Spaces:
Running
Running
Charles Grandjean commited on
Commit Β·
acc767b
1
Parent(s): 984df29
fixes
Browse files- subagents/doc_editor.py +4 -15
- tests/test_bug_fixes.py +239 -0
- utils/doc_editor_tools.py +38 -0
subagents/doc_editor.py
CHANGED
|
@@ -13,7 +13,7 @@ from langchain_core.messages import SystemMessage, HumanMessage, AIMessage, Tool
|
|
| 13 |
from agent_states.doc_editor_state import DocEditorState
|
| 14 |
from utils.doc_editor_tools import (
|
| 15 |
replace_html, add_html, delete_html, inspect_document, attempt_completion,
|
| 16 |
-
_replace_html, _add_html, _delete_html
|
| 17 |
)
|
| 18 |
from prompts.doc_editor import get_doc_editor_system_prompt, get_summary_system_prompt
|
| 19 |
|
|
@@ -40,7 +40,7 @@ class DocumentEditorAgent:
|
|
| 40 |
"""
|
| 41 |
self.llm = llm
|
| 42 |
# Real tools (with _ prefix for internal functions) - these are actually called
|
| 43 |
-
self.tools_real = [_replace_html, _add_html, _delete_html,
|
| 44 |
# Facade tools (exposed to LLM without doc_text parameter)
|
| 45 |
self.tools_facade = [replace_html, add_html, delete_html, inspect_document, attempt_completion]
|
| 46 |
self.llm_with_tools = self.llm.bind_tools(self.tools_facade)
|
|
@@ -181,25 +181,14 @@ class DocumentEditorAgent:
|
|
| 181 |
for tool_call in last_message.tool_calls:
|
| 182 |
tool_name = tool_call['name']
|
| 183 |
|
| 184 |
-
# Special case: inspect_document - add doc_text to message stack
|
| 185 |
-
if tool_name == "inspect_document":
|
| 186 |
-
intermediate_steps.append(
|
| 187 |
-
ToolMessage(
|
| 188 |
-
content=f"Current document:\n{state['doc_text']}",
|
| 189 |
-
tool_call_id=tool_call['id'],
|
| 190 |
-
name=tool_name
|
| 191 |
-
)
|
| 192 |
-
)
|
| 193 |
-
continue
|
| 194 |
-
|
| 195 |
# Get the real tool function (with _ prefix)
|
| 196 |
real_tool_func = next((t for t in self.tools_real if t.__name__ == "_" + tool_name), None)
|
| 197 |
|
| 198 |
if real_tool_func:
|
| 199 |
args = tool_call['args'].copy()
|
| 200 |
|
| 201 |
-
# Inject doc_text for editing tools
|
| 202 |
-
if tool_name in ["replace_html", "add_html", "delete_html"]:
|
| 203 |
args["doc_text"] = state["doc_text"]
|
| 204 |
logger.info(f"π Injecting doc_text ({len(state['doc_text'])}b) into {tool_name}")
|
| 205 |
|
|
|
|
| 13 |
from agent_states.doc_editor_state import DocEditorState
|
| 14 |
from utils.doc_editor_tools import (
|
| 15 |
replace_html, add_html, delete_html, inspect_document, attempt_completion,
|
| 16 |
+
_replace_html, _add_html, _delete_html, _inspect_document, _attempt_completion
|
| 17 |
)
|
| 18 |
from prompts.doc_editor import get_doc_editor_system_prompt, get_summary_system_prompt
|
| 19 |
|
|
|
|
| 40 |
"""
|
| 41 |
self.llm = llm
|
| 42 |
# Real tools (with _ prefix for internal functions) - these are actually called
|
| 43 |
+
