5 items, live from D1

Checklist

Every real, sourced verification/checklist fact, grouped by category unless filtered.

item	source
Sandboxing reduces permission prompts by 84% in Anthropic's own internal usage.	`Making Claude Code more secure and autonomous with sandboxing`
If Claude tries to access something outside the sandbox boundary, the user is notified immediately and can choose whether to allow it -- the boundary lets Claude act freely inside it while still surfacing any attempted escape.	`Making Claude Code more secure and autonomous with sandboxing`
The sandbox is built on OS-level security primitives -- Linux bubblewrap and macOS Seatbelt -- to enforce filesystem and network restrictions at the OS level, not just in the application layer.	`Making Claude Code more secure and autonomous with sandboxing`
Filesystem isolation is particularly important in preventing a prompt-injected Claude from modifying sensitive system files; network isolation prevents a prompt-injected Claude from leaking sensitive information or downloading malware.	`Making Claude Code more secure and autonomous with sandboxing`
Anthropic's Claude Code sandboxing draws two boundaries -- filesystem isolation (read/write confined to the working directory) and network isolation (an allowlist of domains via a proxy) -- so Claude can act freely inside the boundary without asking permission for every step, and is flagged immediately if it tries to leave it.	`Making Claude Code more secure and autonomous with sandboxing`

subagentchecklists