We’ve shipped major improvements to how skills work, enhanced security with agent sandboxing, and refined the mobile experience.
Install Unlimited Skills Without Restrictions
The platform no longer limits how many skills your agent can install. Previously constrained by tier-based slot limits, agents can now add as many capabilities as needed. The Slots panel now displays all installed skills with clear organization - built-in skills appear with badges, and you’ll get friendly warnings if your skill collection grows unusually large.
Why it matters: Your agent can grow its capabilities freely. Add specialized skills for different workflows without worrying about artificial limits. The platform scales with your needs instead of constraining them.
Discover and Install Skills Directly from Chat
Two new built-in skills make it effortless to expand your agent’s capabilities. The find-skills skill searches available skills across registries and GitHub, while install-skill handles installation with AI-driven security analysis. Just ask your agent “find a skill for data visualization” or “install the pdf skill” and it handles the rest.
Why it matters: No more manually hunting through documentation or repositories. Your agent can discover and safely install the exact capabilities it needs mid-conversation. This makes extending functionality as natural as asking a question.
Built-in Skills Stay Current Automatically
All agents now automatically receive newly released built-in skills at the start of each session. When CrossMind adds a new platform skill, your agent gets it without manual intervention.
Why it matters: Your agent stays current with the latest platform capabilities automatically. You benefit from new features immediately without maintenance overhead.
Mobile Copilot Works Smoothly Now
Significantly improved the mobile copilot experience. The autocomplete dropdown now positions correctly on smaller screens, code blocks wrap properly instead of overflowing, and the close button is easier to tap. These changes make working from your phone or tablet genuinely productive.
Why it matters: Remote work happens on all devices. You shouldn’t lose productivity when accessing your agent from mobile - the interface adapts properly so you can work effectively anywhere.
Agent Sandboxing Prevents Cross-Agent Access
Implemented OS-level sandboxing (using Seatbelt on macOS, bubblewrap on Linux) with per-agent permission isolation. Each agent now runs in a restricted environment that prevents it from reading other agents’ files or accessing sensitive system directories like SSH keys, GPG credentials, or AWS configurations.
Why it matters: Security and privacy are fundamental. If you’re running multiple agents - perhaps one for personal work and another for a client project - they’re now properly isolated from each other. An agent can’t accidentally or maliciously access another agent’s data.
Smarter Task Scheduling
Task creation is now schedule-free by default - you create the task container first, then add schedules when ready. Timezone handling is also clearer, with schedule times automatically interpreted in your configured timezone instead of requiring manual UTC conversion.
Why it matters: More flexible workflow. You can define what your agent should do without immediately deciding when it runs. The timezone improvement eliminates a common source of scheduling mistakes.
Streaming Text Shows Thinking Process
The copilot now streams text in real-time with smooth fade-in animations, and displays thinking blocks persistently in messages. You can see your agent’s reasoning process as it happens.
Why it matters: Better transparency into how your agent works. Watching the thinking process helps you understand decisions, catch issues earlier, and build trust in the agent’s reasoning.