event = threading.Event()
+        self._thread: threading.Thread | None = None
+        # ``_active`` reflects whether a frame is currently visible on
+        # screen; ``stop`` uses it to decide whether to emit the final
+        # erase sequence.
+        self._active = False
+        # Guard against concurrent start/stop calls from different
+        # threads (e.g. content-delta handler vs. end_llm_stream).
+        self._lifecycle_lock = threading.Lock()
+
+    def start(self) -> None:
+        """Begin the marquee in a background thread.
+
+        Calling ``start`` while already running is a no-op.
+        """
+        with self._lifecycle_lock:
+            if self._thread is not None and self._thread.is_alive():
+                return
+            self._stop_event.clear()
+            self._active = True
+            thread = threading.Thread(
+                target=self._run,
+                name="ai-thinking-indicator",
+                daemon=True,
+            )
+            self._thread = thread
+            thread.start()
+
+    def stop(self) -> None:
+        """Stop the marquee and erase the current frame from the screen.
+
+        Safe to call when not running.
+        """
+        with self._lifecycle_lock:
+            thread = self._thread
+            if thread is None:
+                return
+            self._stop_event.set()
+            self._thread = None
+        # Wait for the worker outside the lifecycle lock so an in-flight
+        # ``_render_frame`` cannot deadlock against ``start`` from
+        # another thread.
+        thread.join()
+        if self._active:
+            # Wipe the last frame so the model's actual content begins on
+            # a clean line.
+            self._write(self._ERASE_LINE)
+            self._active = False
+
+    def _run(self) -> None:
+        """Background loop: redraw the next frame every ``_FRAME_INTERVAL``."""
+        index = 0
+        while not self._stop_event.is_set():
+            self._render_frame(self._FRAMES[index % len(self._FRAMES)])
+            index += 1
+            # ``Event.wait`` returns immediately when ``set`` is called,
+            # so ``stop`` is responsive even mid-frame.
+            if self._stop_event.wait(self._FRAME_INTERVAL):
+                return
+
+    def _render_frame(self, label: str) -> None:
+        """Emit one frame in place using carriage-return + erase-EOL."""
+        self._write(f"{self._ERASE_LINE}{self._gray(label)}")
+
+
+# ---------------------------------------------------------------------------
+# AI agent console
+# ---------------------------------------------------------------------------
+
+
 class AIAgentConsole:
     """Serialised terminal output for the standalone AI agent.
 
@@ -84,11 +187,30 @@ class AIAgentConsole:
         output_stream: IO[str] | None = None,
         keep_history: bool = False,
         use_color: bool = True,
+        show_reasoning: bool = True,
     ) -> None:
         self._output = output_stream if output_stream is not None else sys.stdout
         self._keep_history = keep_history
         self._use_color = use_color
+        # ``show_reasoning`` controls whether the LLM's chain-of-thought
+        # ("reasoning") deltas are printed to the terminal. The final
+        # answer ("content") is always printed so operators can still see
+        # the action being chosen.
+        self._show_reasoning = show_reasoning
+        # ``_lock`` serialises whole act/game render blocks (coarse grain).
+        # ``_io_lock`` is a finer-grained mutex protecting just the
+        # ``self._output.write`` calls so the thinking-indicator background
+        # thread can interleave safely with the main rendering thread
+        # without being blocked by the coarse lock.
         self._lock = threading.Lock()
+        self._io_lock = threading.Lock()
+        # Animated "thinking..." marquee shown while reasoning output is
+        # suppressed. Created up-front so callers can ``start``/``stop``
+        # idempotently regardless of the show_reasoning flag.
+        self._thinking = _ThinkingIndicator(
+            write_fn=self._write,
+            gray_fn=self._gray,
+        )
 
