01 - Start a New Task#

Starting a task in Helmor means creating an isolated workspace, choosing an agent, and sending a focused prompt. This workflow keeps each piece of work reviewable and self-contained.

Step by step#

1. Pick a repository — select it from the sidebar or add a new one.
2. Create a workspace — click New Workspace, choose a base branch (e.g., main), and let Helmor set up an isolated worktree.
3. Choose an agent — select Claude Code, Codex, or another supported provider in the workspace header.
4. Write a prompt — describe what you want done. Be specific about the goal, relevant files, constraints, and how to verify the result.
5. Send — press Enter. Helmor launches the sidecar, connects to the agent provider, and streams output in real time.

Writing effective prompts#

The best tasks are narrow enough to review in one pass:

✅ Good: "Add input validation to the createUser handler — reject emails without an @ sign and return a 400 response with a clear error message."

❌ Too broad: "Improve error handling across the codebase."

Include:

• Goal — what the end state should look like.
• Scope — which files or modules to touch.
• Constraints — style, performance, backwards compatibility.
• Verification — how you'll confirm it works (test command, expected output).

What happens next#

The agent reads project files, reasons about the task, and starts editing. You'll see:

• Reasoning and tool calls in the session timeline.
• File changes appearing in real time.
• Follow-up questions from the agent — reply inline to unblock it.

When the agent finishes, review the diff before committing.

Related#

• Create a Workspace — workspace setup details.
• Review Agent Changes — what to check after a session.