From fb17b926ecbb025cd29fd2dae3b75bfb0c4871e3 Mon Sep 17 00:00:00 2001
From: Sasha Mitchell <sash.t.mitchell@gmail.com>
Date: Tue, 29 Sep 2026 06:00:39 +0700
Subject: [PATCH] Keep the letters at the ends of an executor code block.

strip("python") deletes every p, y, t, h, o, and n from both ends, so return n came back as return and print(1) came back as rint(1).
---
 agentverse/output_parser/output_parser.py |  12 +--
 tests/test_humaneval_executor.py          | 104 ++++++++++++++++++++++
 2 files changed, 111 insertions(+), 5 deletions(-)
 create mode 100644 tests/test_humaneval_executor.py

diff --git a/agentverse/output_parser/output_parser.py b/agentverse/output_parser/output_parser.py
index 980299292..9177d7bc7 100644
--- a/agentverse/output_parser/output_parser.py
+++ b/agentverse/output_parser/output_parser.py
@@ -374,16 +374,18 @@ def parse(self, output: LLMResult) -> Union[AgentAction, AgentFinish]:
                 text,
                 re.DOTALL,
             )[0]
+            code = parsed_result[4].strip().strip("`").strip()
+            if code.startswith("python3"):
+                code = code[len("python3") :]
+            elif code.startswith("python"):
+                code = code[len("python") :]
+            code = code.strip()
             cleaned_output = {
                 "thought": parsed_result[0].strip(),
                 "reasoning": parsed_result[1].strip(),
                 "criticism": parsed_result[2].strip(),
                 "file_path": parsed_result[3].strip().strip("`"),
-                "code": parsed_result[4]
-                .strip()
-                .strip("```")
-                .strip("python")
-                .strip("python3"),
+                "code": code,
                 "command": parsed_result[5].strip().strip("`"),
             }
         except BaseException as e:
diff --git a/tests/test_humaneval_executor.py b/tests/test_humaneval_executor.py
new file mode 100644
index 000000000..8cb94ecf3
--- /dev/null
+++ b/tests/test_humaneval_executor.py
@@ -0,0 +1,104 @@
+import importlib.util
+import sys
+import types
+import unittest
+from pathlib import Path
+
+
+def _load():
+    root = Path(__file__).resolve().parents[1]
+    pkg = types.ModuleType("agentverse")
+    pkg.__path__ = [str(root / "agentverse")]
+    pkg.__package__ = "agentverse"
+    sys.modules["agentverse"] = pkg
+
+    sub = types.ModuleType("agentverse.output_parser")
+    sub.__path__ = [str(root / "agentverse" / "output_parser")]
+    sub.__package__ = "agentverse.output_parser"
+
+    class _Registry:
+        def __init__(self, name=None):
+            self.name = name
+
+        def register(self, name):
+            def deco(cls):
+                return cls
+
+            return deco
+
+    sub.output_parser_registry = _Registry("OutputParserRegistry")
+    sys.modules["agentverse.output_parser"] = sub
+
+    utils = types.ModuleType("agentverse.utils")
+
+    class AgentFinish:
+        def __init__(self, ret, log):
+            self.ret = ret
+            self.log = log
+
+    utils.AgentAction = type("AgentAction", (), {})
+    utils.AgentFinish = AgentFinish
+    utils.AgentCriticism = type("AgentCriticism", (), {})
+    sys.modules["agentverse.utils"] = utils
+
+    llms = types.ModuleType("agentverse.llms")
+
+    class LLMResult:
+        def __init__(self, content):
+            self.content = content
+
+    llms.LLMResult = LLMResult
+    sys.modules["agentverse.llms"] = llms
+
+    logging_mod = types.ModuleType("agentverse.logging")
+    logging_mod.logger = types.SimpleNamespace(
+        error=lambda *args, **kwargs: None,
+        info=lambda *args, **kwargs: None,
+    )
+    sys.modules["agentverse.logging"] = logging_mod
+
+    pydantic_mod = types.ModuleType("pydantic")
+
+    class BaseModel:
+        def __init__(self, **kwargs):
+            self.__dict__.update(kwargs)
+
+    pydantic_mod.BaseModel = BaseModel
+    sys.modules["pydantic"] = pydantic_mod
+
+    path = root / "agentverse" / "output_parser" / "output_parser.py"
+    spec = importlib.util.spec_from_file_location(
+        "agentverse.output_parser.output_parser", path
+    )
+    mod = importlib.util.module_from_spec(spec)
+    sys.modules[spec.name] = mod
+    spec.loader.exec_module(mod)
+    return mod
+
+
+mod = _load()
+
+
+def _parse(code):
+    text = (
+        "Thought: t\n"
+        "Reasoning: r\n"
+        "Criticism: c\n"
+        "File Path: a.py\n"
+        f"Code: {code}\n"
+        "Command: python a.py"
+    )
+    parsed = mod.HumanevalSolverParser().parse(mod.LLMResult(text))
+    return parsed.ret["output"]["code"]
+
+
+class HumanevalExecutorTest(unittest.TestCase):
+    def test_a_code_line_keeps_its_last_letter(self):
+        self.assertEqual(_parse("return n"), "return n")
+
+    def test_a_print_call_keeps_its_first_letter(self):
+        self.assertEqual(_parse("print(1)"), "print(1)")
+
+
+if __name__ == "__main__":
+    unittest.main()