diff --git a/EvoScientist/stream/formatter.py b/EvoScientist/stream/formatter.py
index f0fdf5e..8035482 100644
--- a/EvoScientist/stream/formatter.py
+++ b/EvoScientist/stream/formatter.py
@@ -109,12 +109,15 @@ class ToolResultFormatter:
         return False
 
     def _is_error(self, content: str) -> bool:
+        head = "\n".join(content.splitlines()[:3])
         error_patterns = [
             'Traceback (most recent call last)',
             'Exception:',
             'Error:',
+            'Error invoking tool',
+            'Failed ',
         ]
-        return any(pattern in content for pattern in error_patterns)
+        return any(pattern in head for pattern in error_patterns)
 
     def _is_markdown(self, content: str) -> bool:
         md_patterns = ['```', '**', '##', '- **']
diff --git a/EvoScientist/stream/utils.py b/EvoScientist/stream/utils.py
index 4244a3d..e125303 100644
--- a/EvoScientist/stream/utils.py
+++ b/EvoScientist/stream/utils.py
@@ -64,18 +64,27 @@ def has_args(args) -> bool:
 
 
 def is_success(content: str) -> bool:
-    """Determine if tool output indicates successful execution."""
+    """Determine if tool output indicates successful execution.
+
+    Only checks the first few lines of output for error patterns to avoid
+    false positives from code/docs that contain "Error:" as literal text.
+    """
     content = content.strip()
     if content.startswith(SUCCESS_PREFIX):
         return True
     if content.startswith(FAILURE_PREFIX):
         return False
+    # Only check the first 3 lines — real tool errors appear at the top,
+    # not buried deep inside file content or command output.
+    head = "\n".join(content.splitlines()[:3])
     error_patterns = [
         'Traceback (most recent call last)',
         'Exception:',
         'Error:',
+        'Error invoking tool',
+        'Failed ',
     ]
-    return not any(pattern in content for pattern in error_patterns)
+    return not any(pattern in head for pattern in error_patterns)
 
 
 def truncate(content: str, max_length: int, suffix: str = "\n... (truncated)") -> str:
diff --git a/tests/test_stream_utils.py b/tests/test_stream_utils.py
index edc2f0d..f67cb5b 100644
--- a/tests/test_stream_utils.py
+++ b/tests/test_stream_utils.py
@@ -35,6 +35,31 @@ class TestIsSuccess:
     def test_whitespace_stripped(self):
         assert is_success("  [OK] with spaces  ") is True
 
+    def test_error_in_code_content_not_false_positive(self):
+        # read_file returning code with "Error:" deep inside should be success
+        content = '#!/usr/bin/env python3\n"""\nSkill Packager\n"""\n\nprint(f"Error: not found")'
+        assert is_success(content) is True
+
+    def test_error_on_line4_not_false_positive(self):
+        content = "line1\nline2\nline3\nError: buried deep\nline5"
+        assert is_success(content) is True
+
+    def test_error_on_first_line(self):
+        assert is_success("Error: file not found\nsome detail") is False
+
+    def test_error_invoking_tool(self):
+        assert is_success("Error invoking tool 'write_file'") is False
+
+    def test_failed_to_uninstall(self):
+        assert is_success("Failed to uninstall skill: Skill not found: latex-paper-en") is False
+
+    def test_failed_to_install(self):
+        assert is_success("Failed to install skill: git clone failed: ...") is False
+
+    def test_failed_in_code_content_not_false_positive(self):
+        content = '#!/usr/bin/env python3\n"""Helper"""\n\nif x:\n    print("Failed to connect")'
+        assert is_success(content) is True
+
 
 # === format_tool_compact ===