     @contextmanager
     def act_log(self, observation: dict[str, Any]) -> Iterator[None]:
@@ -106,13 +228,32 @@ class AIAgentConsole:
 
     def begin_llm_stream(self) -> None:
         self._write(self._gray("AI MODEL STREAM\n"))
+        # When reasoning output is hidden, immediately start the marquee
+        # so the user sees liveness while the model is "thinking" before
+        # any content delta arrives.
+        if not self._show_reasoning:
+            self._thinking.start()
 
     def write_llm_delta(self, kind: str, text: str) -> None:
         if not text:
             return
+        # Skip "reasoning" deltas entirely when reasoning output is hidden;
+        # this keeps the terminal focused on the final answer for users
+        # who do not care about chain-of-thought traces.
+        if kind == "reasoning" and not self._show_reasoning:
+            return
+        # First non-reasoning delta means the model has started speaking
+        # the actual answer; tear down the marquee before printing so the
+        # animation does not collide with the content stream.
+        if kind == "content" and not self._show_reasoning:
+            self._thinking.stop()
         self._write(self._gray(text))
 
     def end_llm_stream(self) -> None:
+        # Defensive stop in case the request finished without ever
+        # producing a content delta (e.g. fallback path / error).
+        if not self._show_reasoning:
+            self._thinking.stop()
         self._write(self._gray("\n"))
 
     def announce_action(
@@ -120,11 +261,17 @@ class AIAgentConsole:
         action: dict[str, Any],
         source: str = "model",
     ) -> None:
+        # Defensive stop: error / fallback paths bypass end_llm_stream, so
+        # we ensure the marquee never leaks into action / warning output.
+        self._thinking.stop()
         body = json.dumps(action, ensure_ascii=False)
         self._write(f"\nAI ACTION ({source}) -> {body}\n")
         self._write("~" * 60 + "\n\n")
 
     def announce_warning(self, message: str) -> None:
+        # Same defensive stop as ``announce_action`` - warnings can fire
+        # before the LLM stream closes (HTTP error, JSON parse error...).
+        self._thinking.stop()
         self._write(f"\nAI WARNING -> {message}\n")
 
     def _gray(self, text: str) -> str:
@@ -133,8 +280,12 @@ class AIAgentConsole:
         return f"{ANSI_GRAY}{text}{ANSI_RESET}"
 
     def _write(self, text: str) -> None:
-        self._output.write(text)
-        self._output.flush()
+        # The thinking-indicator background thread writes from a different
+        # thread than the main /act handler; the fine-grained ``_io_lock``
+        # avoids tearing of escape sequences and keeps stdout consistent.
+        with self._io_lock:
+            self._output.write(text)
+            self._output.flush()
 
 
 # ---------------------------------------------------------------------------
@@ -296,8 +447,11 @@ def _format_action_history(history: list[dict[str, Any]]) -> str:
         return "(no actions yet)"
     # The engine never produces unbounded history within a single hand, but
     # we cap defensively so a malformed payload cannot blow up token usage.
+    # The cap is sized to comfortably cover the worst realistic case (a
+    # 12-handed table running ~10 betting rounds within one hand) so the
+    # LLM never sees a silently truncated history at full ring tables.
     rows = []
-    for record in history[-32:]:
+    for record in history[-128:]:
         rows.append(
             f"- [{record.get('street')}] {record.get('player_id')} -> "
             f"{record.get('action')} amount={record.get('amount', 0)}"
@@ -927,6 +1081,16 @@ def main() -> None:
         action="store_true",
         help="Disable ANSI gray coloring for streamed LLM output.",
     )
+    parser.add_argument(
+        "--hide-reasoning",
+        action="store_true",
+        help=(
+            "Hide the LLM's reasoning/chain-of-thought stream from the "
+            "terminal. The model still performs reasoning; only its "
+            "terminal output is suppressed. The final answer (content) "
+            "is still printed so operators can see the chosen action."
+        ),
+    )
     args = parser.parse_args()
 
