Copilot remembered the bookings. Why couldn't it resume the workflow?

Description and use

The user investigated how GitHub Copilot creates dynamic workflows and found they are not durable, so killing the process mid‑execution discards progress.

Problem

A user wanted to understand how GitHub Copilot creates and executes dynamic workflows, and discovered that the generated workflows are not durable and cannot be resumed after a process kill.

Outcome

The investigation revealed that Copilot’s dynamic workflow generation is a lightweight, non‑durable process: the workflow is registered but not persisted in a durable store, so killing the process mid‑run discards progress.

Workflow

• 1. User prompts Copilot CLI to create a dynamic workflow.
• 2. Copilot calls dynamic_workflows_manage({operation:"guide"}) to get SDK docs path.
• 3. Copilot views workflows.md to learn workflow API.
• 4. Copilot scaffolds an extension.mjs file in extensions/<name>/ folder.
• 5. Copilot writes meta and run code, then reloads the extension into the session.
• 6. User can run the workflow via run_dynamic_workflow; execution is separate from registration.
• 7. If the process is killed during execution, the workflow state is lost because no durable journal is maintained.

Evidence

When you ask Copilot to create a dynamic workflow, there's no magic Markdown compiler turning your English into a state machine.

Source

The model calls dynamic_workflows_manage({ operation: "guide" }) and the runtime hands back the local paths to the bundled SDK documentation.

Source

The host writes the import and joinSession boilerplate; creating and registering the workflow doesn't run it — execution is a separate step via run_dynamic_workflow.

Source

What happens if I kill the process right after the hotel is booked, before the rental car agent finishes?

Source