Dynamic workflows in Copilot CLI and the Copilot app
By Steven Van ·
Copilot CLI, the Copilot app and the SDK can now run multi-agent processes defined in code, with limits on agents, time and AI credits.
Dynamic workflows are now available in Copilot CLI, the GitHub Copilot app and the GitHub Copilot SDK. A workflow is a program, kept inside a Copilot extension, that defines how a task is carried out. GitHub describes it in a docs page. The steps, conditions and handoffs are written in code, and agents handle the parts that need analysis or judgment. Steps can run in sequence, in parallel, or both.
The docs separate this from two existing features. Autopilot lets Copilot keep working without pausing for input. /fleet has Copilot delegate to subagents, deciding each time how to divide the work. In a dynamic workflow, the author defines the process, so reusing a workflow means reusing its steps and rules. Agents can still give different answers on different runs.
Workflows can do the following:
- Run commands, use tools or call other services.
- Split a goal into independent tasks and run them in parallel.
- Pass structured results from one stage to the next.
- Have subagents verify each other's findings.
- Ask for user input, if the client supports it.
- Pause at a checkpoint for review, then resume.
Runs can be limited by concurrent agents, total agents launched, active running time and an approximate AI credit cap. The credit limit is not a hard ceiling, because usage is reported after it occurs. Limits can be set in the prompt, in the workflow code or in personal settings, and the docs advise testing on a small scope before a large run. Runs can be watched, paused, cancelled and resumed through /workflows in the CLI or the Workflows button in the Copilot app.
Workflows can be written by hand or generated by asking Copilot, which includes built-in authoring guidance. By default a workflow is scoped to the current session. It can be copied to a personal extensions directory, or to a repository's extensions directory to share with a project.