     if not args.api_key:
@@ -944,6 +1108,7 @@ def main() -> None:
     console = AIAgentConsole(
         keep_history=args.keep_history,
         use_color=not args.no_color,
+        show_reasoning=not args.hide_reasoning,
     )
     service = AIAgentService(LLMClient(config), prompts, console=console)
     server = create_server(args.host, args.port, service, default_player_id=args.player_id)
@@ -956,6 +1121,7 @@ def main() -> None:
         f"  base_url    : {config.base_url}\n"
         f"  player_id   : {args.player_id}\n"
         f"  stream      : {'on' if config.stream else 'off'}\n"
+        f"  reasoning   : {'hidden (output suppressed)' if args.hide_reasoning else 'visible'}\n"
         f"  clear-screen: {'off (keep history)' if args.keep_history else 'on'}",
         file=sys.stderr,
         flush=True,

texas_holdem/engine.py

@@ -1,6 +1,8 @@
 from __future__ import annotations
 
+from copy import deepcopy
 from random import Random
+from threading import RLock
 from time import time
 
 from texas_holdem.agents import PokerAgent
@@ -8,6 +10,7 @@ from texas_holdem.cards import Deck
 from texas_holdem.evaluator import evaluate
 from texas_holdem.models import (
     ActionRecord,
+    BlindLevel,
     HandSummary,
     Observation,
     PlayerAction,
@@ -53,21 +56,59 @@ class TableGame:
         self.small_blind = small_blind
         self.big_blind = big_blind
         self.rng = rng or Random()
+        self.lock = RLock()
         self.hand_number = 0
         self.button_index: int | None = None
         self.board = []
         self.action_history: list[ActionRecord] = []
         self.hand_summaries: list[HandSummary] = []
+        # ``blind_history`` is an append-only log of every blind level change
+        # (including the initial one). Each entry's ``hand_number`` is the
+        # first hand that played under those stakes, which makes it trivial
+        # to reconstruct the schedule from the outside.
+        self.blind_history: list[BlindLevel] = []
+        self._completed_snapshot: dict[str, object] = self._to_dict_unlocked()
 
     @property
     def is_complete(self) -> bool:
         return len([player for player in self.players if player.stack > 0]) < 2
 
-    def run_hand(self) -> HandSummary:
+    def run_hand(
+        self,
+        small_blind: int | None = None,
+        big_blind: int | None = None,
+    ) -> HandSummary:
+        """Play a single hand.
+
+        ``small_blind`` / ``big_blind`` allow callers to bump the stakes
+        between hands without rebuilding the table. Either both must be
+        provided or both omitted (in which case the previously configured
+        blinds carry over). The resolved blind level is appended to
+        :attr:`blind_history` whenever it changes (including the very first
+        hand) so external observers can replay the schedule.
+        """
+        with self.lock:
+            return self._run_hand_locked(small_blind=small_blind, big_blind=big_blind)
+
+    def _run_hand_locked(
+        self,
+        small_blind: int | None = None,
+        big_blind: int | None = None,
+    ) -> HandSummary:
         if self.is_complete:
             raise GameComplete("game is complete")
 
+        self._apply_blinds_for_hand(small_blind, big_blind)
+
         self.hand_number += 1
+        # Stamp the active blind level onto the upcoming summary so a hand
+        # remains self-describing even after the blinds change later on.
+        active_blinds = BlindLevel(
+            hand_number=self.hand_number,
+            small_blind=self.small_blind,
+            big_blind=self.big_blind,
+        )
+        self._record_blind_level_if_new(active_blinds)
         started_at = time()
         self.board = []
         self.action_history = []
@@ -116,26 +157,69 @@ class TableGame:
             board=list(self.board),
             actions=list(self.action_history),
             awards=awards,
+            blinds=active_blinds,
             showdown_hands=self._collect_showdown_hands(),
             started_at=started_at,
             finished_at=time(),
         )
         self.hand_summaries.append(summary)
+        self._completed_snapshot = deepcopy(self._to_dict_unlocked())
         return summary
 
