Band Band

Recipes & Workflows

This page collects practical workflows for common development scenarios with Band. Each recipe describes a concrete pattern you can adopt and adapt to your projects.

Feature Branch Workflow

Use Band's workspace model to isolate feature work on dedicated branches with their own git worktrees.

  1. Open the Band dashboard and select your project.
  2. Click New Workspace and enter your feature branch name (e.g., feat/user-profiles). Band creates a git worktree for the branch.
  3. Submit a task with your feature requirements: "Implement user profile page with avatar upload, bio field, and settings form."
  4. Review the agent's output in real time. Once the task completes, open the workspace in your IDE to inspect the changes.
  5. Queue follow-up tasks for refinements: "Add validation to the bio field — max 500 characters — and display a character counter."
  6. When the feature is ready, create a pull request directly from the dashboard or your terminal.

Parallel Agent Execution Across Workspaces

Band allows one task per workspace, but you can run tasks across multiple workspaces simultaneously. This is useful for working on independent features in parallel.

  1. Create multiple workspaces for your project, each on a different feature branch.
  2. Submit tasks to each workspace. Band spawns a separate agent instance for each one.
  3. Monitor all running tasks from the dashboard. Each workspace shows its agent's status and output independently.
  4. As tasks complete, review the results in each workspace and submit follow-up prompts as needed.

This pattern works well when you have a backlog of independent issues. Create a workspace per issue, submit the task, and let the agents work in parallel.

Automated PR Review Workflow

Use cronjobs to automatically review incoming pull requests on a schedule.

  1. Create a project-scoped cronjob with the name "PR Review".
  2. Set the prompt to: "List all open pull requests. For each one, review the diff and leave a comment summarizing potential issues, missing tests, and style concerns."
  3. Set the cron expression to 0 9 * * 1-5 (every weekday at 9 AM).
  4. Enable the job. Every weekday morning, Band submits the review task to the default branch workspace.

Backlog Processing with Multiple Agents

Process a queue of issues from your issue tracker by assigning each to a workspace.

  1. From your issue tracker, identify a batch of issues suitable for agent work (bug fixes, refactors, small features).
  2. For each issue, create a workspace with a branch named after the issue (e.g., fix/issue-142).
  3. Submit a task to each workspace referencing the issue: "Fix issue #142: the login form does not validate email format. Add regex validation and display an error message."
  4. Band runs agents in parallel across all workspaces. Monitor progress from the dashboard.
  5. As agents complete, review the changes and create pull requests for the ones that look good.

CI Integration Patterns

Band exposes a web server that you can call from CI pipelines to trigger tasks programmatically.

  1. Ensure the Band server is accessible from your CI environment (running on a known host and port).
  2. In your CI pipeline, use the Band CLI or HTTP API to submit a task: 
    band task submit --workspace "myproject/main" --prompt "Run the full test suite and fix any failures"
  3. Poll the task status or use webhooks to wait for completion.
  4. Based on the task result, continue or fail the CI pipeline.

This pattern is useful for automated fix-and-retry workflows: if CI fails, trigger an agent to diagnose and fix the issue, then re-run the pipeline.

Using Band with Monorepos

Band works naturally with monorepos because projects in Band map to git repositories, and workspaces map to branches. For monorepos with multiple packages:

  1. Add your monorepo as a single project in Band.
  2. Create workspaces for each feature or fix, just as you would in a single-package repo.
  3. In your task prompts, specify which package the agent should focus on: "In the packages/api directory, add rate limiting to the /users endpoint."
  4. Use your project's CLAUDE.md file to document monorepo conventions — workspace structure, shared dependencies, build commands — so the agent follows them automatically.

For very large monorepos, consider creating separate Band projects for independent sub-repositories if they have distinct git histories, or use the additional directories feature to give the agent access to shared configuration.