What's New
Synthetic Teams Core
Foundation for OpenCrew-style synthetic developer teams, enabling coordinated multi-agent workflows.
- Autonomy Ladder (L0-L3) with ToolHook-based enforcement
- L0=observe, L1=reversible, L2=recoverable, L3=irreversible
- 27+ tools classified by default, customizable per-agent
- ApprovalRequester interface for escalation flow
- Agent ACL/Permissions: AllowlistChecker interface enforced in both handoff and spawn paths
- Team Config: 3-layer hierarchy (Intent/Execution/Maintenance) with A2A circuit breaker settings
- Thread-Bound Subagents: ReplyToChannel/ReplyToThread for scoped agent communication
- Autonomy wiring: Agent config-driven autonomy level enforcement in agent loop
MCP Client Enhancements
- Tool discovery and session lifecycle management
- Health checks, auto-reconnect, tool aggregation across servers
- Claude CLI and Codex CLI provider MCP tool passthrough
- Provider MCP capability interface
Test Infrastructure
- Mock MCP stdio server for integration tests
- E2E integration tests for multi-agent system
- 22 new tests for synthetic teams features
Full Changelog
https://github.com/AetherClawTech/aetherclaw/blob/v0.4.0/CHANGELOG.md