self.tools_real = [_replace_html, _add_html, _delete_html, _inspect_document, _attempt_completion]
|
| 44 |
# Facade tools (exposed to LLM without doc_text parameter)
|
| 45 |
self.tools_facade = [replace_html, add_html, delete_html, inspect_document, attempt_completion]
|
| 46 |
self.llm_with_tools = self.llm.bind_tools(self.tools_facade)
|
|
|
|
| 181 |
for tool_call in last_message.tool_calls:
|
| 182 |
tool_name = tool_call['name']
|
| 183 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 184 |
# Get the real tool function (with _ prefix)
|
| 185 |
real_tool_func = next((t for t in self.tools_real if t.__name__ == "_" + tool_name), None)
|
| 186 |
|
| 187 |
if real_tool_func:
|
| 188 |
args = tool_call['args'].copy()
|
| 189 |
|
| 190 |
+
# Inject doc_text for editing tools and inspect_document
|
| 191 |
+
if tool_name in ["replace_html", "add_html", "delete_html", "inspect_document"]:
|
| 192 |
args["doc_text"] = state["doc_text"]
|
| 193 |
logger.info(f"π Injecting doc_text ({len(state['doc_text'])}b) into {tool_name}")
|
| 194 |
|
tests/test_bug_fixes.py
ADDED
|
@@ -0,0 +1,239 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
#!/usr/bin/env python3
|
| 2 |
+
"""
|
| 3 |
+
Test script to verify bug fixes for document editor tools
|
| 4 |
+
Tests that:
|
| 5 |
+
1. All internal functions exist and have __name__ attribute
|
| 6 |
+
2. tools_real contains only functions (no StructuredTools)
|
| 7 |
+
3. Workflow builds successfully
|
| 8 |
+
"""
|
| 9 |
+
|
| 10 |
+
import sys
|
| 11 |
+
import os
|
| 12 |
+
from pathlib import Path
|
| 13 |
+
|
| 14 |
+
# Add parent directory to path
|
| 15 |
+
sys.path.insert(0, str(Path(__file__).parent.parent))
|
| 16 |
+
|
| 17 |
+
def test_internal_functions_exist():
|
| 18 |
+
"""Test that all internal functions exist and have __name__"""
|
| 19 |
+
print("\n" + "=" * 80)
|
| 20 |
+
print("TEST 1: Internal Functions Exist and Have __name__")
|
| 21 |
+
print("=" * 80)
|
| 22 |
+
|
| 23 |
+
from utils.doc_editor_tools import (
|
| 24 |
+
_replace_html, _add_html, _delete_html, _inspect_document, _attempt_completion
|
| 25 |
+
)
|
| 26 |
+
|
| 27 |
+
functions = [
|
| 28 |
+
("_replace_html", _replace_html),
|
| 29 |
+
("_add_html", _add_html),
|
| 30 |
+
("_delete_html", _delete_html),
|
| 31 |
+
("_inspect_document", _inspect_document),
|
| 32 |
+
("_attempt_completion", _attempt_completion)
|
| 33 |
+
]
|
| 34 |
+
|
| 35 |
+
all_passed = True
|
| 36 |
+
for name, func in functions:
|
| 37 |
+
try:
|
| 38 |
+
func_name = func.__name__
|
| 39 |
+
print(f"β
{name}: __name__ = '{func_name}'")
|
| 40 |
+
except AttributeError as e:
|
| 41 |
+
print(f"β {name}: {e}")
|
| 42 |
+
all_passed = False
|
| 43 |
+
|
| 44 |
+
return all_passed
|
| 45 |
+
|
| 46 |
+
|
| 47 |
+
def test_tools_real_are_functions():
|
| 48 |
+
"""Test that tools_real contains only functions"""
