{"version":"2.1.307","anchor":"sandbox-guidance-text-for-windows-style-sandbox-account","canonical_anchor":"sandbox-guidance-text-for-windows-style-sandbox-account","heading":"Model gets new instructions for a sandbox that runs under its own account","tier":"notice","area":"Sandbox","scope":"individual","heads_up":false,"url":"https:\/\/changelogs.core-directive.com\/v\/2.1.307\/e\/sandbox-guidance-text-for-windows-style-sandbox-account","release_url":"https:\/\/changelogs.core-directive.com\/v\/2.1.307","markdown":"### Model gets new instructions for a sandbox that runs under its own account\n\nClaude Code now tells the model how blocked connections appear and where to put temp files when the sandbox runs as a separate account\n\n**What**\n\nA sandbox is the restricted space Claude Code runs commands in. Claude Code has new instructions for the model about a sandbox that runs commands under its own separate user account. They say:\n\n- Results from this sandbox never include a `<sandbox_violations>` block, the section that normally lists what the sandbox blocked.\n\n- A blocked network connection shows up only in the command's own error output.\n\n- Files in the sandbox account's own folder cannot be opened by you or by Claude Code's tools, so temporary files that need opening belong in the scratchpad directory.\n\n**Why**\n\nThe model gets a more accurate picture of how this sandbox behaves. It should spot a blocked connection from the command's error and keep files it needs to share where you can open them.\n\n- Area: Sandbox\n- Tier: You'll notice\n- Useful: 2\/5\n- Signal: 2\/5\n- Scope: individual\n- Heads-up: no"}