Skip to main content

Stages

Each stage defines what the agent does during that phase of work.

Stage Properties

PropertyDescription
NameShort identifier (e.g., “Triage”, “RCA”, “Fix”). Must be unique within a project.
InstructionsDetailed prompt for the agent — what to do, what to look for, what outputs to produce. Supports the / slash command to insert playbook prompts.
TypeEntry (starting point), Work (intermediate), or Terminal (end point).
Auto-approveIf enabled, transitions out of this stage happen without human approval.
Approver configWho can approve: Player (agent suggests based on code ownership) or Individuals (specific team members).
Archive configHow completed Players are archived: Never, Manual, or Automatic (after N days).
Ask a humanOptional guidance on when the agent should stop and escalate to a person.

Stage Types

  • Entry: Work can start here. A workflow can have multiple entry stages.
  • Work: Intermediate stages where processing happens. Requires incoming and outgoing transitions.
  • Terminal: Final stage. No outgoing transitions. Work concludes here.

Using Stages to Trigger Integration Actions

If your project has connectors enabled (e.g., Jira, Linear, Slack), the AI agent can use those integrations as part of any stage. Write stage instructions that tell the agent when and how to use them. Examples:
  • Create a Jira ticket after triage: “After completing your analysis, create a Jira ticket in the PROJ project summarizing the issue, severity, and recommended fix. Include code references.”
  • Post to Slack: “Notify the #engineering channel in Slack with a summary of your findings.”
  • Update a Linear issue: “Update the linked Linear issue with your root cause analysis and change the status to In Progress.”
The agent automatically has access to all connected integrations — you don’t need to configure anything beyond writing clear instructions in the stage’s instruction field. The agent decides how to use the tools based on your instructions and the conversation context.
Integration actions require the relevant connector to be set up at the project or organization level. See Connectors for setup guides.

Transitions

Transitions define the allowed paths between stages.

Creating a Transition

In the workflow builder, draw a connection from one stage to another. A dialog will prompt you to describe the transition conditions — the rules that tell the agent when to take this path.

Transition Conditions

Each transition includes an instruction field. Write specific conditions, not vague ones:
  • Good: “Promote to Fix when root cause has been clearly identified, the issue requires code changes, and failing scenarios demonstrate the bug.”
  • Avoid: “Move to the next stage when ready.”
The agent uses these conditions to decide which transition to request.

Approvals

Auto-approve vs. Manual

  • Auto-approve on: The agent transitions immediately when it determines the stage work is complete.
  • Auto-approve off (default): The agent requests the transition and waits for a human to approve.
When a human reviews a transition request, they are approving the work completed in the current stage — confirming the agent’s output meets expectations before it moves on.

Approver Configuration

  • Player: The agent suggests an approver based on context (e.g., who last modified the affected code). Any project member can approve.
  • Individuals: Specific team members are designated as approvers for that stage. Only they can approve.

Archival

Archive Modes

ModeBehavior
NeverPlayer is never auto-archived from this stage
ManualUsers can archive Players manually
AutomaticPlayers are archived after a configurable number of days (1, 7, 14, or 30)
Archived Players move to a special terminal stage where the AI generates a summary of the conversation and work performed. Archived Players can be restored back to their previous stage at any time.

The Workflow Builder

The visual editor for designing workflows. Access it from Settings > Stages in your project.

Canvas

  • Drag and drop stage nodes to arrange the layout
  • Draw connections between stages to create transitions
  • Click a stage to view or edit its properties in a side panel
  • Click an edge to view or edit the transition rule
Lists all stages grouped by type (Entry, Work, Terminal) with buttons to add new stages.

Validation

The builder displays warning indicators on stages that are missing:
  • Incoming connections (no way to reach this stage)
  • Outgoing transitions (non-terminal stage with no next steps)

Toolbar

  • Zoom in/out and fit view
  • Reorganize — auto-arrange stages in a hierarchical layout
  • Delete all — reset the workflow

Deleting a Stage

When you delete a stage that has active Players, you’ll be prompted to choose a destination stage. All Players currently in the deleted stage are moved to the destination, and any pending transition requests targeting the deleted stage are automatically redirected.

Export & Import

Export your workflow as JSON (stages, transitions, agent rules, layout positions). Import into other projects to replicate workflows across your organization. When importing into a project that already has stages, the import dialog detects conflicting stage names and lets you resolve each one individually:
  • Merge — Overwrites the existing stage with the imported definition
  • Create Copy — Imports the stage as a new stage with an auto-generated unique name (e.g., “Triage (1)”)
When conflicts are found, all are initially set to Merge. You can change individual stages to Copy, or use the bulk action buttons to set all conflicts to the same resolution at once. Stages that don’t conflict with existing names are imported automatically without prompting.

The Workflow Inbox

The inbox is the operational view for managing active Players across your workflow. For full documentation, see Workflow Inbox.