Flow Automation

When to use

• A story asks for record automation and no Apex exists on that object yet.
• Reviewing or changing anything under force-app/main/default/flows/.
• Writing an @InvocableMethod so a flow can call Apex.
• A deploy failed on flow activation or on flow test coverage.
• Wave 1 of the vibe-force workflow: sf-metadata-engineer owns flows/** (see scripts/lib/sf-paths.js); sf-apex-engineer owns the invocable class it calls.

Flow and Apex are not two styles of doing the same thing. They occupy different slots in the order of execution and they interleave by that order, not by your intent.

Decision table: Flow, Apex trigger, or neither

The rule that matters: decide one owner per rule per object. A field maintained by both a flow and a trigger is a race you will debug at 3am.

Where flows sit in the order of execution

Two consequences:

1. A before-save flow mutates the same in-memory record as a before trigger, and runs first. If a before trigger later overwrites the same field, the trigger wins.
2. An after-save flow runs after every after trigger. Anything it writes goes through the whole save procedure again.

The full annotated sequence is in references/order-of-execution-for-flows.md. Trigger-side detail lives in skill sf-apex-development.

Core patterns

1. Before-save flow for same-record field updates

Set triggerType to RecordBeforeSave and recordTriggerType to Create, Update or CreateAndUpdate. Assign to the $Record variable. Do not add an Update Records element for the triggering record - that would turn a free field assignment into a second save.

Metadata shape:

<?xml version="1.0" encoding="UTF-8"?>
<Flow xmlns="http://soap.sforce.com/2006/04/metadata">
    <apiVersion>67.0</apiVersion>
    <label>Opportunity Set Renewal Date</label>
    <processType>AutoLaunchedFlow</processType>
    <status>Active</status>
    <start>
        <object>Opportunity</object>
        <triggerType>RecordBeforeSave</triggerType>
        <recordTriggerType>CreateAndUpdate</recordTriggerType>
        <filterLogic>and</filterLogic>
        <filters>
            <field>StageName</field>
            <operator>EqualTo</operator>
            <value><stringValue>Closed Won</stringValue></value>
        </filters>
        <doesRequireRecordChangedToMeetCriteria>true</doesRequireRecordChangedToMeetCriteria>
    </start>
</Flow>

2. Entry conditions, always

filters on the start element are the flow equivalent of an early return in a trigger handler. doesRequireRecordChangedToMeetCriteria set to true means the flow runs only on the save where the record became matching, not on every subsequent save while it still matches. That single element is the difference between an idempotent flow and one that fires forever.

3. After-save flow for related records

triggerType RecordAfterSave. The triggering record has an Id, so you can create children, update a parent, or publish a platform event. Everything it writes re-enters the save procedure at step 1, which is how flow-to-trigger recursion happens.

4. Asynchronous path when the work can wait

A record-triggered flow can define an asynchronous path that runs after commit (step 20). Use it for callouts and for anything that must not extend the user's save. It is the declarative equivalent of a queueable and it does not need callout=true plumbing.

Scheduled paths (scheduledPaths, API 51.0 and later) run at a defined offset from a date field or from the trigger - a declarative alternative to a scheduled job for "3 days after close".

5. Calling Apex from a flow

public with sharing class RenewalReminderAction {
    public class Request {
        @InvocableVariable(required=true label='Opportunity Id')
        public Id opportunityId;

        @InvocableVariable(label='Days before expiry')
        public Integer leadDays;
    }

    public class Result {
        @InvocableVariable(label='Reminder created')
        public Boolean created;
    }

    @InvocableMethod(
        label='Create Renewal Reminder'
        description='Creates a renewal reminder task for each opportunity'
        category='Renewals'
    )
    public static List<Result> run(List<Request> requests) {
        List<Task> tasks = new List<Task>();
        for (Request r : requests) {
            tasks.add(new Task(
                WhatId = r.opportunityId,
                Subject = 'Renewal reminder',
                ActivityDate = Date.today().addDays(r.leadDays == null ? 30 : r.leadDays)
            ));
        }
        insert as user tasks;

        List<Result> results = new List<Result>();
        for (Integer i = 0; i < requests.size(); i++) {
            Result res = new Result();
            res.created = true;
            results.add(res);
        }
        return results;
    }
}

Hard rules from the Apex Developer Guide:

• The method must be static and public or global, and its class must be an outer class.
• Only one method per class can carry @InvocableMethod.
• The only annotation that may accompany it is @Deprecated.
• At most one input parameter, and it must be a List - of a primitive, an sObject, the generic sObject, a list of those lists, or a user-defined type whose members carry @InvocableVariable.
• The return type, if not null, follows the same list rules.
• @InvocableVariable fields of type List<List<sObject>> are not supported in user-defined classes and cause a runtime error. Use that type only as a direct @InvocableMethod return type.
• Add callout=true when the method calls an external system.

The list-in, list-out signature is not a style choice. It is how the platform bulkifies the action when the flow processes many records. Write the body to match: never loop a DML statement inside it.

Prefer @InvocableMethod over the older Process.Plugin interface. The guide is explicit: the interface does not support Blob, collection or sObject data types and does not support bulk operations, and a class implementing it can only be referenced from flows.

6. Test the flow, not just the Apex it calls

FlowTest is real metadata (API 55.0 and later), suffix .flowtest, stored in the flowtests folder:

<?xml version="1.0" encoding="UTF-8"?>
<FlowTest xmlns="http://soap.sforce.com/2006/04/metadata">
    <label>Renewal reminder is created on Closed Won</label>
    <flowApiName>Opportunity_Set_Renewal_Date</flowApiName>
    <description>Covers the Closed Won entry condition</description>
    <testType>WithAssertion</testType>
    <testPoints>
        <!-- FlowTestPoint entries: parameters and assertions -->
    </testPoints>
</FlowTest>

testType WithAssertion means the test compares the actual flow outcome against the expected outcome that the assertions define. Flow tests cover record-triggered, autolaunched and Data Cloud-triggered flows.

