Workflow and logic-function execution
Workflows turn a trigger into a versioned graph of actions that runs in a workspace. The system validates the graph before publication, moves eligible runs through durable queue jobs, and gives people and AI tools one contract for creating, running, and inspecting automation. The web client also gives operators a view of run health, recent failures, and historical runs. See Automation and AI capabilities and AI applications.
Workflow model
The workflow model joins a reusable workflow identity with versioned content and the runs created from that content. Core workflow records are the main identity for dispatch and execution. Workspace version and run records keep links to those core records for workspace reads and history. The core model is now the normal path, while workspace links remain for upgrade and read compatibility.
Workflow identity and versions
A workflow has one trigger and a set of connected action steps. A draft is the editable version. An active core version is the published definition used for new automated runs. Older versions remain available as history, and a run keeps the version and flow snapshot that started it.
Activation checks the trigger, steps, action settings, and graph before it publishes a version. It changes the old active version to archived, updates the data used by automated dispatch, and updates the command-menu entry used for manual starts. Legacy workspace identifiers remain only for upgrade compatibility. New execution paths use the core workflow identity.
References: packages/…/workflow.entity.ts• L42-106 packages/…/workflow-version.entity.ts• L24-105 packages/…/core-workflow-lifecycle.workspace-service.ts• L130-309 packages/…/workflow-run.workspace-entity.ts• L17-72 packages/…/assert-version-can-be-activated.util.ts• L18-91
Workflow content contracts
Each action has input settings, an output shape, and error behavior. The output shape is the contract that lets later steps use values from earlier steps. The same contract supports the editor, server validation, runtime variable resolution, and AI workflow tools.
The action set includes record changes and searches, messages, email, calendar events, HTTP requests, code, stored logic functions, MCP tools, agents, classification, filters, branches, loops, forms, delays, and event waits. Feature flags can disable an action type for a workspace. Form record choices and other data values are checked against the permissions of the run that will use them.
References: packages/…/workflow-version-step-operations.workspace-service.ts• L91-271 packages/…/workflow-action.type.ts• L28-178 packages/…/workflow-schema.workspace-service.ts• L70-183
Workflow graph
Connections define execution order. A connection can be a normal next step, a branch path, or an iterator loop. Node positions affect only the editor view. Saved content assembles execution links without layout positions, while position updates add or clear values for the editor.
The graph builder creates the needed branch and loop entry nodes so the saved graph and the runtime follow the same paths. Removing a step repairs links and keeps required branch paths usable. The editor also checks loop paths when it marks the last step in an iterator, so a visual loop remains consistent with its saved links.
References: packages/…/assemble-workflow-version-content.util.ts• L6-37 packages/…/apply-workflow-step-positions.util.ts• L8-39 packages/…/remove-step.ts• L164-305 packages/…/isLastStepOfLoop.ts• L8-51
Workflow access
Workflow access controls who can view, change, or start a workflow. Workspace-visible workflows can be seen by workspace members, while private workflows are visible only to their creator. The same visibility rule applies when a caller asks for a version, so a private workflow does not leak through its version records.
Manual starts require workflow access and a valid workspace member or application actor. Application-owned workflows are managed by their application and are not edited as workspace workflows. Application callers can start only workflows in their reachable application scope. Runs keep the owning application identity so step access can enforce the same boundary.
References: packages/…/core-workflow-access.service.ts• L32-195 packages/…/build-core-workflow-visibility-where.util.ts• L7-18 packages/…/is-core-workflow-visible-to-user.util.ts• L4-14 packages/…/can-change-core-workflow-visibility.util.ts• L3-11 packages/…/can-application-start-core-workflow.util.ts• L3-11 packages/…/workflow-trigger-dispatch.workspace-service.ts• L40-110
Runtime execution
The runtime accepts a trigger payload, creates a run from a selected core version, and advances the graph through runnable steps. It stores step state and outputs so another worker can continue after a wait, retry, or queue hand-off.