-    def run_hands(self, max_hands: int, until_one_left: bool = False) -> list[HandSummary]:
-        if max_hands <= 0:
-            raise ValueError("max_hands must be positive")
-        summaries = []
-        for _ in range(max_hands):
-            if self.is_complete:
-                break
-            summaries.append(self.run_hand())
-            if until_one_left and self.is_complete:
-                break
-        return summaries
+    def run_hands(
+        self,
+        max_hands: int,
+        until_one_left: bool = False,
+        small_blind: int | None = None,
+        big_blind: int | None = None,
+    ) -> list[HandSummary]:
+        """Play up to ``max_hands`` hands using a single blind configuration.
+
+        Passing ``small_blind`` / ``big_blind`` bumps the stakes starting
+        with the first hand of this call; subsequent calls can raise them
+        again. Leaving them ``None`` keeps the current level unchanged.
+        """
+        with self.lock:
+            if max_hands <= 0:
+                raise ValueError("max_hands must be positive")
+            summaries = []
+            for _ in range(max_hands):
+                if self.is_complete:
+                    break
+                # Only the first hand of the batch needs to apply the blind
+                # override; after that the engine reuses the stored values.
+                summaries.append(
+                    self._run_hand_locked(
+                        small_blind=small_blind,
+                        big_blind=big_blind,
+                    )
+                )
+                small_blind = None
+                big_blind = None
+                if until_one_left and self.is_complete:
+                    break
+            return summaries
 
     def to_dict(self) -> dict[str, object]:
+        with self.lock:
+            return self._to_dict_unlocked()
+
+    def snapshot_completed(self) -> dict[str, object]:
+        """Return a stable snapshot from the latest completed hand boundary.
+
+        If a hand is currently running under ``self.lock``, this method does
+        not block. It returns the most recent completed hand summary and
+        stacks captured in memory, which is exactly what status endpoints
+        need while a long-running HTTP-agent decision is in progress.
+        """
+        if self.lock.acquire(blocking=False):
+            try:
+                return deepcopy(self._to_dict_unlocked())
+            finally:
+                self.lock.release()
+        return deepcopy(self._completed_snapshot)
+
+    def _to_dict_unlocked(self) -> dict[str, object]:
         return {
             "game_id": self.game_id,
             "status": "complete" if self.is_complete else "running",
@@ -143,8 +227,18 @@ class TableGame:
             "button_seat": None
             if self.button_index is None
             else self.players[self.button_index].seat,
+            # ``small_blind`` / ``big_blind`` mirror the *current* level so
+            # legacy callers keep working. New consumers should prefer the
+            # structured ``blinds`` block which carries the full schedule.
             "small_blind": self.small_blind,
             "big_blind": self.big_blind,
+            "blinds": {
+                "current": {
+                    "small_blind": self.small_blind,
+                    "big_blind": self.big_blind,
+                },
+                "history": [level.to_dict() for level in self.blind_history],
+            },
             "starting_stack": self.starting_stack,
             "players": [player.public_dict() for player in self.players],
             # ``hands`` exposes every finished hand (each entry is the same
@@ -153,6 +247,47 @@ class TableGame:
             "hands": [summary.to_dict() for summary in self.hand_summaries],
         }
 
+    def _apply_blinds_for_hand(
+        self,
+        small_blind: int | None,
+        big_blind: int | None,
+    ) -> None:
+        """Validate and apply optional per-hand blind overrides.
+
+        Splitting this out keeps :meth:`run_hand` focused on the table flow
+        while letting us reuse the validation rules originally enforced by
+        ``__init__``. We require both values to be supplied together so the
+        configuration cannot drift into an inconsistent half-update.
+        """
+        if small_blind is None and big_blind is None:
+            return
+        if small_blind is None or big_blind is None:
+            raise ValueError(
+                "small_blind and big_blind must be provided together"
+            )
+        if small_blind <= 0 or big_blind <= 0 or small_blind > big_blind:
+            raise ValueError("blinds must satisfy 0 < small_blind <= big_blind")
+        self.small_blind = int(small_blind)
+        self.big_blind = int(big_blind)
+
+    def _record_blind_level_if_new(self, level: BlindLevel) -> None:
+        """Append ``level`` to :attr:`blind_history` when it differs.
+
+        Comparing against the latest entry (rather than blindly appending)
+        keeps the log compact: stretches of unchanged stakes only contribute
+        a single record. The very first hand always seeds an entry because
+        the history starts empty.
+        """
+        if not self.blind_history:
+            self.blind_history.append(level)
+            return
+        latest = self.blind_history[-1]
+        if (
+            latest.small_blind != level.small_blind
+            or latest.big_blind != level.big_blind
+        ):
+            self.blind_history.append(level)
+
     def _advance_button(self) -> None:
         if self.button_index is None:
             self.button_index = self._next_index(0, lambda index: self.players[index].stack > 0)
@@ -350,9 +485,18 @@ class TableGame:
         try:
             requested = agent.decide(observation)
         except Exception:
-            requested = PlayerAction("fold")
+            requested = self._default_action(observation.legal_actions)
         return self._coerce_action(requested, observation.legal_actions)
 
