v0.0.2

Refactor Core, Add Langfuse Tracing, and Strengthen Security

This release introduces a major refactor of the core execution flow, adds full observability with Langfuse, and significantly improves security and reliability across the system.

Key changes:

1. Observability with Langfuse

• Introduced LLMWrapper for seamless Langfuse integration
• Each LLM call and tool execution is now tracked as a generation with input/output
• Single trace per request with step-by-step visibility of execution flow

2. Architecture Refactor

• Removed unused pipeline package (7 files) and simplified execution model
• Extended LLM interface with Tool() method for direct tool handling
• Engine now delegates tool execution to LLM.Tool() instead of a shared executor
• Reduced coupling and improved modularity

3. Security Improvements

• Added timeout (5s) for bash execution using CommandContext
• Enabled process isolation via SysProcAttr (cross-platform support)
• Strengthened command validation (whitelist/blacklist, input normalization)
• Improved path security: blocked traversal (../), absolute paths, and symlink escapes

4. Testing and Hardening

• Replaced incorrect test stubs with full unit test coverage
• Added tests for command injection, traversal, symlink escape, and unicode bypass
• Covered edge cases for both command validation and path resolution

5. UX Improvements

• Output limiting now works by line count instead of rune count for better readability
• Added tests for output formatting and argument handling

6. Integration Updates

• Automatic Langfuse wrapping in main when observability is enabled
• Improved error messages and consistency across components