id · name · enabled
id is stable for webhooks and logs; enabled gates live side-effects while validation still explains disabled flows in dry mode.
Scenarios
Directed flows on the board—not a separate automation product
In the Digio workspace, Scenarios are saved graphs: manual triggers, HTTP steps, queue-to-agent tasks, and HTTPS webhooks. The runner walks only nodes reachable from the chosen trigger, orders them with a topological sort (Kahn), rejects cycles, and supports dry-run logging before any outbound call.
Graph + ports
Dry run default
Abort / pause
Flow · Run test
Same execution order in Validate and in live Run test—dry mode skips fetch, task enqueue, and POST.
Persistence
What is stored per scenario
Each scenario is a JSON object on the team: stable id, display name, enabled flag, optional primaryAgentId, optional runOnAgentSchedule (only when primary resolves to a known agent), arrays of nodes and edges. Unknown node types normalize to manual triggers; malformed edges are dropped.
primaryAgentId
When set and valid, add_task nodes with an empty agentId inherit it—so drafts stay valid after you pick a default specialist.
nodes[] · edges[]
Nodes carry canvas x/y plus type-specific fields; edges reference from/to node ids with optional outKey/inKey port names (defaults o0 / i0).
runOnAgentSchedule
Boolean flag stored with the scenario; it only sticks when primaryAgentId is non-empty and matches the roster—used for agent-timed runs in the app shell.
Node palette
Five node types the runner understands
Types are fixed in code (trigger_manual, http_request, add_task, webhook_notify, integration_logo). The overlay palette instantiates defaults with sane ports; normalization clamps string lengths and numeric priority to 1–10.
trigger_manual
Only entry points: no incoming edges allowed in normalization. You pick which trigger Run test uses when a flow has more than one.
http_request
GET or POST with optional JSON body (up to 16k chars). Live run sets Accept, optional Content-Type + body for POST, captures up to 4k of response text for later webhook payloads.
add_task
Queues a standing task: agentId, title (trimmed, required), priority. Live path calls the workspace hook with launchMode immediate; missing title or agent is logged as a skip, not a throw.
webhook_notify
POST only after URL passes the same HTTPS allow-list helper used for spreadsheet status webhooks. Body includes scenario id/name, node id, and a truncated preview of the last HTTP response.
integration_logo
Social or integration marker on the canvas—runner appends a log line for visibility in dry and live modes without performing I/O.
Wiring
Edges, ports, and invalid links
Edges store from, to, outKey, inKey, and their own id. Self-loops and edges into a manual trigger are removed. If an edge names a port missing on a node, normalization falls back to that node’s first declared port—so flows keep running after palette upgrades.
Directed only
Graph is treated as forward from the active trigger: unreachable nodes never execute—handy for experimental branches left on the canvas.
Parallel fan-out
Multiple outgoing edges from one port are allowed; Kahn’s algorithm yields a deterministic linear order among ready nodes—no hidden random shuffle.
Missing endpoints
Edges whose from/to do not resolve to existing node ids are dropped during normalize—Run test never throws on stale graph fragments.
Execution order
Reachability + topological sort + cycle guard
getScenarioExecutionOrder first DFS-marks nodes reachable from the chosen trigger, then builds indegree counts on that subgraph. A queue drains zero-indegree nodes; if the output length is less than reachable size, the graph returns error cycle—Run test stops before any I/O.
no_start
Returned when the trigger id is missing from the node map—usually corrupt paste or a deleted trigger still selected in UI.
no_trigger
executeScenarioFlow aborts early if the normalized graph contains zero trigger_manual nodes—every flow must declare an explicit start.
cycle
Any directed cycle in the reachable subgraph leaves residual indegree; the runner refuses order and surfaces the failure in the log panel instead of looping forever.
http_request
Outbound HTTP with bounded capture
Live mode uses the browser fetch implementation (or an injected test double). AbortSignal from the overlay is passed through so Cancel stops mid-flight. Responses are read as text, not forced JSON parse—binary-safe truncation to 4k chars for lastHttpPreview.
GET vs POST
Method toggles strictly between GET and POST; POST with a non-empty body sets Content-Type application/json and sends the raw string you authored.
