Skip to main content
You can already manage agents from Claude or ChatGPT. The same applies to workflows: any AI you’ve connected to Clearskies can take a workflow through its full lifecycle, create it, refine it, test it against real data, and publish it, from the same conversation where you’re already working.

​
What you can do

  • Create and edit workflows. Describe the workflow you want and the AI builds it, trigger, steps, and the connections between them. Edits to a published workflow go to a draft, so the live version keeps running untouched.
  • Validate before publishing. Validate a draft and get back per-step errors to fix, before anything goes live.
  • Test with a dry run. Run a draft against real trigger data and inspect what each step would do. Test runs never produce side effects: no Slack message or email is sent, and nothing is written to your CRM.
  • Publish and unpublish. Publishing makes the draft live and installs its trigger or schedule. Unpublishing stops the automation without deleting it, it’s reversible.
  • Patch a single step. Change one step, a message, a filter condition, a schedule, without resending the whole workflow.
  • Duplicate a workflow. Copy an existing workflow as a new draft to use as a starting point.
  • Inspect runs. List a workflow’s runs or pull up a specific run to see what happened.
  • Organize with folders. Create folders and move workflows into them to keep a growing set of workflows manageable.

​
How drafts and publishing work

Creating a workflow always creates a draft, nothing runs until you publish. Editing a published workflow writes to its draft while the live configuration keeps running; publish again to roll out the changes, or discard the draft to revert. This is the same draft-to-publish model workflows follow in the app, so a workflow started in Claude can be finished in the app, and vice versa.

​
Grounded in your workspace

Before building, the AI checks what your workspace supports, the available triggers, step types, and the variables each trigger exposes, so what it drafts matches what your workspace can actually run. One rule to know: a workflow that runs an agent can only be published once that agent is published. If you try, the publish fails with a per-step error telling you to publish the agent first. If you’re building both at once, publish the agent first, see Manage agents from Claude or ChatGPT. Questions? We’re here to help at support@clearskies.cc