This diagram shows how a trigger becomes a durable run and how the run moves through action execution.
flowchart LR
T[Trigger payload] --> V[Resolve active core version]
V --> R[Create run with flow snapshot]
R --> Q[Admit run to queue]
Q --> E[Execute runnable steps]
E --> S[Save step state and output]
S --> W{Waiting or finished?}
W -->|Continue| E
W -->|Wake or callback| Q
W -->|End| F[Complete, fail, or stop]
Trigger dispatch
Manual, webhook, schedule, and database-event triggers enter the same run path after trigger-specific checks. A webhook request resolves the core workflow from either its core ID or its legacy workspace ID, then requires a published active version with a webhook trigger. Database events match the event name, watched fields, and record filters. They also pass through application record-read rules before a trigger job is created.
Cron dispatch checks the active core version at the time it runs. It uses a short-lived cache when possible and rebuilds that cache from active workspaces when needed. A stale or incomplete trigger entry cannot start an old or missing definition. Queue refusal is recorded as a failed run with a clear reason.
References: packages/…/workflow-trigger.controller.ts• L50-160 packages/…/core-workflow-trigger.service.ts• L39-77 packages/…/workflow-database-event-trigger.listener.ts• L335-429 packages/…/workflow-cron-trigger-cron.job.ts• L59-121 packages/…/workflow-cron-trigger-cron.job.ts• L269-341 packages/…/workflow-trigger-dispatch.workspace-service.ts• L40-110
Run state
A run stores its workflow and version IDs, trigger payload, flow snapshot, step information, and final status. The snapshot protects inspection and resume behavior when the workflow later receives a new draft or active version.
Run status covers not started, enqueued, running, completed, stopping, stopped, and failed. Step status records more detail, including success, skipped, safely failed, failed, running, and pending. A run can stop without interrupting a step that is already finishing.
References: packages/…/workflow-run.workspace-entity.ts• L17-72 packages/…/workflow-run.workspace-service.ts• L44-114 packages/…/workflow-run.workspace-service.ts• L158-236
Step progression
The executor selects runnable steps from the saved graph and runs independent paths together. Before a step runs, it checks that the action is enabled, that the run application may use it, and that usage is allowed. A step can succeed, be skipped, fail safely, fail the run, or return a wait condition.
Branch actions select their next paths. Iterators repeat a loop body for each item. The executor limits the number of steps handled in one job and queues another job when the limit is reached. The action factory maps each action type to its execution contract.
Usage is tied to the core workflow and its application. A refused quota check becomes a user-facing step result instead of an untracked execution, while a successful step is charged only when it starts its work.
References: packages/…/workflow-executor.workspace-service.ts• L87-113 packages/…/workflow-executor.workspace-service.ts• L115-309 packages/…/workflow-executor.workspace-service.ts• L391-461 packages/…/workflow-action.factory.ts• L34-116
Execution context
Each step runs with a context built from the workspace, workflow application, run creator, and available role permissions. A manual run can act for its workspace member. An automated or application run uses the owning application context.
The context limits which application-owned records, agents, connected accounts, and other targets a step can use. A deleted application or an application with no usable role stops the run instead of widening access. An unrelated application target also stops the step. Actor metadata identifies automated work as the workflow and keeps the workflow ID in the actor context. Work started by a member keeps that member as the acting actor.
Email actions use this context when they call the mail service. If a sender value names a workspace member, the action resolves that member to a connected account that can perform the requested email operation. This keeps the visible workflow actor separate from the mailbox used to send the message.
References: packages/…/workflow-execution-context.service.ts• L41-168 packages/…/workflow-execution-context.type.ts• L6-13 packages/…/build-workflow-actor-metadata.util.ts• L6-20 packages/…/email-workflow-action.base.ts• L76-138
Queue-backed processing
Queues separate trigger intake from work that can take time or wait for another system. The run record is the durable hand-off between trigger jobs, admission jobs, execution jobs, retry jobs, and resume jobs.
