05 - Send Your First Agent Task#
Once you have a workspace, send a prompt and watch the agent work. This page walks through your first session end to end.
Pick an agent#
Before sending, choose your agent in the workspace header:
| Agent | Best for |
|---|---|
| Claude Code | Multi-file edits, debugging, reasoning-heavy tasks |
| Codex | Implementation, refactoring, code generation |
Make sure the agent CLI is installed, on your PATH, and authenticated with its provider.
Write a good first prompt#
A strong first task is narrow, verifiable, and self-contained:
"Add a
--verboseflag to thebuildcommand that prints each step as it runs. Update the existing test to cover the new flag."
Weak first tasks ("refactor the whole codebase", "make it better") produce sprawling diffs that are hard to review.
Send the task#
- Type your prompt in the composer at the bottom of the workspace.
- Press Enter (or click Send).
- Helmor starts a session, launches the sidecar, and streams agent output in real time.
You'll see:
- Agent reasoning and tool calls in the session timeline.
- File edits appearing as the agent writes them.
- Follow-up questions if the agent needs clarification — reply inline.
When the session finishes#
The agent signals completion. You now have three options:
- Review the diff — check changed files against the original task.
- Continue — send a follow-up prompt to refine the result.
- Discard — reset the workspace to its pre-session state.
Troubleshooting first sessions#
| Problem | Fix |
|---|---|
| Session hangs immediately | Verify agent CLI works outside Helmor (claude --version). |
| "Sidecar not found" error | Run bun run dev:prepare if developing locally. |
| Agent produces no output | Check provider credentials and API key validity. |
Next: Learn how to Review Agent Changes before committing.