|
| 49 |
+
print("\n" + "=" * 80)
|
| 50 |
+
print("TEST 2: tools_real Contains Only Functions")
|
| 51 |
+
print("=" * 80)
|
| 52 |
+
|
| 53 |
+
# Import agent - this will fail if tools_real has StructuredTools
|
| 54 |
+
try:
|
| 55 |
+
from subagents.doc_editor import DocumentEditorAgent
|
| 56 |
+
|
| 57 |
+
# We need a mock LLM to initialize
|
| 58 |
+
class MockLLM:
|
| 59 |
+
def bind_tools(self, tools):
|
| 60 |
+
return self
|
| 61 |
+
|
| 62 |
+
async def ainvoke(self, messages):
|
| 63 |
+
from langchain_core.messages import AIMessage
|
| 64 |
+
return AIMessage(content="Test response")
|
| 65 |
+
|
| 66 |
+
llm = MockLLM()
|
| 67 |
+
agent = DocumentEditorAgent(llm=llm)
|
| 68 |
+
|
| 69 |
+
print(f"β
Agent initialized successfully")
|
| 70 |
+
print(f"π¦ tools_real has {len(agent.tools_real)} items")
|
| 71 |
+
|
| 72 |
+
# Check that all are functions
|
| 73 |
+
all_passed = True
|
| 74 |
+
for i, tool in enumerate(agent.tools_real):
|
| 75 |
+
try:
|
| 76 |
+
name = tool.__name__
|
| 77 |
+
print(f"β
Tool {i}: {name} (has __name__)")
|
| 78 |
+
except AttributeError as e:
|
| 79 |
+
print(f"β Tool {i}: {e}")
|
| 80 |
+
all_passed = False
|
| 81 |
+
|
| 82 |
+
return all_passed
|
| 83 |
+
|
| 84 |
+
except Exception as e:
|
| 85 |
+
print(f"β Failed to initialize agent: {e}")
|
| 86 |
+
import traceback
|
| 87 |
+
traceback.print_exc()
|
| 88 |
+
return False
|
| 89 |
+
|
| 90 |
+
|
| 91 |
+
def test_workflow_builds():
|
| 92 |
+
"""Test that workflow builds successfully"""
|
| 93 |
+
print("\n" + "=" * 80)
|
| 94 |
+
print("TEST 3: Workflow Builds Successfully")
|
| 95 |
+
print("=" * 80)
|
| 96 |
+
|
| 97 |
+
try:
|
| 98 |
+
from subagents.doc_editor import DocumentEditorAgent
|
| 99 |
+
|
| 100 |
+
class MockLLM:
|
| 101 |
+
def bind_tools(self, tools):
|
| 102 |
+
return self
|
| 103 |
+
|
| 104 |
+
async def ainvoke(self, messages):
|
| 105 |
+
from langchain_core.messages import AIMessage
|
| 106 |
+
return AIMessage(content="Test response")
|
| 107 |
+
|
| 108 |
+
llm = MockLLM()
|
| 109 |
+
agent = DocumentEditorAgent(llm=llm)
|
| 110 |
+
|
| 111 |
+
print(f"β
Workflow built successfully")
|
| 112 |
+
# Note: CompiledStateGraph doesn't expose nodes() and edges() methods directly
|
| 113 |
+
# The important thing is that it built without errors
|
| 114 |
+
|
| 115 |
+
return True
|
| 116 |
+
|
| 117 |
+
except Exception as e:
|
| 118 |
+
print(f"β Failed to build workflow: {e}")
|
| 119 |
+
import traceback
|
| 120 |
+
traceback.print_exc()
|
| 121 |
+
return False
|
| 122 |
+
|
| 123 |
+
|
| 124 |
+
def test_tools_callable():
|
| 125 |
+
"""Test that internal tools are callable"""
|
| 126 |
+
print("\n" + "=" * 80)
|
| 127 |
+
print("TEST 4: Internal Tools Are Callable")
|
| 128 |
+
print("=" * 80)
|
| 129 |
+
|
| 130 |
+
from utils.doc_editor_tools import (
