Run Workflows
A workflow is a declarative, multi-stage pipeline that runs several agent steps toward one outcome. Wayland ships 178 workflows in total: 107 ForgeFlow workflows grouped by category, plus 71 bundled workflows. This guide covers the ForgeFlow library. This guide covers finding, inspecting, and running them from the desktop.
Open the Workflows library
Section titled “Open the Workflows library”Open Workflows from the sidebar. Workflows are grouped by category (Featured first, then Business Operations, Career, Content Creation, Creative Projects, Cross-Domain, Lifestyle, and Software & Engineering). Use the search box in the filter rail to jump to one by name, or pick a category to narrow the list.
Inspect a workflow
Section titled “Inspect a workflow”Click a workflow card to open its detail view. It shows what the workflow does (“When you launch this”), an excerpt of its steps, and the skills it activates as it runs (“Uses these skills”). Click any skill chip to open that skill in the Skills page. This lets you see up front what the workflow will use before you run it.
Run a workflow
Section titled “Run a workflow”From the detail view:
- Under Run on, pick the backend agent and model the workflow should run on. If the picker says no models are available, connect a provider first in Settings > Models.
- Click Launch now. Wayland opens a new conversation with the workflow loaded as the agent’s first directive and works through its stages, reporting back as it goes. You can interrupt or change direction at any point.
If you had already started this workflow and a recent session is still in flight, the detail view offers to resume that session or start fresh instead of showing Launch now.
To run it later instead of now, click Schedule to create a scheduled task pre-filled with this workflow. See Schedule Recurring Tasks.
Approvals during a run
Section titled “Approvals during a run”A workflow is meant to reach a finished deliverable, so it works through its stages without stopping for a prompt at every tool call. One that paused on each file read would be a checklist, not an automation.
How far it proceeds on its own depends on the backend you picked under Run on:
- Claude Code runs under Wayland’s guarded-auto mode. Routine calls proceed; anything Wayland classifies as destructive is still put in front of you.
- Other backends run in their own full-auto mode for the duration of the workflow.
Two things follow, and both are worth deciding once rather than per run:
- Pick the workspace deliberately. Launch a workflow against the project directory it is meant to touch. The agent works where you point it, so the scope of the run is the scope of the folder you opened.
- Choose the backend to match the blast radius. For a workflow that edits code or moves files in a repository you care about, Claude Code gives you the destructive-call gate for free.
Unattended runs stop rather than guess
Section titled “Unattended runs stop rather than guess”A workflow you schedule runs when nobody is watching, so it is held to a different rule than one you launch yourself.
An attended session waits indefinitely for you — you are the thing it is waiting on. A scheduled run instead carries a deadline, kept strictly shorter than the gap to that conversation’s next run. If a held decision reaches that deadline it resolves as a denial. Never as an approval, never cached, and never persisted, so an expired hold cannot quietly become a standing permission on the next run.
Wayland raises the expiry as its own signal rather than a generic failure, so you can tell the difference between a task that broke and a task that was waiting on you.
Build a workflow
Section titled “Build a workflow”Click Build a workflow at the top of the library to author a new one. The builder has two tabs:
- Write it myself opens an editor where you give the workflow a name, description, category, and tags, then write its body in Markdown.
- Describe it lets you describe the workflow in plain language and have Wayland draft the body for you, which you can then refine in the editor.
When you save, Skill Guard scans the body and reports a verdict (clean, flagged for review, or blocked).
If you would rather author workflows as ForgeFlows in the RON format and validate and run them from the engine, see the references below.