+    def _default_action(self, legal_actions: list[dict[str, object]]) -> PlayerAction:
+        by_action = {str(action["action"]): action for action in legal_actions}
+        for action_type in ("check", "call", "fold"):
+            if action_type in by_action:
+                legal = by_action[action_type]
+                return PlayerAction(action_type, int(legal.get("amount") or 0))
+        legal = legal_actions[0]
+        return PlayerAction(str(legal["action"]), int(legal.get("amount") or 0))
+
     def _coerce_action(
         self,
         requested: PlayerAction,
@@ -482,7 +626,7 @@ class TableGame:
         swallow individual exceptions so a flaky remote endpoint cannot
         break the table flow.
         """
-        snapshot = self.to_dict()
+        snapshot = self._to_dict_unlocked()
         for agent in self.agents.values():
             try:
                 agent.on_game_update(snapshot)

texas_holdem/models.py

@@ -130,6 +130,28 @@ class Observation:
         }
 
 
+@dataclass(slots=True)
+class BlindLevel:
+    """A snapshot of the blind configuration that took effect at a given hand.
+
+    The structure is intentionally append-only: every time the blinds change
+    (or the very first hand seeds the initial values) we push a new
+    ``BlindLevel`` so callers can reconstruct how the stakes evolved over the
+    course of the game without losing any prior state.
+    """
+
+    hand_number: int
+    small_blind: int
+    big_blind: int
+
+    def to_dict(self) -> dict[str, object]:
+        return {
+            "hand_number": self.hand_number,
+            "small_blind": self.small_blind,
+            "big_blind": self.big_blind,
+        }
+
+
 @dataclass(slots=True)
 class PotAward:
     amount: int
@@ -152,6 +174,10 @@ class HandSummary:
     board: list[Card]
     actions: list[ActionRecord]
     awards: list[PotAward]
+    # ``blinds`` records the exact blind level used by this hand. Storing it
+    # on the summary (rather than only on the game) guarantees historical
+    # hands remain self-describing even after the blinds are raised later.
+    blinds: BlindLevel | None = None
     showdown_hands: dict[str, list[Card]] = field(default_factory=dict)
     started_at: float = field(default_factory=time)
     finished_at: float = field(default_factory=time)
@@ -161,6 +187,7 @@ class HandSummary:
             "game_id": self.game_id,
             "hand_number": self.hand_number,
             "button_seat": self.button_seat,
+            "blinds": self.blinds.to_dict() if self.blinds else None,
             "board": [str(card) for card in self.board],
             "actions": [record.to_dict() for record in self.actions],
             "awards": [award.to_dict() for award in self.awards],