|
| 131 |
+
_replace_html, _add_html, _delete_html, _inspect_document, _attempt_completion
|
| 132 |
+
)
|
| 133 |
+
|
| 134 |
+
import asyncio
|
| 135 |
+
|
| 136 |
+
# Test _replace_html
|
| 137 |
+
async def test_replace():
|
| 138 |
+
result = await _replace_html(
|
| 139 |
+
doc_text="Hello World",
|
| 140 |
+
search="World",
|
| 141 |
+
replace="Universe",
|
| 142 |
+
expected_matches=1
|
| 143 |
+
)
|
| 144 |
+
return result
|
| 145 |
+
|
| 146 |
+
# Test _inspect_document
|
| 147 |
+
async def test_inspect():
|
| 148 |
+
result = await _inspect_document(
|
| 149 |
+
doc_text="Test document"
|
| 150 |
+
)
|
| 151 |
+
return result
|
| 152 |
+
|
| 153 |
+
# Test _attempt_completion
|
| 154 |
+
async def test_attempt():
|
| 155 |
+
result = await _attempt_completion(
|
| 156 |
+
message="Test complete"
|
| 157 |
+
)
|
| 158 |
+
return result
|
| 159 |
+
|
| 160 |
+
all_passed = True
|
| 161 |
+
|
| 162 |
+
# Run tests
|
| 163 |
+
print("\nTesting _replace_html...")
|
| 164 |
+
try:
|
| 165 |
+
result = asyncio.run(test_replace())
|
| 166 |
+
if result.get("ok"):
|
| 167 |
+
print(f"β
_replace_html returned: {result}")
|
| 168 |
+
else:
|
| 169 |
+
print(f"β οΈ _replace_html failed (expected for invalid HTML): {result}")
|
| 170 |
+
except Exception as e:
|
| 171 |
+
print(f"β _replace_html error: {e}")
|
| 172 |
+
all_passed = False
|
| 173 |
+
|
| 174 |
+
print("\nTesting _inspect_document...")
|
| 175 |
+
try:
|
| 176 |
+
result = asyncio.run(test_inspect())
|
| 177 |
+
if result.get("ok"):
|
| 178 |
+
print(f"β
_inspect_document returned: {result}")
|
| 179 |
+
else:
|
| 180 |
+
print(f"β _inspect_document failed: {result}")
|
| 181 |
+
all_passed = False
|
| 182 |
+
except Exception as e:
|
| 183 |
+
print(f"β _inspect_document error: {e}")
|
| 184 |
+
all_passed = False
|
| 185 |
+
|
| 186 |
+
print("\nTesting _attempt_completion...")
|
| 187 |
+
try:
|
| 188 |
+
result = asyncio.run(test_attempt())
|
| 189 |
+
if result.get("ok"):
|
| 190 |
+
print(f"β
_attempt_completion returned: {result}")
|
| 191 |
+
else:
|
| 192 |
+
print(f"β _attempt_completion failed: {result}")
|
| 193 |
+
all_passed = False
|
| 194 |
+
except Exception as e:
|
| 195 |
+
print(f"β _attempt_completion error: {e}")
|
| 196 |
+
all_passed = False
|
| 197 |
+
|
| 198 |
+
return all_passed
|
| 199 |
+
|
| 200 |
+
|
| 201 |
+
if __name__ == "__main__":
|
| 202 |
+
print("\n" + "=" * 80)
|
| 203 |
+
print("BUG FIX VERIFICATION TESTS")
|
| 204 |
+
print("=" * 80)
|
| 205 |
+
|
| 206 |
+
results = {}
|
| 207 |
+
|
| 208 |
+
# Test 1: Internal functions exist
|
| 209 |
+
results["test_internal_functions_exist"] = test_internal_functions_exist()
|
| 210 |
+
|
| 211 |
+
# Test 2: tools_real contains only functions
|
| 212 |
+
results["test_tools_real_are_functions"] = test_tools_real_are_functions()
|
