Claude Code makes manual mode the default
Claude Code v2.1.200 renamed the default permission mode to Manual, which reviews every action. CI pipelines relying on the old behavior need updating.
Claude Code v2.1.200, released July 3, changed what “default” means. The permission mode previously labeled “default” is now called “Manual” in the CLI, the --help output, the VS Code extension, and the JetBrains extension. The underlying config value used by hooks and SDK integrations remains "default", but the behavior the label describes is now accurate: Manual reviews every proposed action and waits for your approval.
This is a breaking change for any script, CI job, or background agent that started Claude Code without specifying a permission mode and expected it to proceed without interactive prompts. Those workflows will stall at the first approval request after upgrading.
What to change
If you run Claude Code non-interactively:
--permission-mode acceptEditsaccepts file writes automatically but still prompts before running commands. Use this for most automation that touches files but not shell execution.--permission-mode bypassPermissionsskips all prompts. Use this for fully automated runs in trusted, sandboxed environments.
The alias manual also works in place of default as a config value from v2.1.200 onward, for readability.
Hooks and SDK integrations that already set "defaultMode": "default" explicitly in their config are unaffected. Only setups relying on the old unlabeled default need to change.
Why this matters
The rename is accurate rather than just cosmetic. New users who try Claude Code without reading the docs will now see that the starting mode asks before every action, which matches reality. It also reduces the chance that someone accidentally runs an agent session with full autonomy because the previous default sounded neutral.
For existing automated setups, running coding agents remotely covers the full permission configuration for CI, headless, and background agent use.