iCode Agent · Guide · Workflow

Fix a failing test.

The core loop of a coding agent: reproduce the failure, understand it, change the code, and prove the change. iCode Agent runs that loop against your real repository.

The task

Start iCode Agent in the project and describe the goal:

Fix the failing tests in this repository.

Naming a specific test narrows the work, but iCode Agent can find the failure itself by running your test command.

What happens

1. Explore. It reads the relevant files and searches the tree to locate the failing code.
2. Reproduce. It runs your test command and reads the actual failure.
3. Repair. It makes a targeted edit — not a rewrite — and shows the diff.
4. Verify. It runs the tests again and reports the result. If they still fail, it keeps going.

Every step is visible: each tool call and a live preview of its output stream past as the agent works.

Steering it

If the agent heads the wrong way, you do not wait for it to finish:

• Type a message and press Enter to redirect it mid-task.
• Press Esc to abort, or Alt+Enter to queue a follow-up.
• Press Ctrl+O to expand a tool result you want to read in full.

Good prompts for this workflow

Fix the failing tests and explain what the bug was.
The test in tests/auth.test.ts is failing — find out why before changing anything.
Make the failing test pass without changing the test itself.