Configuration
Permission modes
Permission mode controls how much freedom an agent has when editing your code. Set it in Settings → General; it maps directly to Claude Code's permission flags.
The four modes#
| Mode | Behavior |
|---|---|
| Autonomous | Skip permission prompts entirely — the agent edits, runs commands, and proceeds on its own. Recommended for the parallel-agents workflow, and the default. |
| Accept edits | Auto-accept file edits, but still gate other actions. |
| Plan | Plan first and don't modify files — great for scoping a change before committing to it. |
| Strict | Claude Code's default permission behavior. |
Why autonomous is the default#
NeuralCode's whole model is supervising agents by result — you review the diff and run checks after the fact, rather than approving each step. Autonomous mode is what makes running many agents in parallel practical. The safety comes from isolation: each agent works only inside its own worktree, so nothing touches your main checkout until you merge.
Know your default
Out of the box, agents run autonomously and can edit files and run commands without prompting. If you prefer a tighter leash, switch to Plan or Accept edits in Settings → General.
What's affected#
The mode applies to agent runs you start in the chat. NeuralCode's own internal helpers — like the self-review pass — always run read-only and never modify your code regardless of mode.