NEW Hidden Features You MUST Enable In Your Claude Code Setup!

Key Concepts

• Autodream (/dream): A memory consolidation feature that prunes stale information to improve context accuracy over time.
• Flicker Mode: An experimental renderer that stabilizes the terminal UI and adds mouse support.
• MCP (Model Context Protocol): A framework enabling Claude Code to interact with GUI applications and external tools.
• Hooks: Programmable triggers that automate logic at specific points in the agent’s lifecycle (e.g., session start, pre-tool use).
• Git Worktrees: A feature allowing multiple parallel sessions within the same repository.
• Effort Level: A setting to control the model's reasoning time and depth.

1. Memory Management and Performance

• Autodream: Addresses context memory degradation. By invoking consolidate my memory or using the dream command, the agent performs a "cleanup" process, pruning irrelevant notes and retaining useful insights. This ensures the AI’s memory remains accurate across long-term sessions.
• Flicker Mode: Solves terminal UI glitchiness. By setting the environment variable CLAUDE_CODE_NO_FLICKER=1, users enable an experimental renderer that replaces limited ANSI rendering with a stable viewport, enabling mouse support and smoother performance.

2. Workflow and Productivity Tools

• Powerup Command (/powerup): Provides interactive, in-terminal lessons to help users discover and master hidden features within Claude Code.
• Codeex Plugin: An external integration that allows users to hand off tasks to a secondary agent. This is particularly useful for bypassing rate limits on Pro plans and obtaining a "second opinion" on high-stakes code refactors or migrations.
• Session Management:
  • Teleport (/teleport): Enables moving active sessions between devices (e.g., mobile to desktop).
  • Remote Control (/remote-control): Allows users to control a local terminal session from a phone, tablet, or browser, keeping the execution environment local while providing a remote interface.
  • Branching (/branch): Allows users to fork sessions to experiment with new ideas without impacting the primary workflow.

3. Advanced Automation and Customization

• Output Styles: Users can modify the agent’s behavior via /output-style.
  • Default: Optimized for software engineering.
  • Explanatory: Provides educational insights.
  • Learning: Guides the user to write code themselves rather than just providing diffs.
• Hooks: These provide deterministic control over the agent. Examples include:
  • Session Start: Automatically loading specific context.
  • Pre-tool Use: Logging every bash command executed.
  • Permission Request: Routing sensitive actions to a messaging app (e.g., WhatsApp) for manual approval.
• Dynamic Shell Embedding: Allows embedding shell outputs directly into prompts. The model receives the result of the command rather than the command itself, keeping the context clean and secure.

4. Computer Use and Integration

• Computer Use via MCP: Currently in research preview for Pro/Max plans on macOS. It allows Claude Code to open apps, interact with GUI elements (Swift UI, Electron, etc.), identify bugs, and verify fixes.
• Git Worktrees: Deep integration allows running multiple parallel sessions in the same repository. The cloud worktree command or the get worktree create hook allows for custom logic in managing parallel development branches.

5. Synthesis and Conclusion

Claude Code has evolved into a highly sophisticated tool, though many of its most powerful features remain hidden behind configuration files and experimental flags. By leveraging Autodream for memory, Hooks for automation, and Computer Use for GUI interaction, users can significantly increase their productivity. The transition from a basic CLI tool to a highly customizable, multi-device, and context-aware agent environment is facilitated by these undocumented or experimental features, which allow for granular control over reasoning depth, session management, and external tool integration.