Skip to main content
You don’t have to leave your AI assistant to manage agents. From any AI you’ve connected to Clearskies, Claude or ChatGPT, you can take an agent through its full lifecycle: create it, read its current setup, update it, publish it, and manage it from there, all in conversation. It’s the same set of agents you’d build in the app, managed from wherever you’re already working.

​
What you can do

  • Create and edit agents. Describe what you want and it’s set up as a draft. Edits to a published agent go to a draft, so the live version keeps working untouched.
  • Read and list agents. Pull up an agent to check its model, command, instructions, and web-search setting, or list every agent in your workspace.
  • Validate before publishing. Validate a draft and get back the errors to fix, before anything goes live.
  • Publish and unpublish. Publishing makes the draft live so workflows and reps can use it. Unpublishing takes an agent out of service without deleting it, it’s reversible.
  • Discard a draft. Throw away pending edits and go back to the published version.
  • Duplicate an agent. Copy an existing agent as a new draft to use as a starting point.
  • Delete an agent. Remove one you no longer need.
  • Review draft history. See the changes made to an agent’s draft and who made them.
  • Organize with folders. Create, rename, and delete folders, and move agents into them to keep a growing set manageable.

​
Drafts and publishing work the same way

Managing an agent from Claude or ChatGPT follows the same rules as the app. Creating an agent makes a draft; publishing is what makes it usable. Editing a published agent writes to its draft while the live version keeps working, publish again to roll out the changes, or discard the draft to go back to the published version. A workflow can reference only a published agent, and it can’t publish while it points at one that’s still a draft, so publish the agent before wiring it into a workflow, whether you built it in the app or from your assistant.

​
Grounded in your workspace

Before building, the AI checks what your workspace supports, the available models, the tools an agent can use, and the rules a command has to follow, so what it drafts is something your workspace can actually publish.

​
In practice

Working in Claude with Clearskies connected, you ask it to build an agent that drafts follow-up emails, set the model, and publish it. Then you have it create a workflow that runs the agent after every call, the agent and the workflow both set up without leaving the conversation. Questions? We’re here to help at support@clearskies.cc