feature

Anthropic added OS-level sandboxing to Claude Code, cutting permission prompts by 84% while blocking prompt injection attacks. The sandbox runtime is open source, and Claude Code now runs securely in the cloud.

Anthropic shipped Tool Search Tool, Programmatic Tool Calling, and Tool Use Examples — three features that cut token overhead by 85%, eliminate inference passes on intermediate results, and teach correct API usage through examples instead of schemas.

Anthropic cracked the long-running agent problem with a two-phase approach: an initializer that scaffolds the environment, and a coding agent that works incrementally across context windows using git, feature checklists, and mandatory browser testing.

Anthropic's research team spent a year building evals for Claude Code and multi-agent systems. Here's what actually works: start with 20-50 real failures, combine deterministic and model-based graders, and read the transcripts. Teams that invest early ship faster.