Run admission
The runner validates the core workflow version and creates the run before it queues work. Manual runs are admitted directly to a run job. Automated runs first enter a workspace admission path, which applies workspace limits and selects not-started runs for execution.
Each run job uses the run ID as its queue identity. Its scope includes the workflow's owning application and, when configured, a workflow-specific rate limit. Workspace admission uses a lock and updates the not-started count so concurrent workers do not admit the same run twice. A queue that refuses a trigger creates a failed run rather than losing the event.
References: packages/…/core-workflow-runner.service.ts• L50-137 packages/…/workflow-trigger-dispatch.workspace-service.ts• L40-110 packages/…/workflow-run-enqueue.workspace-service.ts• L40-160
Waiting and resume
A waiting step records the condition that must resolve instead of keeping a worker open. The condition can be a delay, record event, form answer, inbox response, or agent callback. When it is ready, the system claims the pending step and queues ordinary continuation.
The wait is armed before the step is marked pending. A late response or a second worker then sees that the step is no longer claimable and does not resume the run twice. When a run ends, its waits and suspended agent calls are released.
References: packages/…/workflow-executor.workspace-service.ts• L463-573 packages/…/run-workflow.job.ts• L223-251 packages/…/workflow-run.workspace-service.ts• L192-212
Retry and recovery
The runtime retries eligible step failures with a delayed job. A failed queue hand-off changes the run or step back to a failed state instead of leaving work marked as active with no job. Duplicate or late jobs check the saved run state before they change step state.
References: packages/…/workflow-executor.workspace-service.ts• L759-841 packages/…/run-workflow.job.ts• L40-95 packages/…/workflow-run-enqueue.workspace-service.ts• L192-219
Logic-function execution
Logic functions provide reusable code that a workflow can call with structured input. New workflow code steps store TypeScript source in the step. Older runs can still use a stored logic-function ID, which keeps upgrade compatibility without making a separate resource part of new code-step creation.
Code-step contracts
A code step exposes input fields, source code, an expected output shape, and the output shape produced by a test or run. Workflow variables fill only the input object. The source stays unchanged, and the exported entry function receives the resolved input.
The editor and server use the output shape to make later variables safe to select. Stored logic functions can also declare the output shape used by a legacy logic-function step. A new code step created through the workflow tools must include inline source.
References: packages/…/workflow-version-step-operations.workspace-service.ts• L117-215 packages/…/workflow-schema.workspace-service.ts• L275-334 packages/…/create-workflow-version-step.tool.ts• L40-67
Code-step execution
At run time the code action resolves the input, builds an execution context, and runs either inline source or a legacy logic function. Inline code uses the workflow application and the run's user context. A legacy function must be runnable in the workspace and reachable from the run application.
Execution returns data or an action error. The runtime persists code-step logs for run inspection. A stale legacy build is refreshed before execution. The frontend shows a legacy step as read-only when its source cannot be moved into the step, while normal code steps remain editable.
References: packages/…/code.workflow-action.ts• L41-177 packages/…/WorkflowEditActionCode.tsx• L86-182 packages/…/WorkflowReadonlyActionCode.tsx• L21-53
External tool actions
External tool actions resolve workflow variables before they call a service outside the graph. HTTP requests and MCP tools use the same action result and error model, so a later step can consume the returned value and the run view can show the call outcome.
An MCP tool step identifies an active server and a named tool. The runtime checks that the tool still exists, removes only null inputs that the tool schema does not require, and calls it with the run user's workspace identity. The run application must be allowed to reach the MCP server. Step tests apply the same checks and require confirmation for destructive tools.
The HTTP editor keeps an expected response sample and a derived output schema. Its test path resolves variables before sending the request and shows the response or error without changing the saved run contract.
