Describe what this PR does / why we need it

Add security and risk control capabilities for Agent applications:

• Sensitive Word Filtering: DFA-based filtering with customizable dictionaries
• Data Masking: Auto-detect and mask PII (phone, ID card, email, bank card, IP, password)
• Security Callback (Java): Optional automated security checks during Agent execution

Includes bug fixes:

• Fixed IP address detection in Chinese text context (Python)
• Fixed security callback blocking mechanism (Java)

Does this pull request fix one issue?

• Yes
• Fixes 安全能力贡献 #30

Describe how you did it

Core Implementation:

• SensitiveWordFilterTool: DFA algorithm, 4 replace strategies (asterisk/delete/custom/detect_only)
• DataMaskingTool: 6 PII types with flexible masking strategies
• SecurityCallback: Implements existing Callback interface for automated checks

Design:

• Implements BaseTool interface, consistent with project architecture
• New security package under core/tools, minimal coupling
• No breaking changes to existing code

Files Changed (13 total):

• Documentation: README.md, README_CN.md, Security-Guide.md, Security-Guide_CN.md
• Java: 3 tools + 1 test class
• Python: 2 tools + 1 example
• Config: 2 pom.xml files (added dependencies)

Describe how to verify it

Run Tests:

Java:

cd ali-agentic-adk-java/ali-agentic-adk-extension/ali-agentic-example
mvn test -Dtest=SecurityAgentTest

Python:

cd ali-agentic-adk-python
python examples/security_demo/security_example.py

Test Results:

• Java: 5/5 tests passed
• Linter: 0 errors

Special notes for reviews

Dependencies Added:

• Java: ali-langengine-core, Spring Framework, Apache POI, Commons-IO, Groovy (test)
• Python: None (standard library only)

Production Notes:

• Default sensitive word dictionary is basic, use comprehensive wordlist in production
• DFA algorithm optimized for large dictionaries (10K+ words)
• SecurityCallback is optional, no impact on existing features

Future Enhancements:

• Load dictionaries from external files
• Tiered sensitivity levels
• Fine-grained masking strategies