Skip to content

Guides

How it actually works

Ask Driver to explain one implementation flow end to end. Include the codebase, accessible branch, and an example input, such as the request that starts invoice generation.

  1. Ask in Driver: β€œOn this branch, how does this invoice request move through rendering, storage, and delivery? Cite the entry point and each handoff, including plan and configuration branches.”
  2. Ask Driver to identify the owning subsystems and trace their calls. With prepared Advanced context, ask it to consult the Atlas and relationships before following the source. Open the citations for each handoff.
  3. Request a table with step, component, input/output, source, and failure behavior. Ask Driver to distinguish a direct call from an asynchronous queue or an external service whose implementation is outside the available code.
  4. Supply the relevant design document or point to one accessible through a configured documentation connection. Ask Driver to list differences between the documented flow and the implementation, then review them with the area owner.
  5. Copy the reviewed flow, codebase branch, and any known revision into the architecture issue or team documentation. Keep the unresolved external steps visible for the next reader.

If you work in an MCP-connected coding assistant, ask the assistant to use get_atlas_index, get_atlas_detail, and get_related for this flow when Advanced context is ready. Basic source search and file reads can trace the implementation directly. Use the resulting flow to help new engineers get oriented and revisit it when the implementation changes.