texas_holdem/server.py

@@ -25,8 +25,8 @@ class PokerRequestHandler(BaseHTTPRequestHandler):
             if path == ["games"]:
                 self._json({"games": MANAGER.list_games()})
                 return
-            if len(path) == 2 and path[0] == "games":
-                self._json(MANAGER.get_game(path[1]).to_dict())
+            if len(path) == 2 and path[0] in {"game", "games"}:
+                self._json(MANAGER.get_game_state(path[1]))
                 return
             self._json({"error": "not found"}, HTTPStatus.NOT_FOUND)
         except KeyError as exc:
@@ -35,23 +35,37 @@ class PokerRequestHandler(BaseHTTPRequestHandler):
     def do_POST(self) -> None:
         path = self._path_parts()
         try:
-            if path == ["games"]:
+            if path in (["game"], ["games"]):
                 game = MANAGER.create_game(self._read_json())
-                self._json(game.to_dict(), HTTPStatus.CREATED)
+                self._json(game.snapshot_completed(), HTTPStatus.CREATED)
                 return
-            if len(path) == 3 and path[0] == "games" and path[2] == "hands":
+            if len(path) == 3 and path[0] in {"game", "games"} and path[2] == "hands":
                 body = self._read_json()
                 count = int(body.get("count", 1))
                 until_one_left = bool(body.get("until_one_left", False))
-                summaries = MANAGER.run_hands(path[1], count, until_one_left)
-                self._json({"hands": summaries, "game": MANAGER.get_game(path[1]).to_dict()})
+                small_blind, big_blind = self._extract_blinds(body)
+                summaries = MANAGER.run_hands(
+                    path[1],
+                    count,
+                    until_one_left,
+                    small_blind=small_blind,
+                    big_blind=big_blind,
+                )
+                self._json({"hands": summaries, "game": MANAGER.get_game_state(path[1])})
                 return
-            if len(path) == 4 and path[0] == "games" and path[2] == "hands" and path[3] == "run":
+            if len(path) == 4 and path[0] in {"game", "games"} and path[2] == "hands" and path[3] == "run":
                 body = self._read_json()
                 count = int(body.get("count", 1))
                 until_one_left = bool(body.get("until_one_left", False))
-                summaries = MANAGER.run_hands(path[1], count, until_one_left)
-                self._json({"hands": summaries, "game": MANAGER.get_game(path[1]).to_dict()})
+                small_blind, big_blind = self._extract_blinds(body)
+                summaries = MANAGER.run_hands(
+                    path[1],
+                    count,
+                    until_one_left,
+                    small_blind=small_blind,
+                    big_blind=big_blind,
+                )
+                self._json({"hands": summaries, "game": MANAGER.get_game_state(path[1])})
                 return
             self._json({"error": "not found"}, HTTPStatus.NOT_FOUND)
         except KeyError as exc:
@@ -78,6 +92,25 @@ class PokerRequestHandler(BaseHTTPRequestHandler):
             raise ValueError("request body must be a JSON object")
         return payload
 
+    @staticmethod
+    def _extract_blinds(body: dict[str, Any]) -> tuple[int | None, int | None]:
+        """Parse optional blind overrides from a /hands POST body.
+
+        Callers may omit both keys (keep current level), or supply both to
+        raise the blinds for the upcoming batch. Providing only one is
+        treated as a client error and surfaced via ``ValueError`` so the
+        handler can reply with 400.
+        """
+        raw_small = body.get("small_blind")
+        raw_big = body.get("big_blind")
+        if raw_small is None and raw_big is None:
+            return None, None
+        if raw_small is None or raw_big is None:
+            raise ValueError(
+                "small_blind and big_blind must be provided together"
+            )
+        return int(raw_small), int(raw_big)
+
     def _json(self, payload: dict[str, Any], status: HTTPStatus = HTTPStatus.OK) -> None:
         body = json.dumps(payload, ensure_ascii=True).encode("utf-8")
         self.send_response(status)

texas_holdem/service.py

@@ -5,13 +5,14 @@ from threading import RLock
 from typing import Any
 from uuid import uuid4
 