References: packages/…/mcp-tool.workflow-action.ts• L20-89 packages/…/mcp-tool-call.service.ts• L21-101 packages/…/useHttpRequestOutputSchema.ts• L16-76 packages/…/useTestHttpRequest.ts• L41-140
Database-event logic-function triggers
Database events can also start application logic functions without a workflow step. The trigger worker finds active functions for the event, removes records that the application cannot read, applies watched-field and condition filters, and queues each admitted call.
Batch mode groups events by the user context that caused them and splits large groups into bounded jobs. This keeps one job from mixing permissions while reducing queue work for safe batches. Application limits and queue memory limits can refuse the batch, in which case the event is recorded rather than run without its required scope.
References: packages/…/call-database-event-trigger-jobs.job.ts• L48-190 packages/…/find-logic-functions-triggered-by-event-name.ts• L6-31 packages/…/transform-event-batch-to-event-payloads.ts• L20-123
AI-agent execution
An AI-agent action gives a workflow a controlled way to ask an agent to work with workflow data or a person. The action carries a prompt, optional human-input instructions, recipient and conversation scope, and an output format.
Agent interaction
An agent can finish in one turn or pause for a person, tool result, or later turn. The workflow run stores the conversation link and resumes through the same queue-backed run path. The agent result becomes the step output, so later actions can use named response fields.
References: packages/…/ai-agent.workflow-action.ts• L117-187 packages/…/workflow-agent-conversation.workspace-service.ts• L23-81
Agent permissions
An agent action uses the workflow execution context rather than an unrestricted identity. Its agent and target resources must be reachable from the run application. When an agent continues after a pause, the runtime rebuilds its context from the same run permissions.
The editor presents the agent scope as object-level CRUD grants and other permission labels. A grant for all objects remains distinct from a grant for one object, so removing a displayed row removes only that grant. This lets the workflow author review the scope without changing the underlying role model.
References: packages/…/build-workflow-agent-run-execution-context.util.ts• L7-21 packages/…/workflow-execution-context.service.ts• L110-168 packages/…/WorkflowAiAgentPermissionList.tsx• L22-131 packages/…/getWorkflowAiAgentScopePermissions.ts• L14-95
Frontend management
The web client manages workflow drafts, runs, and step output through core workflow data. It keeps local edits responsive while using the server version ID and saved run snapshot as the source of truth for durable state. It also presents action-specific checks for the accounts and tools that a step will use.
Workflow editing
The editor reads a workflow's current trigger and steps, creates or selects a draft, and sends step, edge, trigger, and position changes to the server. Local flow keys identify the trigger or steps with unsaved values. A refetch replaces only server-owned parts that have no local edit, while the server version ID remains authoritative.
Graph editing supports node selection, step insertion, duplication, and connection changes. New connections are persisted after the displayed graph changes, and a failed reconnection rolls the graph back to the last saved edge. Layout changes do not change execution links. New code steps carry source directly in the step, so ordinary edits do not need a separate legacy resource ID.
References: packages/…/buildFlowWithLocalEdits.ts• L14-46 packages/…/WorkflowDiagramCanvasBase.tsx• L241-319 packages/…/WorkflowDiagramCanvasEditable.tsx• L85-221 packages/…/useCreateStep.ts• L19-101 packages/…/useDuplicateStep.ts• L11-59 packages/…/core-workflow-version-mutation.workspace-service.ts• L64-125 packages/…/core-workflow-version-write.service.ts• L109-274
Trigger and action management
Trigger panels expose manual, schedule, webhook, and database-event settings. The cron panel validates the selected interval or expression and labels schedule values as UTC before the server validates publication. A published manual trigger also controls its command-menu visibility and record-selection scope.
Action panels keep the editor contract close to the server contract. HTTP request testing resolves the current variables before the call, output samples become selectable fields, and MCP settings save as one action update after the user pauses. Email panels select a send-capable connected account, support variable senders, and expose draft-account permission gaps. Calendar panels select a calendar account, handle missing create-event permission, and support time zone, attendee, invitation, and conferencing choices. AI-agent panels show the object and field scope that the agent will use.
