Skills
A Skill is a named, reusable automation you define once and run on demand, on a schedule, or automatically when a Workspace event occurs. Instead of manually sending a follow-up email, updating a task, or drafting a monthly report — you describe the workflow once and dpaperwork executes it.
What a Skill can do
Each Skill is a sequence of Steps. Available step types:
| Step | What it does |
|---|---|
| Assign | Assigns a task to a team member or to Automation |
| Create Record | Creates a new Record in any Domain (e.g. a Project, a Contact, a Deal) |
| Notify | Sends an Inbox notification to a specific team member |
| Generate Document | Drafts a client-facing Document using context from the current Record |
| Wait | Pauses the Skill for a set duration before continuing |
| Condition | Branches based on a field value or Record state — different paths for different cases |
Steps can be combined in any order. A typical Skill might: check a condition → generate a document → assign a task → notify a team member.
Triggers
Skills run in three ways:
On demand — run it manually from the Skills page or via Chat:
"Run the monthly retainer report Skill for all active clients"
Event trigger — fires automatically when something happens in your Workspace:
- A deal moves to Closed Won
- A deal moves to Closed Lost
- A Cycle closes
- A Project is created
Scheduled trigger — fires on a recurring schedule you define (e.g. every Monday at 9am, first of the month).
Authority Levels
Every Skill declares an Authority Level — the permission matrix that defines what it's allowed to do. You set this when creating the Skill:
- Domain access — which Domains it can read from, and which it can write to
- Integration access — whether it's allowed to use Gmail (e.g. to send a document)
- Action scope — what specific actions it may take without human approval (e.g. "draft but not send", "assign tasks but not close Deals")
Automation operates strictly within these rails. If a step requires something outside the Authority Level, the Skill escalates to you rather than proceeding.
Skills are also scoped to the Domains their creator can access — a Skill can only read from or write to Domains the person who set it up is allowed into. This keeps automation inside the same access boundaries as the team.
Approvals
Some steps are marked as requiring human-in-the-loop approval — for example, sending a client-facing email or closing a Deal. When a Skill run reaches an approval-gated step, it pauses and posts an approval card for you: as an interrupt-strip item right above the greeting on Today, and in the Inbox archive.
The approval card shows exactly what the step is about to do. From there you can:
- Approve — the Skill continues from that step
- Reject — the Skill stops and the run is marked as rejected, with the reason recorded in the run history
Approval requests are distinct from escalations: an escalation means Automation is stuck and needs input to proceed; an approval means Automation knows exactly what it wants to do and is asking for a go-ahead before doing it.
Creating a Skill
Via Chat:
"Create a Skill that runs at deal close — generates a project kick-off document, assigns a setup task to me, and notifies the team"
dpaperwork drafts the Skill structure for you to review, adjust, and save.
From the Skills page: click New Skill, give it a name, add steps one by one, configure the trigger, and set the Authority Level.
Run history
Every time a Skill runs, the result is logged: start time, steps completed, steps skipped, and any escalations triggered. Access the run history from the Skill detail page. If a run stalled, the history shows exactly which step failed and why.
Escalations
When Automation is working through a Skill and hits something it can't resolve — a missing field, an ambiguous condition, an action outside its Authority Level — it escalates straight to you, the Founder.
Escalations include the Skill name, the step that failed, and what's needed to unblock it. Once you resolve it, the Skill can be re-run from where it left off.
Try it in a walkthrough
Walkthroughs follow one job end to end, in real Chat exchanges: