Agents Gain Instant Terminal State Awareness

VS Code's updates let AI agents detect terminal pauses precisely, even in hidden background terminals. For a PowerShell script prompting for a name via Read-Host, the agent identifies the input need, relays your response (e.g., "James Monte Magno"), injects it automatically, and reads the output ("Hello, James Monte Magno"). A bottom-left indicator shows the hidden terminal, which you can expose to inspect executed scripts. This prevents agents from hanging on inputs, enabling completion without manual exposure.

In foreground terminals, agents prompt for the same input but recognize when you intervene, updating their context from your manual entry. Result: agents stay synchronized regardless of terminal visibility, reducing workflow friction.

Effortless User Takeover for Sensitive Commands

Agents default to handling inputs but yield control seamlessly. Click the hidden terminal indicator to enter data yourself—the agent detects your intervention, incorporates the update (e.g., your name entry), and proceeds. This "focus terminal" option suits hands-on or sensitive commands, blending automation with manual precision.

Trade-off: Full foreground support means agents interact with your active session, but one-click takeover ensures you retain authority without disrupting agent flow.

Multi-Prompt Handling Scales to Complex Init Flows

For commands like npm init, agents parse multiple sequential prompts—up to 9 questions (package name, version, description, entry point, etc.)—collecting answers in batch before injecting them all at once. Enter responses like "Tiny Tool Town" for name and "tiniest tool town ever" for version; the agent feeds them into the terminal in real-time, visible upon exposure.

Outcome: Agents manage interactive CLIs that would otherwise require tedious step-by-step babysitting, accelerating setup for projects while allowing overrides.