References: packages/…/WorkflowEditTriggerCronForm.tsx• L65-145 packages/…/WorkflowEditActionEmailBase.tsx• L128-227 packages/…/WorkflowEditActionCreateCalendarEvent.tsx• L56-101 packages/…/useHttpRequestOutputSchema.ts• L37-76 packages/…/useMcpToolStepForm.ts• L13-50 packages/…/WorkflowEditActionMcpTool.tsx• L144-176 packages/…/WorkflowAiAgentPermissionList.tsx• L97-129
Run management
The workflow show page has Flow, Overview, and Runs views. A workflow with existing runs opens on Overview when the reader can access runs. A workflow with no runs stays on Flow, so starting the first run does not move the author away from the editor. A requested version opens on Flow, while a stored tab choice or URL hash can select another visible view.
Overview presents run count, failure count, failure rate, average duration, daily volume, and recent failed runs. The run list shows status, start time, duration, and version, then loads more history as the reader reaches the end. If run records are not readable, run summary fields remain unavailable instead of appearing as zero values.
Run insights listen for run changes and refetch with the user's time zone. Recent failed runs and run-list rows open the saved workflow run in a side panel, where the operator can inspect the run snapshot and step details.
References: packages/…/useCoreWorkflowShowPageActiveTab.ts• L16-53 packages/…/CoreWorkflowShowPageActiveTab.tsx• L27-65 packages/…/CoreWorkflowOverviewTab.tsx• L18-95 packages/…/CoreWorkflowRunsTab.tsx• L35-173 packages/…/useCoreWorkflowRunInsights.ts• L9-47 packages/…/CoreWorkflowTableRow.ts• L4-10
Run inspection
The run view builds its graph from the saved flow and marks nodes with persisted step state. It shows step input, output, errors, logs, and iteration details. A pending form step can open in the side panel so a person can provide the input that resumes the run. Historical views use core run data even when the workflow was reached through an older workspace identifier.
The visualizer derives its diagram from the run snapshot rather than the current draft. It can select the next step that needs attention and open that step in the side panel. Retry and stop actions update the run record, while live run updates keep the open view current.
References: packages/…/WorkflowRunVisualizerEffect.tsx• L103-227 packages/…/useRunWorkflowRunOpeningInSidePanelEffects.ts• L30-123 packages/…/WorkflowRunStepInputDetail.tsx• L21-127
Manual run launch
The frontend creates a local run record before it asks the core API to execute a version. This gives the run view an immediate enqueued state and lets the client subscribe to updates for that run. The hook in packages/…/useRunWorkflowVersion.tsx then opens the run in a side panel after the server accepts the request.
Manual starts use the core workflow and version IDs. A command can run once with a test trigger payload or build one payload per selected record. The client also saves pending workflow edits before it tests a single record, so the server runs the version that the author just approved.
If the run mutation fails, the client stops listening for that run and returns the error. The optimistic record is not a substitute for server execution. It only gives the operator an immediate view while the queue-backed runtime changes the durable run state.
References: packages/…/useRunWorkflowVersion.tsx• L30-169 packages/…/TriggerWorkflowVersionEngineCommand.tsx• L41-87 packages/…/TestWorkflowSingleRecordCommand.tsx• L31-63 packages/…/core-workflow-run.resolver.ts• L24-78
AI workflow tools
AI workflow tools expose the same lifecycle as the web client. A tool caller can create, validate, activate, deactivate, delete, run, and inspect workflows. It can also manage steps, edges, positions, output schemas, and workflow runs.
The complete-workflow tool creates a trigger, steps, and edges in one operation. Code steps created through it carry inline TypeScript source. The step tool gives logic-function steps a stored function ID and gives code steps source and input data. Run tools require a signed-in workspace member because a run can cause side effects.
