{"version":"2.1.300","anchor":"sandbox-and-prefetch-wording-generalised","canonical_anchor":"sandbox-and-prefetch-wording-generalised","heading":"Sandbox rule text now covers all commands, not just Bash","tier":"notice","area":"Sandbox","scope":"both","heads_up":false,"url":"https:\/\/changelogs.core-directive.com\/v\/2.1.300\/e\/sandbox-and-prefetch-wording-generalised","release_url":"https:\/\/changelogs.core-directive.com\/v\/2.1.300","markdown":"### Sandbox rule text now covers all commands, not just Bash\n\nThe strict sandbox text now says every command the model runs must be sandboxed, and the prefetch memory message no longer names a 64 MiB limit\n\n**Unclear.** It is not clear whether the sandbox now actually covers more commands or only the wording changed.\n\n**What**\n\nA sandbox is a fenced-off area where commands run with limited access to your machine and network. The text for the strict sandbox setting no longer mentions Bash. It now reads: \"All commands invoked by the model must run in the sandbox unless they are explicitly listed in excludedCommands.\"\n\nClaude Code can load files ahead of time in case they are needed, which is called prefetching. When it stops doing this to save memory, its message no longer says 64 MiB. It now says the unused prefetched files \"would exceed their limit in memory\".\n\n**Why**\n\nThe sandbox wording now reads as applying to every command the model runs, and the prefetch message suggests the memory limit may no longer be fixed at 64 MiB. Commands listed in `excludedCommands` remain the exception.\n\n- Area: Sandbox\n- Names: `excludedCommands`\n- Tier: You'll notice\n- Useful: 2\/5\n- Signal: 2\/5\n- Scope: both\n- Heads-up: no"}