What the Agent Can Do

Agent is the built-in agent that makes changes. It is described in the product as a “General-purpose AI agent for software engineering tasks”. It reads and edits your code, runs commands such as builds and tests, and opens pull requests when you ask. It works in a cloud environment with a copy of your repository, on its own branch. Agent is selected by default. To pick it, select the agent picker below the prompt box and choose Agent under Built-in.

Use cases

• Fix a bug and add a regression test.
• Build a small feature that follows your existing patterns.
• Refactor code across several files.
• Update dependencies and fix what breaks.
• Write or update tests and documentation.
• Carry out a plan you made with Plan.

What the agent can do

The agent can only reach what is in its cloud environment: the repositories you picked and the tools installed there. To add tools, services, or setup steps, use an environment. In a thread, the agent opens a pull request only when you ask. Learn more on the Pull requests page.

How the agent works

1. Prepare: Bold Create starts a cloud environment, copies your repository, and creates the agent’s working branch from the branch you picked.
2. Read your instructions: the agent reads repository instruction files if they exist, such as AGENTS.md, CLAUDE.md, .github/copilot-instructions.md, .github/instructions/*.instructions.md, and .boldcreate/boldcreate-instructions.md.
3. Explore: it reads and searches the code it needs.
4. Plan: for longer tasks, it writes a to-do list.
5. Change and check: it edits files and runs commands, such as your tests, and fixes what fails.
6. Report: it ends with a summary of what it did. You review the changes in the Git panel. Follow-up messages in the same thread continue on the same working branch, with the full context of the thread. Put your project’s build, test, and style rules in AGENTS.md at the root of the repository. The agent reads it at the start of every run, so you don’t have to repeat those rules in prompts.

Follow the agent’s progress

You see the agent’s work in real time while it runs.

• Response text: the agent’s messages, with Markdown formatting and code blocks. Select Copy on a code block to copy it.
• Reasoning and steps: a collapsible block shows what the agent is thinking and the steps it took. Steps are grouped into summaries such as Read 3 files, Edited 2 files, Ran 5 commands, and 3 searches. Expand a group to see each step, for example the file it edited and the lines changed, or the command it ran.
• Sidebar status: the thread shows Running in the sidebar until the run ends.

Use the Todos list

For longer tasks, the agent writes a to-do list and updates it as it works. The list appears in the Todos dock above the composer. The badge shows how many items are not done yet.

• Select Todos to expand or collapse the list.
• Items show as pending, in progress, or completed. Completed items are struck through. The to-do list helps you see where the agent is and what is left. You don’t edit it. To change the plan, send a follow-up message.

Give the agent a task

1. Start a new thread, or open an existing one.

2. Select the agent picker and choose Agent.

3. Describe the task. Say what to change, where, how to check it, and what not to touch.

   The /orders endpoint returns every order before paging. Change it to page
   in the database query using the existing limit and offset parameters.
   Keep the response shape the same, and run the order tests when done.

4. Press Enter.

5. When the run ends, open the Git panel with Open workspace panel and review the changes.

6. Send follow-ups for corrections, or ask the agent to open a pull request.

Stop or retry a run

• Stop: select Stop response while the agent is working. Changes it already made stay on its working branch.
• Retry: hover over the last response and select Retry to send your last message again.

Best practices

• Start with Ask or Plan for unfamiliar code or larger changes, then switch to Agent in the same thread.
• Say how to verify: name the test command or the result you expect.
• Keep tasks small: one change per thread is faster to run and easier to review.
• Use an environment when your project needs services, extra tools, or several repositories.
• Review before you merge: check the diff in the Git panel and the pull request in your Git provider.

Troubleshooting

• The agent can’t run your tests: a tool or service is missing in the cloud environment. Add it with setup commands in an environment.
• The run ends with Error: open the thread to read the message. See Run stopped or shows Error.
• This organization has N active runs (limit M). Wait for a run to finish and retry.: your organization reached its limit of runs at the same time. Wait for a run to finish, then send again.