References: packages/…/workflow-tool.workspace-service.ts• L76-114 packages/…/create-workflow-version-step.tool.ts• L40-159 packages/…/create-complete-workflow.tool.ts• L77-180
Change checklist
When these files or sections are changed, remember to consider these:
- Keep core workflow IDs, workspace version links, run snapshots, and automated dispatch maps aligned.
- Keep activation checks aligned with trigger types, schedule rules, action validity, and graph content.
- Keep inline code source, inferred input, expected output, test result, and saved output shape aligned across the editor, server, and AI tools.
- Keep MCP server and tool reach checks aligned between step creation, step testing, and run execution.
- Keep legacy logic-function compatibility isolated to old runs and read-only legacy editor states. Do not add a legacy resource requirement to new inline code steps.
- Keep run queue IDs, workflow application scope, workflow-specific limits, admission counts, and refused-dispatch failure records aligned.
- Keep wait claims aligned with pending step state, callbacks, queue continuation, and final run cleanup.
- Keep execution context, actor metadata, email account selection, agent permissions, and application reach aligned across manual and automated runs.
- Keep workflow usage checks and billing spenders tied to the same core workflow identity used by run creation.
- Keep the run flow snapshot aligned with run diagrams, step logs, form responses, retry, stop, and stale-response handling.
- Keep database-event batching aligned with read permissions, watched fields, conditions, actor context, job size, and application limits.
- Keep run insights, access-aware summary fields, time-zone handling, tab defaults, and run-list version labels aligned with the core workflow run API.
- Keep manual launch state aligned across the optimistic run record, the core run mutation, the run SSE subscription, and side-panel opening.
- Keep AI tool inputs aligned with the same workflow IDs and activation rules used by the web client.
Important failure modes
| Trigger | Consequence | Guard |
|---|---|---|
| A core version has no valid workflow identity, trigger, or steps | The run is rejected before work is queued | Core run creation checks the complete execution definition |
| A cron setting has an invalid interval or expression | Activation is rejected or the schedule would not match the editor | The client validates input and activation validates the final settings |
| A database event is not readable by the workflow application | No workflow trigger job is created for that record | Dispatch filters events through application read permissions |
| A workflow run exceeds its hard limit | The run is created as failed and does not enter execution | Runner admission checks the workspace throttle limit |
| A webhook names a missing, inactive, or non-webhook workflow | The request is rejected and no run is created | The trigger endpoint resolves the core workflow and checks its published version and trigger type |
| A private workflow is requested by another member | The workflow is hidden as not found | Access checks apply the creator or workspace visibility rule to workflows and versions |
| A concurrent draft edit changes the expected version | The save is rejected and the editor must reload | Draft writes use an edit lock and compare the expected trigger and steps |
| A trigger job is too large or queue memory is full | The workflow does not start | Dispatch records a failed run with the queue refusal reason |
| A second worker delivers a response for a pending step | The run could resume twice | Pending-step claims are locked and one-time |
| Inline code has a build or execution error | The code step returns an error and its run can retry or fail | The code action records the result and the executor applies step error policy |
| A legacy function or MCP server belongs to an unreachable application | The step is forbidden | Execution checks the run application's reach before calling it |
| An MCP server is inactive or no longer offers the selected tool | The step is rejected before the external call | Runtime lookup checks the server and tool again |
| An AI-agent scope omits the object or field access it needs | The agent cannot read or change the intended data | The editor exposes the effective scope and the run rebuilds permissions from its execution context |
| A local frontend edit is followed by a server refetch | Unsaved values could be replaced by newer server data | Local flow keys overlay only the matching trigger or steps |
