v0.1.0

[0.1.0] - 2026-02-07

Added

• Native Multi-LLM Adapters: Native support for openai, claude, and
  gemini API formats.
• Multimodal Support: Full support for image inputs on compatible models
  (e.g., GPT-4o, Claude 3.5 Sonnet).
• Micro-kernel Architecture: A completely redesigned Agent Core with
  Gateway, Registry, Policy, and Executor.
• Improved Bot Channels: Added image/photo support for both Telegram and
  Feishu.
• Headless Mode: Efficient execution without heavy SDK dependencies.
• Skill Loading Strategies: Support for eager (preload all) and lazy
  (on-demand via tools) skill loading.
• New Built-in Tools:
  • list_skills: To see available knowledge extensions.
  • read_skill: To explicitly read skill instructions when needed.
• Improved Policy Engine: Granular control over shell commands and tool
  permissions.
• Session Isolation: Support for independent user sessions and working
  directories.

Changed

• Refactored src/ai to move away from @anthropic-ai/claude-agent-sdk.
• Unified configuration schema for multiple LLM providers.
• Improved logging with structured output.
• Optimized tool execution flow (Proposal -> Policy check -> Execution).

Removed

• Removed src/ai/claude-sdk.ts and src/ai/claude-cli.ts.
• Removed dependency on heavy external Agent SDKs.