iCode Agent · First task

Fix a failing test.

The fastest way to see how iCode Agent works is a small bug. You give it the goal; it does the investigation, the change and the verification.

1. Start iCode Agent in your project

cd ~/my-project
icode-agent

2. Ask for the fix

Type this into the prompt:

Fix the failing tests in this repository.

3. Watch it work

You see every step as it happens — the tool calls and a live preview of their output:

▸ read src/math.ts
✓ read src/math.ts
▸ bash bun test math.test.ts
✓ bash bun test math.test.ts      ← observes the failure
▸ edit src/math.ts                ← repairs the code
✓ edit src/math.ts
▸ bash bun test math.test.ts
✓ bash bun test math.test.ts      ← verifies: all tests pass

Fixed the add() function ... All tests pass.

A failure is not an error to the agent — it is information. It keeps repairing and re-verifying until the task is done or it needs you.

4. Stay in control

• Steer — type a message and press Enter while it works; it changes direction.
• Queue — Alt+Enter queues a follow-up for after the current task.
• Interrupt — Esc aborts; queued messages return to the editor.
• Expand — Ctrl+O shows full tool output instead of a preview.

5. Go further

Once a task is done, your work is saved. Exit and return with --continue, or open /session to manage sessions.