Fix deployment and load latest Codex models

bot/codex_runner.py

@@ -1,11 +1,13 @@
 """Runs tasks through the OpenAI Codex Python SDK and yields progress events."""
 
 import asyncio
+import shutil
 from dataclasses import dataclass
 from typing import Any, AsyncIterator
 
 from openai_codex import ApprovalMode, AsyncCodex, CodexConfig, Sandbox
-from openai_codex.types import ReasoningEffort, ThreadItem, ThreadTokenUsage, Turn
+from openai_codex.types import ThreadItem, ThreadTokenUsage, Turn
+from pydantic import BaseModel, ConfigDict, Field
 
 from . import config
 
@@ -73,6 +75,32 @@ class ModelOptions:
         return tuple(efforts)
 
 
+class _ModelEffort(BaseModel):
+    model_config = ConfigDict(populate_by_name=True, extra="ignore")
+
+    reasoning_effort: str = Field(alias="reasoningEffort")
+
+
+class _Model(BaseModel):
+    model_config = ConfigDict(populate_by_name=True, extra="ignore")
+
+    id: str
+    model: str
+    display_name: str = Field(alias="displayName")
+    hidden: bool = False
+    is_default: bool = Field(default=False, alias="isDefault")
+    supported_reasoning_efforts: list[_ModelEffort] = Field(
+        default_factory=list,
+        alias="supportedReasoningEfforts",
+    )
+
+
+class _ModelListResponse(BaseModel):
+    model_config = ConfigDict(populate_by_name=True, extra="ignore")
+
+    data: list[_Model]
+
+
 def _shorten(text: str, limit: int = TOOL_INPUT_PREVIEW_LEN) -> str:
     text = " ".join(text.split())
     if len(text) > limit:
@@ -193,24 +221,15 @@ def _approval_mode() -> ApprovalMode:
         ) from exc
 
 
-def _effort(value: str | None) -> ReasoningEffort | None:
-    if value is None:
-        return None
-    try:
-        return ReasoningEffort(value)
-    except ValueError as exc:
-        valid = ", ".join(effort.value for effort in ReasoningEffort)
-        raise RuntimeError(f"Unsupported reasoning effort={value!r}; use one of: {valid}") from exc
-
-
 def _codex_config() -> CodexConfig:
-    return CodexConfig(codex_bin=config.CODEX_BIN)
+    return CodexConfig(codex_bin=config.CODEX_BIN or shutil.which("codex"))
 
 
 def _reasoning_efforts(raw_options: Any) -> tuple[str, ...]:
     efforts: list[str] = []
     for option in raw_options or []:
-        value = getattr(getattr(option, "reasoning_effort", None), "value", None)
+        raw_value = getattr(option, "reasoning_effort", None)
+        value = getattr(raw_value, "value", raw_value)
         if value and value not in efforts:
             efforts.append(value)
     return tuple(efforts)
@@ -223,7 +242,14 @@ async def list_model_options() -> ModelOptions:
             if config.OPENAI_API_KEY:
                 await codex.login_api_key(config.OPENAI_API_KEY)
 
-            response = await codex.models()
+            # The app-server catalog evolves faster than the beta Python SDK's
+            # generated enums. Parse this one response permissively so newly
+            # introduced reasoning levels appear without an SDK release.
+            response = await codex._client.request(
+                "model/list",
+                {"includeHidden": False},
+                response_model=_ModelListResponse,
+            )
     except Exception as exc:
         return ModelOptions(models=(), error=str(exc))
 
@@ -318,7 +344,6 @@ async def run_task(
             prompt,
             approval_mode=_approval_mode(),
             cwd=config.WORK_DIR,
-            effort=_effort(effort),
             model=selected_model,
             sandbox=_sandbox(),
         )