Changelog

v0.6.0 (current)

• Permission system v2: fine-grained per-tool permissions and approval flows
• unwrap-free codebase: all product code uses proper error propagation
• TUI improvements: markdown rendering, plan mode, scroll, session management
• Multi-agent enhancements: pipeline, team orchestration, shared memory bus
• Improved transcript compaction with LLM-driven summarisation

v0.5.0

• HTTP API — axum-based REST server with sessions and SSE streaming
• Terminal UI — ratatui-based TUI with streaming tool indicators and plan mode
• Multi-Agent — agent pool, shared memory, messaging bus, pipeline & team orchestration
• Python SDK — pip install recursive-client
• TypeScript SDK — npm install recursive-client
• Loop Mode — recursive loop for self-scheduling autonomous agent runs

v0.2.0

• Skill system v2: refs, scripts, params, injection modes, composition
• MCP HTTP+SSE transport
• MCP resources and prompts support
• Feature flags: mcp, web_fetch, anthropic
• Structured error types
• 5 runnable examples
• 367+ tests

v0.1.0

• Minimal ReAct agent loop
• OpenAI-compatible LLM provider
• Filesystem tools: read, write, list, patch
• Shell tool with sandboxing and timeout
• Mock provider for offline testing
• CLI: run, repl, tools commands
• Hook system for lifecycle observation
• Transcript compaction
• MCP stdio transport
• Skill system v1