Dawn agents can now help teams create and update reusable Skills and Prompt Commands while keeping changes reviewable and protected from stale edits.
Dawn Team
Product Engineering
Dawn agents now get better as you use them. Ask an agent to turn the current conversation into a Skill or a Prompt Command (a sharper PR review checklist, a cleaner release-note format, a standardized support reply), and the improvement becomes reusable workspace content right from the thread.
The best agent instructions usually emerge from real use, not from a quiet afternoon in the Skills editor. When a team improves how an agent works in conversation, that lesson should outlive the thread it came from.
Skills tell agents how to work.
They can define a review checklist, a support tone, a planning format, an incident response process, a writing style, or even when to reach for a different tool, like the built-in Skill that tells the agent when a JavaScript script is the right way to handle a request. Until now, creating a good Skill meant switching into the Skills screen, writing the markdown manually, and translating the lesson from the conversation into durable instructions.
Now the agent can help with that step. You can ask Dawn to turn the guidance from the current conversation into a Skill, refine an existing Skill, or update the wording so the agent follows the same process next time.
Prompt Commands are for work that starts with the same framing again and again.
A good PR review prompt, ticket planning prompt, test-plan prompt, release-note prompt, or customer-summary prompt usually includes more than one sentence. It carries criteria, formatting rules, and reminders about what context to inspect. That repeated framing belongs in a command.
Dawn can now help create or update those commands. If a team keeps asking for the same task in chat, the agent can turn the pattern into a slash command that accepts a short brief at runtime.
For example, after refining a useful review process, you can ask:
Turn this into a workspace Prompt Command called /api-review.Or after improving a planning format:
Update our ticket planning command so it asks for risks, dependencies, and test notes.That makes the best version of the work easier to reuse.
When an agent creates or updates a Skill or Prompt Command, the change is tied to an auditable action. Teams can see what changed and why. The agent helps draft and apply reusable instructions; it does not silently rewrite workspace content.
Dawn also protects against stale edits. If the content changed after the agent read it, Dawn does not blindly overwrite the newer version. That matters in active workspaces where a teammate may be editing a command while another person is discussing changes with an agent.
The result is a safer authoring flow: agents can help, but they do not erase newer work by accident.
The best agent instructions often emerge from use.
Someone asks for a review. The first answer is close but misses a class of risk. The user corrects it. The agent tries again and gets closer. At the end of that thread, the team has learned something about how the agent should work next time.
Before this change, that learning was easy to lose. Now it can become a Skill or Prompt Command while the context is still fresh.
That helps teams improve Dawn through normal use. They do not need to treat prompt design as a separate project. When a conversation produces a better pattern, the agent can help preserve it.
Open a Dawn conversation and ask the agent to draft a Skill or Prompt Command from the current work. Be specific about whether the instruction should become a reusable Skill or a Prompt Command, and whether you want a new one or an update to existing workspace content.
Useful starting prompts include:
Create a workspace Skill from this review checklist.Draft a Prompt Command for this release-note workflow.Update the existing support reply Skill with the tone rules we just agreed on.The existing guides are here:
Dawn has always been shaped by Skills and Prompt Commands. Now agents can help maintain those reusable instructions directly from the conversations where the best improvements are discovered.