tenseleyflow/loader / 515c51e

Browse files

fix: improve content extraction for bracket-format write tool calls

- Better parsing of file_path with quotes
- Walk backward to find matching end quote for content
- Add smarter bracket patterns with lookahead for common endings
- Add debug logging for write tool extraction
Authored by espadonne
SHA
515c51e3c1f96d4022f7cbcd23a9fd85af25743f
Parents
2031f1a
Tree
28bf827

2 changed files

StatusFile+-
A index.html 1 0
M src/loader/agent/loop.py 33 7
index.htmladded
@@ -0,0 +1,1 @@
1
+<html><body><img src='sort.jpg' style='display: block; margin: auto;
src/loader/agent/loop.pymodified
@@ -1441,10 +1441,13 @@ class Agent:
14411441
14421442
         # First, try to extract bracket format: [calls bash tool with: ...]
14431443
         # or [USE bash tool: ...] or similar variations
1444
+        # Note: Using (.+?) with re.DOTALL to capture content that may span patterns
1445
+        # The ] at end acts as anchor, but we need to handle ] inside content
14441446
         bracket_patterns = [
1447
+            r'\[calls?\s+(\w+)\s+tool\s+with:\s*(.+?)\](?=\s*(?:\n|$|[A-Z]|Done|Created|Error))',
1448
+            r'\[USE\s+(\w+)\s+tool:\s*(.+?)\](?=\s*(?:\n|$|[A-Z]|Done|Created|Error))',
14451449
             r'\[calls?\s+(\w+)\s+tool\s+with:\s*([^\]]+)\]',
14461450
             r'\[USE\s+(\w+)\s+tool:\s*([^\]]+)\]',
1447
-            r'\[(\w+)\s+tool:\s*([^\]]+)\]',
14481451
         ]
14491452
14501453
         for pattern in bracket_patterns:
@@ -1469,15 +1472,38 @@ class Agent:
14691472
                         ))
14701473
                     elif tool_name == "write":
14711474
                         # write tool: file_path=..., content="..."
1472
-                        file_path_match = re.search(r'file_path[=:]\s*([^,\s]+)', args_str)
1473
-                        content_match = re.search(r'content[=:]\s*["\'](.+)["\']', args_str, re.DOTALL)
1474
-                        if not content_match:
1475
-                            # Try without quotes
1476
-                            content_match = re.search(r'content[=:]\s*(.+)', args_str, re.DOTALL)
1475
+                        # Handle quoted file paths
1476
+                        file_path_match = re.search(r'file_path[=:]\s*["\']?([^"\'`,]+)["\']?', args_str)
1477
+
1478
+                        # For content, find the content= part and extract everything after it
1479
+                        # Handle both quoted and unquoted content
1480
+                        content_start = re.search(r'content[=:]\s*', args_str)
1481
+                        file_content = ""
1482
+                        if content_start:
1483
+                            rest = args_str[content_start.end():]
1484
+                            # Check if content starts with a quote
1485
+                            if rest.startswith('"'):
1486
+                                # Find matching end quote (handle escaped quotes)
1487
+                                end_idx = len(rest) - 1
1488
+                                # Walk backward to find the last quote
1489
+                                while end_idx > 0 and rest[end_idx] != '"':
1490
+                                    end_idx -= 1
1491
+                                if end_idx > 0:
1492
+                                    file_content = rest[1:end_idx]
1493
+                            elif rest.startswith("'"):
1494
+                                end_idx = len(rest) - 1
1495
+                                while end_idx > 0 and rest[end_idx] != "'":
1496
+                                    end_idx -= 1
1497
+                                if end_idx > 0:
1498
+                                    file_content = rest[1:end_idx]
1499
+                            else:
1500
+                                # No quotes - take everything
1501
+                                file_content = rest.strip()
1502
+
1503
+                        debug(f"  write: file_path={file_path_match.group(1) if file_path_match else None}, content_len={len(file_content)}")
14771504
14781505
                         if file_path_match:
14791506
                             file_path = file_path_match.group(1).strip('"\'')
1480
-                            file_content = content_match.group(1) if content_match else ""
14811507
                             tool_calls.append(ToolCall(
14821508
                                 id=f"bracket_{tool_name}_{len(tool_calls)}",
14831509
                                 name=tool_name,