Skip to content
Wayland Docs

Workflows (ForgeFlows)

A ForgeFlow is a declarative, multi-stage agent pipeline. You describe a fan-out, pipeline, or verify topology once, and the engine runs it as a coordinated set of sub-agents. The alternative is asking a single model to orchestrate the stages turn by turn; a ForgeFlow makes the structure explicit instead.

The Workflows library grouped into categories

How a ForgeFlow runs

Section titled “How a ForgeFlow runs”

A ForgeFlow is authored in RON as one Workflow(...) document with three parts: meta (name and description), an optional schemas table, and an ordered list of phases. Each phase holds steps, and there are three step kinds:

  • Agent: one sub-agent call.
  • Pipeline: an ordered chain of sub-agent calls, optionally fanning over an array so each item streams through the stages.
  • Parallel: sibling sub-agent calls that run at the same time and fold their outputs together.

When it runs, the ForgeFlow lowers to the engine’s execution-graph form and runs through a dedicated executor over the same sub-agent dispatch path the Delegate tool uses. There is no separate runtime: the stages are sub-agents. That means a ForgeFlow reuses the engine’s existing tool, provider, and permission machinery rather than introducing a parallel system.

Three ways it surfaces

Section titled “Three ways it surfaces”

The same ForgeFlow concept appears in three places:

  • The Workflow tool: the model runs an inline ForgeFlow mid-conversation.
  • The wayland-core workflow CLI (alias forgeflows): validate, list, and run saved .ron ForgeFlows from the command line.
  • Shadow-mode detection: a telemetry-only signal that flags turns that look like a ForgeFlow. It is off by default and never prompts or reroutes anything.

The in-app library

Section titled “The in-app library”

The desktop Workflows page ships a library of vendored workflows grouped by category, from software and engineering to business operations and content creation. A detail view shows a workflow’s body and the skills it depends on, so you can see what it activates before you run it. You can run one from the library or build your own. See Run and Author Workflows.

Next

Section titled “Next”