Fix/general run

[suluyana]

No description provided.

[gemini-code-assist]

Code Review

This pull request introduces several significant enhancements to the ms-agent framework, including a new RepetitionGuardCallback to detect stuck loops, an isolated git-based snapshot and rollback utility for agent output directories, a TaskManager for managing background tasks, and a LocalSearchTool for on-demand codebase search via sirchmunk. Additionally, Tavily Search and Extract APIs have been integrated, and FileSystemTool has been upgraded with grep, glob, and safer edit_file capabilities. The code review identified several critical issues: a potential AttributeError in the repetition guard when retrieving configuration, a possible crash in the snapshot utility if git is missing, potential option-parsing errors in ripgrep when search patterns start with a hyphen, a missing path validation in edit_file, and incomplete dictionary handling in _save_transcript.

Important

The consumer version of Gemini Code Assist on GitHub is being sunset. Starting June 18, 2026, new organization installations will be blocked, and all code review activity will officially cease on July 17, 2026.
For more details on the timeline and next steps, please review the Help Documentation.

[gemini-code-assist]

Using getattr() on a plain dictionary (which guard_cfg becomes if it is None) will raise an AttributeError because dictionaries do not support attribute-based lookup for keys. Since both dict and DictConfig support the .get() method, you should use .get() instead of getattr() to safely retrieve the configuration values.

Suggested change

	guard_cfg = getattr(config, "repetition_guard", None)
	if guard_cfg is None:
	guard_cfg = {}
	self.threshold: int = int(getattr(guard_cfg, "threshold", _DEFAULT_THRESHOLD))
	self.lookback: int = int(getattr(guard_cfg, "lookback_rounds", _DEFAULT_LOOKBACK))
	self.max_warnings: int = int(getattr(guard_cfg, "max_warnings", _DEFAULT_MAX_WARNINGS))
	guard_cfg = getattr(config, "repetition_guard", None)
	if guard_cfg is None:
	guard_cfg = {}
	self.threshold: int = int(guard_cfg.get("threshold", _DEFAULT_THRESHOLD))
	self.lookback: int = int(guard_cfg.get("lookback_rounds", _DEFAULT_LOOKBACK))
	self.max_warnings: int = int(guard_cfg.get("max_warnings", _DEFAULT_MAX_WARNINGS))

[gemini-code-assist]

If git is not installed on the system, _git will raise a FileNotFoundError (which inherits from OSError). Since restore_snapshot only catches subprocess.CalledProcessError, a missing git executable will cause a crash. You should catch Exception or OSError to make it robust against missing dependencies.

Suggested change

	except subprocess.CalledProcessError as e:
	logger.warning(f'[snapshot] restore failed: {e.stderr.strip()}')
	return False, 0
	except (subprocess.CalledProcessError, OSError) as e:
	logger.warning(f'[snapshot] restore failed: {e}')
	return False, 0

[gemini-code-assist]

If pattern starts with a hyphen (e.g., -foo), rg (ripgrep) will interpret it as a command-line option rather than the search pattern, which can cause errors or unexpected behavior. To prevent this, use the -e option to explicitly specify the pattern.

Suggested change

	if output_mode == 'files_with_matches':
	args.extend(['-l', pattern, str(file_path)])
	elif output_mode == 'count':
	args.extend(['-c', pattern, str(file_path)])
	else:
	args.extend(['-n', pattern, str(file_path)])
	if output_mode == 'files_with_matches':
	args.extend(['-l', '-e', pattern, str(file_path)])
	elif output_mode == 'count':
	args.extend(['-c', '-e', pattern, str(file_path)])
	else:
	args.extend(['-n', '-e', pattern, str(file_path)])

[gemini-code-assist]

Similar to _grep_rg_file, if pattern starts with a hyphen, rg will treat it as a command-line option. Use the -e option to safely pass the pattern to rg.

Suggested change

	if output_mode == 'files_with_matches':
	args.extend(['-l', pattern, str(root)])
	elif output_mode == 'count':
	args.extend(['--count-matches', pattern, str(root)])
	else:
	args.extend(['-n', pattern, str(root)])
	if output_mode == 'files_with_matches':
	args.extend(['-l', '-e', pattern, str(root)])
	elif output_mode == 'count':
	args.extend(['--count-matches', '-e', pattern, str(root)])
	else:
	args.extend(['-n', '-e', pattern, str(root)])

[gemini-code-assist]

If path is not provided (it defaults to None), calling self.get_real_path(path) on line 1010 will raise a TypeError (which is caught by the except block but results in a confusing error message). It is better to validate that path is not None or empty at the beginning of the method.

Suggested change

	if old_string is None:
	return 'Error: `old_string` is required.'
	if new_string is None:
	return 'Error: `new_string` is required.'
	if not path:
	return 'Error: `path` is required.'
	if old_string is None:
	return 'Error: `old_string` is required.'
	if new_string is None:
	return 'Error: `new_string` is required.'

[gemini-code-assist]

The _save_transcript method only writes messages that have a to_dict attribute. If any message in the list is already a plain dictionary, it will be silently ignored. You should handle both Message objects (with to_dict) and plain dictionaries, similar to how SubAgentStreamWriter does.

Suggested change

	for msg in messages:
	if hasattr(msg, 'to_dict'):
	f.write(
	json.dumps(msg.to_dict(), ensure_ascii=False)
	+ '\n')
	for msg in messages:
	if hasattr(msg, 'to_dict'):
	f.write(
	json.dumps(msg.to_dict(), ensure_ascii=False)
	+ '\n')
	elif isinstance(msg, dict):
	f.write(
	json.dumps(msg, ensure_ascii=False)
	+ '\n')

[alcholiclg]

这个 PR 里有比较多的脚本在头部新增了 from __future__ import annotations，主要解决的问题是？如果暂时没有类型注解延迟求值需求的话是不是可以先不加这些。

[suluyana]

主要解决兼容py3.9的问题

[alcholiclg]

这个模块目前的状态是完整的吗？看起来互相调用的一批文件还没有提交。

[alcholiclg]

这里做顺序调整是主要是为了做什么处理吗

[alcholiclg]

增加重试次数主要是为了处理什么 case？

[suluyana]

解决 API 调用因瞬时错误（rate limit、网络超时、502/503）导致任务失败的问题。与这个同时添加的还有retry的delay从1s改到3s