-from texas_holdem.agents import build_agent
+from texas_holdem.agents import build_agent, http_agent_endpoint_from_spec
 from texas_holdem.engine import TableGame
 
 
 class GameManager:
     def __init__(self) -> None:
         self._games: dict[str, TableGame] = {}
+        self._http_endpoint_owners: dict[str, str] = {}
         self._lock = RLock()
 
     def create_game(self, payload: dict[str, Any]) -> TableGame:
@@ -29,12 +30,19 @@ class GameManager:
         big_blind = int(payload.get("big_blind", 10))
 
         specs = []
+        http_endpoints: set[str] = set()
         for seat, raw_spec in enumerate(players):
             if not isinstance(raw_spec, dict):
                 raise ValueError("each player must be an object")
             player_id = str(raw_spec.get("id") or raw_spec.get("player_id") or f"p{seat + 1}")
             name = str(raw_spec.get("name") or player_id)
-            agent = build_agent(raw_spec.get("agent", raw_spec), rng)
+            agent_spec = raw_spec.get("agent", raw_spec)
+            if not isinstance(agent_spec, dict):
+                raise ValueError("agent spec must be an object")
+            endpoint = http_agent_endpoint_from_spec(agent_spec)
+            if endpoint is not None:
+                http_endpoints.add(endpoint)
+            agent = build_agent(agent_spec, rng, player_id=player_id)
             specs.append((player_id, name, agent))
 
         game = TableGame(
@@ -46,9 +54,18 @@ class GameManager:
             rng=rng,
         )
         with self._lock:
+            self._release_completed_http_endpoints_locked()
             if game_id in self._games:
                 raise ValueError(f"game already exists: {game_id}")
+            for endpoint in http_endpoints:
+                owner = self._http_endpoint_owners.get(endpoint)
+                if owner is not None and owner != game_id:
+                    raise ValueError(
+                        f"http agent endpoint already belongs to game {owner}: {endpoint}"
+                    )
             self._games[game_id] = game
+            for endpoint in http_endpoints:
+                self._http_endpoint_owners[endpoint] = game_id
         return game
 
     def get_game(self, game_id: str) -> TableGame:
@@ -58,14 +75,54 @@ class GameManager:
             except KeyError as exc:
                 raise KeyError(f"game not found: {game_id}") from exc
 
+    def get_game_state(self, game_id: str) -> dict[str, object]:
+        return self.get_game(game_id).snapshot_completed()
+
     def list_games(self) -> list[dict[str, object]]:
         with self._lock:
-            return [game.to_dict() for game in self._games.values()]
+            games = list(self._games.values())
+        return [game.snapshot_completed() for game in games]
 
-    def run_hands(self, game_id: str, count: int = 1, until_one_left: bool = False) -> list[dict[str, object]]:
+    def run_hands(
+        self,
+        game_id: str,
+        count: int = 1,
+        until_one_left: bool = False,
+        small_blind: int | None = None,
+        big_blind: int | None = None,
+    ) -> list[dict[str, object]]:
+        """Run ``count`` hands, optionally raising the blinds first.
+
+        ``small_blind`` / ``big_blind`` are forwarded to the engine so the
+        blinds can change between batches. Leaving them as ``None`` keeps
+        the previously configured level, which preserves the original
+        no-argument behaviour.
+        """
         game = self.get_game(game_id)
-        with self._lock:
-            return [
-                summary.to_dict()
-                for summary in game.run_hands(count, until_one_left=until_one_left)
-            ]
+        summaries = [
+            summary.to_dict()
+            for summary in game.run_hands(
+                count,
+                until_one_left=until_one_left,
+                small_blind=small_blind,
+                big_blind=big_blind,
+            )
+        ]
+        if game.is_complete:
+            with self._lock:
+                self._release_http_endpoints_for_game_locked(game_id)
+        return summaries
+
+    def _release_completed_http_endpoints_locked(self) -> None:
+        for game_id, game in list(self._games.items()):
+            if game.lock.acquire(blocking=False):
+                try:
+                    if game.is_complete:
+                        self._release_http_endpoints_for_game_locked(game_id)
+                finally:
+                    game.lock.release()
+
+    def _release_http_endpoints_for_game_locked(self, game_id: str) -> None:
+        for endpoint, owner in list(self._http_endpoint_owners.items()):
+            if owner == game_id:
+                del self._http_endpoint_owners[endpoint]