Agent intelligence
Reasoning loops with progressive trust, shadow verification, and parity to modern IDE agent UX.
Tool execution sits behind policy guards, secret redaction on logs, and shell / Unicode safety hardening.
Multi-agent paths include voting, handoffs, and background schedulers without losing a single-writer mental model.
Context stacks support compaction, micro-compaction, scratchpad bridges, and away-summary recovery.
Editor parity spans tab prediction, speculative edits, LSP integration, and client / server verify-after-edit.