| 213 |
+
|
| 214 |
+
# Test 3: Workflow builds
|
| 215 |
+
results["test_workflow_builds"] = test_workflow_builds()
|
| 216 |
+
|
| 217 |
+
# Test 4: Tools are callable
|
| 218 |
+
results["test_tools_callable"] = test_tools_callable()
|
| 219 |
+
|
| 220 |
+
# Summary
|
| 221 |
+
print("\n" + "=" * 80)
|
| 222 |
+
print("TEST SUMMARY")
|
| 223 |
+
print("=" * 80)
|
| 224 |
+
|
| 225 |
+
passed = sum(1 for v in results.values() if v)
|
| 226 |
+
total = len(results)
|
| 227 |
+
|
| 228 |
+
for test_name, result in results.items():
|
| 229 |
+
status = "β
PASSED" if result else "β FAILED"
|
| 230 |
+
print(f"{status}: {test_name}")
|
| 231 |
+
|
| 232 |
+
print("\n" + "=" * 80)
|
| 233 |
+
if passed == total:
|
| 234 |
+
print(f"β
ALL TESTS PASSED ({passed}/{total})")
|
| 235 |
+
else:
|
| 236 |
+
print(f"β SOME TESTS FAILED ({passed}/{total} passed)")
|
| 237 |
+
print("=" * 80 + "\n")
|
| 238 |
+
|
| 239 |
+
sys.exit(0 if passed == total else 1)
|
utils/doc_editor_tools.py
CHANGED
|
@@ -187,6 +187,44 @@ async def _delete_html(doc_text: str, search: str, expected_matches: int = 1) ->
|
|
| 187 |
return await _replace_html(doc_text, search, "", expected_matches)
|
| 188 |
|
| 189 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 190 |
@tool
|
| 191 |
async def replace_html(search: str, replace: str, expected_matches: int = 1) -> Dict[str, Any]:
|
| 192 |
"""
|
|
|
|
| 187 |
return await _replace_html(doc_text, search, "", expected_matches)
|
| 188 |
|
| 189 |
|
| 190 |
+
async def _inspect_document(doc_text: str) -> Dict[str, Any]:
|
| 191 |
+
"""
|
| 192 |
+
Internal implementation of inspect_document (with doc_text parameter).
|
| 193 |
+
|
| 194 |
+
Returns the current document state for inspection.
|
| 195 |
+
|
| 196 |
+
Args:
|
| 197 |
+
doc_text: The HTML document content
|
| 198 |
+
|
| 199 |
+
Returns:
|
| 200 |
+
Dict with 'ok' (bool) and 'content' (str)
|
| 201 |
+
"""
|
| 202 |
+
logger.info(f" π inspect_document | size:{len(doc_text)}b")
|
| 203 |
+
return {
|
| 204 |
+
"ok": True,
|
| 205 |
+
"content": doc_text
|
| 206 |
+
}
|
| 207 |
+
|
| 208 |
+
|
| 209 |
+
async def _attempt_completion(message: str) -> Dict[str, Any]:
|
| 210 |
+
"""
|
| 211 |
+
Internal implementation of attempt_completion.
|
| 212 |
+
|
| 213 |
+
Signals that document editing is complete.
|
| 214 |
+
|
| 215 |
+
Args:
|
| 216 |
+
message: Summary message describing what was changed
|
| 217 |
+
|
| 218 |
+
Returns:
|
| 219 |
+
Dict with 'ok' (bool) and 'message' (str)
|
| 220 |
+
"""
|
| 221 |
+
logger.info(f" β
attempt_completion | {message}")
|
| 222 |
+
return {
|
| 223 |
+
"ok": True,
|
| 224 |
+
"message": message
|
| 225 |
+
}
|
| 226 |
+
|
| 227 |
+
|
| 228 |
@tool
|
| 229 |
async def replace_html(search: str, replace: str, expected_matches: int = 1) -> Dict[str, Any]:
|
| 230 |
"""
|