Terminal AI

Terminal AI adds three AI-powered hooks to the embedded terminal. Each fires inline — no separate panel, no context switch.

Question hook

Type a question or an unfamiliar command. Terminal AI routes it to a fast, cheap model and prints a suggestion inline, prefixed Agentissimo:. Useful for quick lookups without interrupting your flow.

Error assist

When a command exits non-zero, Terminal AI reads the error output and prints a fix hint inline. Fires automatically on failures above a configurable exit-code threshold (default: 1).

Destructive guard

Commands matching catastrophic patterns — rm -rf /, drop table, and similar — are intercepted before they run. Two modes:

• confirm — blocks the command and asks y/N before proceeding.
• warn — runs the command but prints a recovery note after.

Set mode: off to disable.

Configuration

All three hooks are on by default. Configure per project in agentissimo.yaml under terminalAi:

terminalAi:
  enabled: true
  model: claude-haiku-4-5
  question: { enabled: true }
  errorAssist: { enabled: true, minExitCode: 1 }
  destructiveGuard: { enabled: true, mode: confirm }