Components
Permission Gate
Human-in-the-loop approval interface for AI agent actions with risk assessment
Demo
Click the terminal below and press space to trigger permission requests. Approve with y, deny with n, or always-allow with a.
Features
- Action type badges for file edits, file creation, shell commands, API calls, and deletions
- Risk level indicators from low to critical with color-coded severity
- Three approval modes: allow once, deny, or always-allow for a given action type
- Decision history showing past approvals and denials
- Realistic demo scenarios including destructive operations like force push and .env modification
Keybindings
spacetrigger next permission requestyallow actionndeny actionaalways allow this action typetabtoggle decision historyrreset all decisions