Status line in log
Each live step logs HTTP status, OK/FAIL against response.ok, and byte length—enough to debug APIs without dumping secrets by default.
Dry-run line
Dry mode prints method + URL and explicitly marks simulation—no network tab noise while you teach the flow to others.
add_task
Queue work to a specific agent
The runner validates agentId and title after trim. Dry run echoes the tuple and marks simulation. Live run checks agentId against the current dashboard roster; unknown ids log agent missing and skip enqueue—protecting you from stale flows after a hire leaves.
Priority 1–10
Normalized at graph load: non-numeric becomes 5; values clamp to the inclusive 1–10 band used across standing tasks.
Immediate launch
onAddTask receives launchMode immediate—the same urgency path as pressing Run on a card, not a silent backlog draft.
Title length cap
Titles persist up to 500 chars at normalization; extremely long template strings survive import without breaking JSON storage.
webhook_notify
HTTPS callback with structured JSON
Live POST JSON includes event scenario_step, scenarioId, scenarioName, nodeId, and lastHttpPreview (first 1200 chars of the most recent HTTP response text). That lets your receiver correlate notify steps with upstream API payloads without re-fetching Digio internals.
URL policy
Invalid or non-HTTPS endpoints log skip bad webhook—same helper as spreadsheet status webhooks, so security rules stay consistent across products.
Errors stay in-log
Network failures append human-readable error strings; AbortError propagates as a halted run with ok false aborted, not a silent drop.
Dry-run preview
Dry mode prints truncated URL plus simulation marker—ideal when demoing a customer endpoint without firing their on-call.
Safety
Dry run is the default in the overlay
Run test boots with dryRun true so opening someone else’s scenario never fires webhooks accidentally. Validation still walks order: disabled scenarios print a disabled note in dry logs so reviewers see intent without side effects.
Simulated only
No fetch call is made; method and URL appear in the log with the dry-run copy string from product strings.
No enqueue
add_task branches log the would-be title and agent id, proving field mapping without touching the kanban or To do queue.
No POST
Webhook nodes log URL length-safe preview plus simulation—receiver teams can validate routing rules before go-live.
Still traced
integration_logo nodes always append a label line in both modes so documentation subgraphs show up in exported run logs.
Live iteration
Pause, resume, and cancel semantics
Between nodes the runner awaits scenarioWaitResume: if the overlay sets paused, the loop sleeps in 90ms slices until cleared or aborted. AbortError unwinds as ok false aborted; other errors on HTTP/webhook append to the log but continue when the code path allows—matching the forgiving behavior operators expect in demos.
Per-node await
Every step checks signal and pause gate before work—tight loop for integration_logo, meaningful for long HTTP calls.
Fetch injection
fetchImpl is overridable for unit tests or future server-side replay—production UI passes window.fetch.
Trigger pick
When triggerNodeId is empty or stale, the runner chooses the first normalized manual trigger—predictable for single-entry flows, explicit selector for multi-branch.
Overlay UX
Team scope, agent filter, integrations flyout
ScenariosOverlay receives scenarios[], agents[], teams[], activeTeamId, canEdit, and optional prefilterAgentId. The sidebar can filter flows whose primaryAgentId matches a selected specialist; the header scope control can emit select-team when you pick another dashboard—keeping lists consistent with the board you are viewing.
Local editing buffer
Changes stay in component state until you save—Run test always reads the latest normalized graph including unsaved canvas tweaks when the parent wires it that way.
Integrations entry
autoOpenIntegrationsFlyout can expand the credentials menu once—useful when launching Scenarios from an Assets or integration shortcut without hunting the dock.
Help string
Status info pulls scenarios lead, hotkeys line, and footer copy from uiStrings—so keyboard power users see the same guidance as mouse operators.
Where scenarios plug in
Same agents, tables, and API surface
Task steps use real agents from your roster; web rules match Tables. Heavier scheduled automation also lives in To do, Backlog, and Agent API—Scenarios are great for guided checks and manual test hops you design on the canvas.
Read next
Related product pages
Automation layers, templates with standing tasks, and integrations for inbound triggers.
Model cross-system checks, then flip dry run off once
Build the graph with manual triggers, validate order and skips, then run live with pause/rescue patterns your team already uses on the board.