| A graph reconnection fails after the local edge changes | The editor could show a link that the server did not save | The edit queue restores the last saved edge |
| A run view reads the current workflow instead of its snapshot | Historical step state can no longer match the displayed graph | The visualizer builds from the saved run flow |
| A reader cannot access workflow runs | Run counts can look like real zero values | Core workflow row types use null for hidden run summaries |
| The frontend creates an optimistic run but the core mutation fails | The temporary run cannot progress and live updates must stop | The launch hook removes the run subscription and surfaces the mutation error |
| A tool caller starts a run without a signed-in member | The side-effecting run is refused | The run tool requires member authentication |
Key files
| File | Why to read it |
|---|---|
| packages/…/workflow.entity.ts | Core workflow identity, visibility, and published-version pointers |
| packages/…/workflow-version.entity.ts | Core version content, status, and workspace compatibility links |
| packages/…/core-workflow-lifecycle.workspace-service.ts | Version validation, activation, deactivation, schedule cache, and manual command entries |
| packages/…/core-workflow-access.service.ts | Member visibility, edit rights, and application start boundaries |
| packages/…/workflow-trigger.controller.ts | Webhook resolution and public trigger checks |
| packages/…/core-workflow-run.resolver.ts | Authenticated core mutation for manual run launch |
| packages/…/workflow-version.workspace-entity.ts | Workflow version states and saved execution content |
| packages/…/workflow-run.workspace-entity.ts | Run status, snapshot, step state, and core identity links |
| packages/…/assert-version-can-be-activated.util.ts | Publication and trigger validity rules |
| packages/…/workflow-version-step-operations.workspace-service.ts | Action creation, feature gating, and graph support nodes |
| packages/…/workflow-schema.workspace-service.ts | Output contracts for actions and variables |
| packages/…/assemble-workflow-version-content.util.ts | Separation of saved execution links from editor positions |
| packages/…/core-workflow-runner.service.ts | Core version resolution, run creation, and admission choice |
| packages/…/workflow-trigger-dispatch.workspace-service.ts | Trigger queue admission and refused-run recording |
| packages/…/run-workflow.job.ts | Start, resume, retry, and awaited-step queue handling |
| packages/…/workflow-run.workspace-service.ts | Run snapshots, step state, stop, and cleanup |
| packages/…/workflow-executor.workspace-service.ts | Step progression, branching, waits, retries, quota, and completion |
| packages/…/workflow-execution-context.service.ts | Application, member, role, actor, and permission context for a run |
| packages/…/code.workflow-action.ts | Inline and legacy logic-function execution |
| packages/…/mcp-tool.workflow-action.ts | MCP tool lookup, permission checks, and calls |
| packages/…/email-workflow-action.base.ts | Workflow actor and connected-account handling for email |
| packages/…/workflow-cron-trigger-cron.job.ts | Cron cache, schedule checks, and trigger dispatch |
| packages/…/workflow-database-event-trigger.listener.ts | Database-event filtering and trigger admission |
| packages/…/call-database-event-trigger-jobs.job.ts | Database-event logic-function admission and queueing |
| packages/…/useRunWorkflowVersion.tsx | Frontend run launch, optimistic run state, live updates, and side-panel opening |
| packages/…/WorkflowDiagramCanvasEditable.tsx | Editable graph changes and persistence |
| packages/…/buildFlowWithLocalEdits.ts | Server-flow and local-edit merging |
| packages/…/useCoreWorkflowShowPageActiveTab.ts | Workflow page tab defaults and run access |
| packages/…/CoreWorkflowOverviewTab.tsx | Run health summary and recent failures |
| packages/…/CoreWorkflowRunsTab.tsx | Historical run list and version context |
| packages/…/useCoreWorkflowRunInsights.ts | Live run insight refresh and time-zone input |
| packages/…/WorkflowEditActionCode.tsx | Code editing, testing, and output-shape refresh |
| packages/…/WorkflowRunVisualizerEffect.tsx | Run snapshot diagram and step selection |
| packages/…/useRunWorkflowRunOpeningInSidePanelEffects.ts | Opening the run step that needs attention |
| packages/…/workflow-tool.workspace-service.ts | AI workflow tool surface |
| packages/…/create-workflow-version-step.tool.ts | AI step creation contract for code and logic-function steps |
| packages/…/create-complete-workflow.tool.ts | Complete workflow creation contract |