Skip to main content

Using the AI Agent — Interactive Mode

Interactive mode is a live, back-and-forth session with the coding agent. The agent runs continuously while your workspace is open, and you can send prompts, answer its questions, and watch it work in real time.

Use interactive mode for:

  • Complex or multi-step tasks where you want to guide the agent
  • Work where the agent may need to ask clarifying questions
  • Exploratory tasks where the outcome depends on what the agent discovers

The workspace interface

When a workspace is open, the interface has three panels:

Left sidebar — file browser showing the workspace repository. Updated automatically as the agent writes files. Also shows git status (modified files, unpublished commits).

Centre — the chat area. Agent responses appear here as markdown. The agent's current status is shown at the top (thinking, executing, idle).

Right — workspace controls: model switcher, git actions, session controls.


Sending a prompt

Type in the input box at the bottom and press Enter. A few things to know:

  • The input box is disabled while the agent is working. It re-enables when the agent finishes and is waiting for your next instruction.
  • Use @filename to include file content. Mentioning @src/config/rbac.py in your prompt automatically includes that file's content, giving the agent context without you needing to paste it.
  • Be specific about scope. Instead of "fix the config", try "update the RBAC roles in src/config/rbac.py to add a Visitor role with read-only access to zones A and B".

When the agent asks a question

The agent may pause and ask you something — a yes/no confirmation, a choice between options, or a free-text question. When this happens:

  • A prompt overlay appears in the chat
  • The input box remains disabled until you respond
  • Click your answer (for choice questions) or type and submit (for free-text)

The agent resumes immediately after you respond.

If you reconnect after a brief disconnect, any unanswered questions are shown again automatically.


Agent status indicators

StatusWhat it means
ThinkingThe agent is processing your prompt or reasoning about the next step
ExecutingThe agent is actively running commands, reading files, or writing code
WaitingThe agent has a question for you
IdleThe agent has finished; ready for your next prompt

The input box is only active when the agent is Idle or Waiting.


Stopping the agent mid-run

Click the Stop button in the workspace header to interrupt the agent immediately. The agent process is terminated; any files it was writing may be in a partial state. Check git diff before continuing.


Checking the agent's work

After the agent finishes a task:

  1. Review the file changes in the left sidebar — modified files are highlighted
  2. Use the Diff view to see exactly what changed
  3. Run a quick sanity check by asking the agent to summarise what it did: "What files did you change and why?"
  4. Commit when you're satisfied (see Git Operations)