Feature News & Roadmap

Commander.js-based CLI with 100+ slash commands, argument parsing, session management. The backbone of every Claude Code interaction.

File operations, code execution, web search, task management, agent spawning, MCP integration. Every tool documented with parameters, examples, and security notes.

React-based terminal UI with Yoga layout, ANSI colors, keyboard/mouse handling, and responsive design.

Streaming messages, extended thinking, prompt caching, multi-model support, and cost tracking.

Comprehensive security: bash command validation, path traversal prevention, read-only enforcement, and permission framework.

Full Model Context Protocol: server registry, tool discovery, resource access, OAuth, and enterprise auth.

Persistent MEMORY.md with auto-loading, nested includes, 200-line/25KB limits, project-scoped.

Native vim keybindings: motions, operators, text objects, insert/normal/visual mode transitions.

Tamagotchi-style companion: 18 species, 5 rarity tiers, deterministic PRNG generation, AI-generated personality.

Resume previous conversations, session snapshots, conversation history, and session chooser UI.

Two-way VS Code integration: file ops, terminal control, permission prompts, multi-session support (up to 32).

Speech-to-text input for hands-free coding. Conditionally loaded via VOICE_MODE feature flag.

Cron scheduling (CronCreate/Delete/List) and remote trigger execution for automated workflows.

Execute Claude Code on remote machines via SSH. Extends beyond local development.

Direct API connection mode bypassing standard routing. For enterprise deployments.

Bundled workflow scripts for common development patterns. Triggered via WorkflowTool.

Always-on proactive AI: watches logs, acts autonomously, push notifications, GitHub webhooks.

Multi-agent swarm: team creation, worker spawning, inter-agent messaging, parallel execution.

Background memory consolidation via Opus subagent: Orient, Gather, Consolidate, Prune.

Safety for Anthropic employees in public repos. Strips codenames, internal info, attribution.

AI-powered classifier for auto-mode permission decisions. Determines safe vs unsafe tool calls.

System monitoring for runtime behavior and performance metrics.

Remote 30-minute Opus sessions for complex multi-step planning. Context teleportation.

Full web browser capabilities beyond simple fetch.

Multiple environments per host:dir pairing.

Sandboxed filesystem for agent experimentation.

Advanced conversation compaction via SnipTool.

Mobile push notifications from KAIROS assistant.

PR subscription webhooks for proactive monitoring.

Referenced in source but purpose unclear - possibly a guidance/navigation system for agents.