Ready for first real test

tools/python_exec.py

@@ -1,13 +1,14 @@
 """run_python — the calculator. Agents call this for ALL arithmetic instead of doing
 mental math (LLMs are unreliable at it). Executes in an isolated subprocess with a hard
-timeout. Kept math-only by convention; runs model-generated code on this machine, so
-scope is deliberately narrow.
+timeout and a **scrubbed environment** (no account secrets), so model-generated code can't
+read API keys. For heavier data/analysis work, agents use run_analysis (tools/sandbox.py).
 """
 
 from __future__ import annotations
 
 import subprocess
 import sys
+import tempfile
 
 from langchain_core.tools import tool
 
@@ -20,14 +21,17 @@ def run_python(code: str) -> str:
 
     Use for ANY arithmetic: position sizing, percentages, P/E, returns, averages, ratios.
     You MUST print() the result. `math` and `statistics` are importable.
-    Example: `print(round(1000 / 23.55, 4))`. No network or file access; 5s limit.
+    Example: `print(round(1000 / 23.55, 4))`. No access to secrets; 5s limit.
     """
+    from tools.sandbox import scrubbed_env
+
     try:
         proc = subprocess.run(
             [sys.executable, "-I", "-c", code],
             capture_output=True,
             text=True,
             timeout=_TIMEOUT_S,
+            env=scrubbed_env(tempfile.gettempdir()),
         )
     except subprocess.TimeoutExpired:
         return f"ERROR: execution timed out ({_TIMEOUT_S}s limit)."