Skip to content

First-Agent-Run.md

Latest commit

 

History

History
108 lines (70 loc) · 3.53 KB

First-Agent-Run.md

File metadata and controls

108 lines (70 loc) · 3.53 KB
Raw

First Agent Run

This guide walks through the first successful Coder Studio agent workflow: open a project, start a recommended first-run provider, watch the session output, and review the resulting Git diff.

What This Guide Solves

Use this to verify that Coder Studio can support a real AI coding loop, not just launch the UI.

Prerequisites

  • Coder Studio installed: npm install -g @spencer-kit/coder-studio
  • Node.js 24 or newer: node --version
  • At least one Provider CLI installed. For the first trial, Claude Code or Codex is recommended:
    • Claude Code: npm install -g @anthropic-ai/claude-code
    • Codex: npm install -g @openai/codex
  • The Provider CLI works in a normal terminal:
    • claude --version
    • codex --version

1. Launch Coder Studio

coder-studio open

Your browser should open automatically. If it does not, run:

coder-studio status

Then manually open the local URL printed by the command.

2. Open A Local Project

On the welcome screen, choose Open Workspace and select a small repository you can safely test with.

For the first run, prefer:

  • a personal repository with a README
  • a Git repository with a clean working tree
  • a project that does not contain production secrets

3. Start A Recommended First-Run Provider

Inside the workspace, create a new session and choose a detected provider. For the first trial, Claude or Codex is the recommended path because the setup and troubleshooting docs are the most detailed.

If the provider is missing:

  1. Install the CLI in your normal terminal.
  2. Verify it with claude --version or codex --version.
  3. Refresh Coder Studio or reopen the workspace.

4. Use A Safe First Task

Do not start with a large refactor. Use a small documentation task first:

Please read the README and add a short "Who this is for" section. Keep the change to 1-2 paragraphs and do not modify source code.

This is a good first test because:

  • the change is small
  • the Git diff is easy to inspect
  • the result is easy to revert

5. Review Output And Changes

While the session runs, watch the agent output. When it finishes:

  1. Open the Git view.
  2. Inspect changed files.
  3. Open the diff.
  4. Decide whether the result matches your request.

You can ask the same session for a follow-up change, edit the files yourself, or discard the change with Git.

6. Check Progress From Mobile

For phone or tablet access, read Mobile and Remote Access.

Mobile is best for:

  • checking whether an agent is still running
  • reading terminal output
  • browsing files and diffs
  • monitoring long tasks

Use desktop for the first full trial. Mobile is a continuation and review surface, not the main heavy editing environment.

FAQ

Can I try Coder Studio without Claude or Codex installed? Yes. You can open workspaces, browse files, and use terminals, but you cannot start a provider's agent session until its CLI is installed.

I installed a provider but Coder Studio cannot find it. What now? Run which <provider-command> and <provider-command> --version in a normal terminal. If the command is missing, the install directory is probably not in PATH. See Agent Providers.

What if the agent makes a bad change? Review the Git diff first. You can edit the file, ask for a follow-up, or discard the change with Git.

Next Steps