repl

libs/tools.py

@@ -4,19 +4,22 @@ from re import search
 from dataclasses import dataclass
 from typing import Union
 
+from langchain_core.tools import Tool
+from langchain_experimental.utilities import PythonREPL
+
 @tool
 def add(a: float, b: float) -> str:
     """Adds a+b and returns the sum"""
     af = float(a)
     bf = float(b)
-    return f"{a} + {b} = {a+b}"
+    return f"{af} + {af} = {af+bf}"
 
 @tool
 def multiply(a: float, b: float) -> str:
     """Multiplies a*b and returns the product"""
     af = float(a)
     bf = float(b)
-    return f"{a} * {b} = {a*b}"
+    return f"{af} * {bf} = {af*bf}"
 
 @tool
 def get_current_date_and_time() -> str:
@@ -99,10 +102,13 @@ def get_notes_in_timespan(begin: str, to: str) -> str:
     try:
         begin_d = datetime.strptime(begin, "%Y/%m/%d")
         to_d = datetime.strptime(to+" 23:59", "%Y/%m/%d %H:%M")
-    except: return "Error: Invalid input. Date format is %Y/%m/%d"
+    except ValueError:
+        return "Error: Invalid input. Date format is %Y/%m/%d"
 
-    try: assert begin_d < to_d
+    try:
-    except: return "Error: from time has to be before to time."
+        assert begin_d < to_d
+    except AssertionError:
+        return "Error: from time has to be before to time."
 
     filtered_entries = [entry for entry in note_entries if begin_d <= entry.time <= to_d]
 
@@ -128,9 +134,12 @@ def get_notes_containing(patterns: Union[list[str], str]) -> str:
     exaples:
     {"patterns": [ "Aunt(ie)?", "Sabine" ]} # Looks for Notes related to Aunt Sabine"""
 
-    if isinstance(patterns, list): big_pattern = '|'.join(f"({s})" for s in patterns)
+    if isinstance(patterns, list):
-    elif isinstance(patterns, str): big_pattern = patterns
+        big_pattern = '|'.join(f"({s})" for s in patterns)
-    else: return f"Error: Invalid Input type. `patterns` can either be a list of strings or a single string. But got {type(patterns)}."
+    elif isinstance(patterns, str):
+        big_pattern = patterns
+    else:
+        return f"Error: Invalid Input type. `patterns` can either be a list of strings or a single string. But got {type(patterns)}."
 
     filtered_entries = [entry for entry in note_entries if search(big_pattern.lower(), entry.content.lower())]
 
@@ -147,7 +156,29 @@ def get_notes_containing(patterns: Union[list[str], str]) -> str:
     return ret
                                
 @tool
-def write_note(content: str) -> str:
+def write_note(command: str) -> str:
     """Write a not with the current time to the database."""
-    return content
+    return command
+                               
+@tool
+def save_python_repl(command: str):
+    """Simulates the normal python repl but with certain patterns blocked for savety reasons"""
+    python_repl = PythonREPL()
+    blocked_patterns = [
+        "^os\\.",
+        "^subprocess\\.",
+        "^with open\\(",
+    ]
+    valid = True
+    for pattern in blocked_patterns:
+        if search(pattern, command):
+             valid = False
+             break
+
+    if valid:
+        return python_repl.run(command)
+    else:
+        return f"Command not executed, becaise the blocked pattern `{pattern}` was found in the command."
+    
+
 

suite_settings/tests.py

@@ -1,7 +1,7 @@
 from libs.classes import Test
 from libs.runnables import basic_prompt, one_tool_call_answer, agent_with_tools
 from libs.validators import regex_match_any, system_human_answer_match
-from libs.tools import add, multiply, get_current_date_and_time, get_notes_in_timespan, get_notes_containing, write_note
+from libs.tools import add, multiply, get_current_date_and_time, get_notes_in_timespan, get_notes_containing, write_note, save_python_repl
 from textwrap import dedent
 from langchain_core.messages import HumanMessage, SystemMessage, ToolMessage, AIMessage
 
@@ -49,6 +49,19 @@ tests = {
         validator=regex_match_any,
         validation_input={"patterns": ["6134205", "6.134.205", "6,134,205"]},
     ),
+    363: Test(
+        name="Complex Multiplication Python",
+        runnable=one_tool_call_answer,
+        runnable_input={
+            "system_msg": 'You are a helpful assistant.',
+            "human_msg": 'Is 31515261 divisible by 425? If not, whats the remainder?',
+            "tools": { "python_repl": save_python_repl },
+        },
+        validator=regex_match_any,
+        validation_input={
+            "patterns": [ "236", "two ?hundred and thirty ?six", "two ?hundred thirty ?six" ]
+        }
+    ),
     283: Test(
         name="Notes from last Saturday",
         runnable=agent_with_tools,
@@ -65,7 +78,7 @@ tests = {
         validator=system_human_answer_match,
         validation_input={
             "criteria": dedent("""- containing the information that the Human should call Wolfgang 
-    - just one single conversational answer, without any AI fragments (A/B versions, "end of message" parts, unfitting discalimers or notes, what specific tool was used to get the answer, etc.)""")
+            - just one single conversational answer, without any AI fragments (A/B versions, "end of message" parts, unfitting discalimers or notes, what specific tool was used to get the answer, etc.)""")
         },
     ),
     260: Test(
@@ -119,7 +132,6 @@ tests = {
             - just one single conversational answer, without any AI fragments (A/B versions, "end of message" parts, unfitting discalimers or notes, what specific tool was used to get the answer, etc.)""")
         },
     ),
-    # 363: Test(),
     # 600: Test(),
     # 221: Test(),
     # 985: Test(),