feat: add builtin tool of llm firewall

[BhAem]

No description provided.

[cuericlee]

llm_firewall-》llm_shield，

[Copilot]

Pull Request Overview

This PR adds a new LLM Firewall plugin for content moderation and security filtering that integrates with Volcengine's LLM Firewall service to provide real-time content moderation for LLM requests.

• Introduces LLMFirewallPlugin class that analyzes user inputs for various risks before allowing requests to reach the language model
• Adds configuration support for the LLM Firewall plugin with AppID configuration
• Updates documentation with usage examples and configuration instructions

Reviewed Changes

Copilot reviewed 3 out of 3 changed files in this pull request and generated 5 comments.

File	Description
veadk/tools/builtin_tools/llm_firewall.py	New plugin implementation with content moderation logic, history building, and risk detection
docs/content/5.tools/1.builtin-tools.md	Documentation update with LLMFirewallPlugin usage example and configuration notes
config.yaml.full	Configuration template addition for LLM Firewall app_id

---

💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.

[Copilot]

The docstring states that ValueError is raised if required environment variables are missing, but the init method doesn't actually raise this exception. The getenv() function will raise ValueError if the environment variable is not found (based on the veadk.utils.misc implementation), but this happens inside getenv, not explicitly in init. The documentation should clarify that the ValueError comes from the getenv() calls for VOLCENGINE_ACCESS_KEY, VOLCENGINE_SECRET_KEY, and TOOL_LLM_FIREWALL_APP_ID.

Suggested change

	ValueError: If required environment variables are missing
	ValueError: If required environment variables are missing. This exception is raised by the
	`getenv` calls for `VOLCENGINE_ACCESS_KEY`, `VOLCENGINE_SECRET_KEY`, and
	`TOOL_LLM_FIREWALL_APP_ID`.

[Copilot]

Potential exception if decision_type is None. The code calls int(decision_type) without checking if decision_type is None first. If decision_type is None (which can happen when getattr returns None), this will raise a TypeError. Consider adding a null check: if decision_type is not None and int(decision_type) == 2 and risk_info:

Suggested change

	if int(decision_type) == 2 and risk_info:
	if decision_type is not None and int(decision_type) == 2 and risk_info:

[Copilot]

Potential IndexError if parts list is empty. While line 152 checks if parts, it assumes parts[0] exists when checking hasattr(parts[0], \"text\"). If parts is a truthy but empty container, this will raise an IndexError. The condition should be if parts and len(parts) > 0 and hasattr(parts[0], \"text\") or simplified to check the length explicitly.

Suggested change

	if parts and hasattr(parts[0], "text"):
	if parts and len(parts) > 0 and hasattr(parts[0], "text"):

[Copilot]

The conversion of set to list at line 260 is unnecessary since the set is only used once for joining. You can directly join the set: reason_text = \", \".join(risk_reasons) if risk_reasons else \"security policy violation\". This simplifies the code and avoids the intermediate variable.

Suggested change

	risk_reasons_list = list(risk_reasons)
	# Generate blocking response
	reason_text = (
	", ".join(risk_reasons_list)
	if risk_reasons_list
	# Generate blocking response
	reason_text = (
	", ".join(risk_reasons)
	if risk_reasons

[Copilot]

The broad Exception catch at line 224 makes it difficult to diagnose and handle different failure scenarios. Consider catching more specific exceptions (e.g., network errors, timeout errors, API errors) and logging them with appropriate severity levels. This would also help distinguish between transient failures and configuration issues.

[cuericlee]

add additional call back

before_agent_callback=governance.before_agent_callback,
after_agent_callback=governance.after_agent_callback,

[cuericlee]

comment as #269 TODO

[cuericlee]

comment as #TODO

[cuericlee]

/lgtm