Apex called by a flow still needs ordinary Apex tests: a FlowTest does not produce Apex coverage.

Metadata and deployment

Because FlowDefinition silently wins over status, a repository that contains both is a repository where nobody can predict which version goes live. Pick the Flow object and delete the definitions.

Deploying an active flow into production can require flow test coverage depending on org configuration. Two ways out: ship FlowTest metadata, or deploy the flow inactive and activate it afterwards. Details and the error text are in skill sf-deployment-strategies.

Limits

Flows run inside the same transaction as the DML that triggered them and share the Apex per-transaction governor limits. A record-triggered flow doing a Get Records inside a loop burns the same 100 SOQL queries an Apex loop would. CPU time is calculated for the executing code and for processes called from it.

Code Analyzer's Flow rule DbInLoop detects Get, Create, Update and Delete Records elements inside loops and names the 150 DML and 100 SOQL ceilings directly. The fix is the same as in Apex: collect into a collection variable, act once outside the loop, use the In operator to fetch by a collection of ids.

Full limit tables: skill sf-governor-limits.

Anti-patterns

Update Records on the triggering record in a before-save flow

The whole value of a before-save flow is that assigning to $Record costs nothing. Adding an Update Records element for that same record converts it into an extra save cycle and re-enters the order of execution. Assign to $Record and finish.

A flow and a trigger both maintaining one field

Flow:    Opportunity.Renewal_Date__c = CloseDate + 365   (step 3, before save)
Trigger: Opportunity.Renewal_Date__c = CloseDate + 180   (step 4, before)

The trigger wins, today. Change the flow's step and the answer changes. Decide one owner.

No entry condition

A record-triggered flow with no filters runs on every save of every record of that object, forever, including saves caused by unrelated automation. Add filters and doesRequireRecordChangedToMeetCriteria.

Get Records inside a loop

Loop over Opportunities
  Get Records: Line Items where OpportunityId = current item Id   <-- one query per iteration

Fix: one Get Records with the In operator over a collection of ids before the loop, then work in memory. Detected by Code Analyzer rule DbInLoop; vf-check analyzer fails on it.

An invocable method that loops DML

@InvocableMethod(label='Bad')
public static void run(List<Request> requests) {
    for (Request r : requests) {
        insert new Task(WhatId = r.recordId);   // one DML per request
    }
}

The platform hands you a list precisely so you can do one DML. Build the list, insert once.

Process.Plugin in new code

No bulk support, no sObject or collection types, referenceable only from flows. Use @InvocableMethod.

Shipping FlowDefinition files alongside active flows

The definition's activeVersionNumber overrides the flow's status. Remove flowDefinitions/ from the package unless you are maintaining a pre-API-44 org deliberately.

Verification

# Static analysis, including the Flow rules
node "${CLAUDE_PLUGIN_ROOT}/scripts/checks/vf-check.mjs" analyzer --changed

# Metadata XML shape and api version drift, plus the whole local gate
node "${CLAUDE_PLUGIN_ROOT}/scripts/checks/vf-check.mjs" local --changed

# Deploy validation before touching a shared org
node "${CLAUDE_PLUGIN_ROOT}/scripts/checks/vf-check.mjs" deploy-validate --target-org <alias>

# Which flows exist in the org, and which version is active
sf data query --query "SELECT Id, MasterLabel, ProcessType, TriggerType, Status, VersionNumber FROM FlowDefinitionView" --use-tooling-api --target-org <alias>

# Retrieve one flow to compare the org against the repository
sf project retrieve start --metadata Flow:Opportunity_Set_Renewal_Date --target-org <alias>

# After deploy: did the flow actually run?
sf apex get log --number 1 --target-org <alias> | grep -E "FLOW_|CODE_UNIT_STARTED"

Debug logs record flow execution under FLOW_* markers; log setup is in skill sf-debugging-logs. Post-deploy probes are in skill sf-post-deploy-verification.

References

• references/flow-metadata-reference.md - Flow, FlowDefinition, FlowTest fields, trigger type enumerations
• references/order-of-execution-for-flows.md - the full save sequence, annotated for flow authors
• references/invocable-apex.md - @InvocableMethod and @InvocableVariable rules, recipes, testing
• references/workflow-to-flow-migration.md - workflow rule to flow mapping tables, the before/after split, /vf-migrate-workflow

Process Builder is a separate migration with separate rules: skill sf-process-builder-migration (criteria-node chaining, scheduled actions, /vf-migrate-process).

Sibling skills: sf-apex-development (triggers, handler shape, order of execution), sf-minimal-change (declarative versus code, standard features first), sf-governor-limits (the limits flows share with Apex), sf-code-analyzer-quality (the Flow rule set), sf-deployment-strategies (flow activation and coverage on deploy), sf-project-structure (flows/ in the metadata directory map), sf-debugging-logs (FLOW_* log markers), sf-async-apex-patterns (when the async path needs to be Apex instead).

Sources: Apex Developer Guide "Triggers and Order of Execution", "InvocableMethod Annotation", "Passing Data to a Flow Using the Process.Plugin Interface"; Metadata API Developer Guide Flow, FlowDefinition, FlowTest; Salesforce Code Analyzer Flow rule DbInLoop. All Summer '26 / API version 67.0.