Workflows and skills

Turn a job you repeat into an agent of your own — deterministic, cheaper to run, and schedulable.

What a workflow is

A workflow is an agent you created.

You describe a job once. Monokrew turns it into a fixed sequence of steps — a graph of what happens, in what order, using which capabilities. From then on, running it follows that graph rather than working the job out afresh.

That difference matters more than it sounds.

Why build one instead of just asking

You can ask the agent to do anything in Agent Chat, and for a one-off that is the right tool. A workflow earns its place when the job comes back.

• It costs less to run. The steps are already decided, so the agent is not re-planning the job every time. A workflow spends fewer tokens than the same request typed fresh.
• It is the same every time. A fixed graph does the same thing on Tuesday that it did on Monday. Asking in plain language leaves room to be interpreted differently.
• It keeps its place when it pauses. If a step waits for your approval, the workflow holds everything it has done so far. Answer an hour later, or the next morning, and it carries on with the context intact rather than starting over.
• It can run without you. Schedule it, or let a connected app trigger it.

Tip — A good sign that something should be a workflow: you have typed nearly the same prompt three times.

Building one

Describe it

Write what you want in plain English, the same way you would ask in chat.

Optionally, let Monokrew rewrite your description into structured steps first. This is worth doing when the job has several parts — it turns a paragraph into a numbered sequence you can check before anything is built.

Let it build the graph

Monokrew generates the workflow graph and matches every step against what your workspace can actually do — bundled skills, skills you have installed, MCP servers, and connected toolkits.

Each step in the graph shows where it stands:

Warning — A workflow with orange steps cannot run. It is saved as a draft until every step is resolved, which is why a workflow you just built may not appear in My Workflows.

Resolve what is missing

For each orange step, connect what it needs: install a skill, add an MCP server, or authorise a toolkit. Then re-check the workflow and the step turns green.

If nothing covers a step, that is worth knowing before the job matters — better to find out here than halfway through a scheduled run.

Add the human parts

Two things you can place into the graph before saving:

• Human in the loop — a point where the workflow pauses and waits for you. Use it before anything that publishes, spends, or sends on your behalf.
• Attachment placeholders — a slot for a file you will provide when the workflow runs. This is what lets one workflow handle a different spreadsheet each week rather than being fixed to one.

You can also add special instructions for particular steps — details the agent should follow that were not obvious from your original description.

Save it

Once every step is green, save it. It moves out of drafts and into My Workflows, ready to run or schedule.

The library

Your workflows are organised across five views.

My Workflows

Your finished workflows. From here you can run one, give it a schedule, view its graph, open its details, or move it to archived.

Scheduled

Everything with a schedule. You can run one immediately without waiting for its next slot, pause it, or change when it runs.

Tip — Pausing is better than deleting when you are unsure. A paused schedule keeps the workflow and its settings; you can start it again without rebuilding anything.

Drafts

Where a workflow waits when it has steps nothing covers yet. Open one, resolve the orange steps, and move it to My Workflows.

If a workflow you built is not where you expected it, look here first.

Archived

Workflows you have set aside. Archiving does not delete anything — you can recheck an archived workflow's skills later and move it back to My Workflows if your workspace has gained what it was missing.

Running one

Run a workflow directly, on a schedule, or from an event in a connected app — when something happens in a toolkit you have authorised, it can start the workflow for you.

You can also run one from a channel. Ask for it by name in Slack, WhatsApp or anywhere else you have connected, and the result comes back in the same place. See Connecting a channel.

Skills

Skills are what extend the agent beyond what it can do on its own.

Monokrew ships with a set of skills and MCP servers already available. Beyond those, you can install skills from ClawHub, and provide API keys for any that need one.

A skill you install becomes available to every workflow and to Agent Chat — which is what turns an orange step green.

See Secrets, MCP servers and toolkits for where keys and connections are managed, and Installing skills for the ClawHub library.

What it costs

Running a workflow spends billing tokens, and a longer graph spends more than a short one. Because the steps are already planned, the same job costs less as a workflow than as a fresh request.

Building and saving a workflow is not the expensive part — running it is. See How billing tokens work.

Next steps

A workflow is for a job you repeat. For a whole function run for you — testing, search, social, customers — see the agent library.

To let a workflow reach you where you already work, continue to Connecting a channel.