v0.4.3 — Hooks System, LLM Tool, and Stability Fixes

Introduces a comprehensive hooks system for automating workflows on app and agent events, a new call_llm tool for secondary AI calls, and important stability fixes.

🪝 Hooks System

• New event-driven hooks architecture for running commands and injecting prompts on app/agent events
• Event bus with support for session lifecycle, permission changes, label updates, and scheduler ticks
• Command hooks execute shell commands with permission-based security and sandboxing
• Prompt hooks inject context into agent conversations when events fire
• Cron-based scheduling via the new SchedulerService
• Event rate limiter and SIGKILL fallback for runaway hook processes
• Session metadata exposed as environment variables for hook scripts
• Source auto-activation when referenced in prompt hooks
• Event logger for debugging hook execution

🧠 LLM Tool (call_llm)

• New session-scoped tool enabling secondary Claude calls for specialized subtasks
• Supports summarization, classification, extraction, and analysis workflows
• Attachment-based file and image loading with line range support
• Predefined output formats plus custom JSON Schema support
• Extended thinking mode for complex reasoning tasks
• Parallel execution support for multiple simultaneous calls

🔌 Codex Improvements

• Simplified Codex connection setup by removing third-party provider options
• Bumped Codex version to craft-v0.4.1

🐛 Bug Fixes

• Fixed Cmd+Enter sending message twice when Enter is configured as the send key (Fixes #212, #197)
• Fixed settings navigation — clicking "App" no longer fails after selecting another page
• Fixed update_user_preferences replacing notes instead of appending
• Fixed prompt hooks not firing and source auto-activation

🏗️ Build & Infrastructure

• Use bun-linux-x64-baseline in CI for non-AVX2 CPU compatibility (Fixes #218)
• Refactored hooks: removed legacy API and global state, added comprehensive test coverage
• Improved config watcher and source config validation