Use when designing or reviewing schedule-triggered flows for recurring automation, replacement of time-based workflow patterns, bounded record selection, idempotent processing, and escalation to Apex when volume is too high. Triggers: 'scheduled flow design', 'nightly flow job', 'time based workflow replacement', 'schedule triggered flow limits'. NOT for record-triggered scheduled paths or large-scale batch processing that should be built directly in Batch Apex.
Standardmäßig ist der Prompt ausgewählt, der zuerst die Quelle prüft. Sie können zu einem direkten Befehl wechseln oder eine lokale Kopie herunterladen.
Quelldateien prüfen
Lesen Sie SKILL.md und alle von SkillsMP angezeigten Begleitdateien, bevor Sie sich für eine Installation entscheiden.
Mit Codex oder Claude installieren Kopieren Sie diesen Prompt, fügen Sie ihn in Codex, Claude oder einen anderen Assistant ein und lassen Sie die Skill-Seite prüfen und installieren.
Ein direkter Befehl überspringt den Prüf-Prompt. Prüfen Sie die Quelle, bevor Sie ihn ausführen.
Use when designing or reviewing schedule-triggered flows for recurring automation, replacement of time-based workflow patterns, bounded record selection, idempotent processing, and escalation to Apex when volume is too high. Triggers: 'scheduled flow design', 'nightly flow job', 'time based workflow replacement', 'schedule triggered flow limits'. NOT for record-triggered scheduled paths or large-scale batch processing that should be built directly in Batch Apex.
["when should i use a scheduled flow","nightly automation with schedule triggered flow","scheduled flow versus batch apex","time based workflow replacement with flow","scheduled flow volume and retry design","scheduled flows isn't working"]
inputs
["what recurrence is needed and whether the record set can be bounded tightly","how much data the flow will inspect or mutate on each run","what should happen on rerun, partial failure, or duplicate processing"]
outputs
["scheduled-flow design for record selection, processing, and escalation boundaries","review findings for unbounded scope, weak retry behavior, and volume risk","guidance on when to stay in Flow versus move heavy work to Apex"]
dependencies
[]
version
2.1.0
author
Pranav Nagrecha
updated
"2026-07-08T00:00:00.000Z"
Use this skill when automation needs to run on a recurring cadence rather than directly from a record event. Schedule-triggered flows work well for bounded recurring tasks such as reminders, renewals, or cleanup logic. They become risky when teams treat them like a general-purpose batch engine and let the record scope or side effects grow without discipline.
The three most common failure modes of scheduled flows in production: (1) unbounded start filter that touches half the org's records on every run, (2) missing idempotency so the same records get "reminded" on every nightly run, (3) silent failure because nobody is monitoring the schedule-trigger error queue. This skill exists to prevent all three — and to know when to stop and use Batch Apex instead.
Before Starting
Check for salesforce-context.md in the project root. If present, read it first.
Gather if not available:
What recurrence is required, and is the automation tied to a business calendar or to a record event that might fit a scheduled path better?
How many records could match the criteria on a normal day, and what is the worst-case volume during seasonal or data-load spikes?
How will the flow avoid duplicate work if it runs again, partially fails, or overlaps with another automation?
What is the expected runtime budget (each scheduled-flow invocation must finish within async limits)?
Who owns the error-email recipient for this schedule? What's the SLA on response?
Core Concepts
The core design problem in scheduled flows is not the schedule itself. It is bounded scope plus repeat-safe behavior. A recurring automation that cannot explain which records it will touch, why it will touch them only once when appropriate, and when it should escalate out of Flow is not ready for production.
Choose The Right Time Primitive
Three options; only one is a schedule-triggered flow:
Primitive
When to use
When NOT
Schedule-triggered flow
Periodic scan of records matching time-based criteria independently of save events (nightly lease-renewal reminders, quarterly data cleanup)
When triggered from a record event — use scheduled path instead
Scheduled Path on record-triggered flow
Record event → wait N days → do work (send follow-up email 14d after Case created)
When there's no triggering record — schedule-triggered is the answer
Batch Apex scheduled via System.schedule
High-volume periodic processing, complex joins, large record sets
When the workload truly fits Flow's model (small set, simple operations, admin-maintainable)
A wrong primitive choice is the most common scheduled-flow design error. If the requirement starts "when a record is X and Y days old", it's likely a scheduled path, not a schedule-triggered flow.
Two Execution Modes: With Or Without An Object
The object in the Start element is optional, and that single choice changes the flow's entire execution model. Decide it before you draw anything else.
Object selected (+ criteria)
No object selected
What runs
The platform queries matching records and creates one interview per record
A single interview, once per scheduled run
$Record
Available — the flow's logic operates on one record
Not available
Batching
Interviews run in batches (default 200, settable 1–200 in the Start element)
N/A — one interview
Get + Loop over the same object
Anti-pattern — the platform already iterates for you
Correct and expected
Trailhead states it directly: "When the flow runs, it finds all of that object's records that match the criteria, and makes them available through $Record," and "If you don't set an object and criteria, $Record isn't available." Trailhead also notes $Record "actually refers to multiple records" — meaning it spans the whole matching set across interviews; within any single interview it resolves to exactly one record, which is why you author for one.
The consequence most builders miss: when an object is selected, you author the flow as if it handles exactly one record. Adding a Get Records + Loop over that same object re-fetches the whole set inside every interview — N interviews × N iterations instead of N. The record scope belongs in the Start element's criteria, not in a Loop.
Conversely, an object-less schedule-triggered flow is just a scheduled job. There is no $Record and nothing iterates for you, so Get Records + Loop (or Transform / Collection Filter / Collection Sort) is the correct shape — a loop at the top of this mode is not an anti-pattern.
Bounded Selection Comes First
The start criteria should narrow the job to a realistic set of records. Wide-open scans with complex branching downstream create both performance risk and support ambiguity. Good scheduled flows have clear filters, explicit stop conditions, and a manageable result set per run (< 250 records per run is a good rule of thumb for pure-Flow execution).
Concrete selection criteria should combine:
Type/status filter — Status = 'Open' AND RecordType = 'Support'.
Processing marker — Last_Reminded_Date__c != TODAY (see idempotency below).
Filters that rely on computed fields or formula predicates WILL slow the scheduled query. Prefer indexed fields + date ranges + processing flags.
Idempotency Matters In Recurring Automation
A recurring flow should be able to answer: "has this record already been processed for this job window?" Without idempotency, the same records get reminded, updated, or emailed every run — a frequent source of "the system spammed our customers" incidents.
Idempotency patterns:
Pattern
Implementation
Tradeoff
Processing flag
Processed_Today__c = true set at end of processing; reset nightly
Simple; requires a reset job
Last-processed date
Last_Processed_At__c = <timestamp>; filter on Last_Processed_At__c < TODAY
No reset needed; field can stale
Status transition
Move record to a terminal status that the filter excludes
Cleanest; requires business-meaningful status
External lock
Write to a custom "job log" object; check before processing
Most robust; highest overhead
Flow Is Not Always The Final Execution Engine
When the recurring job becomes large (> 500 records per run), computationally heavy (complex joins or multi-step transformations), or deeply iterative (nested loops with DML), the right answer may be to use the scheduled flow only as a lightweight orchestration layer or move the workload into Batch Apex entirely. The schedule does not justify using the wrong execution model.
Signals Flow is no longer the right engine:
Record count per run > 500.
CPU time approaches the async limit (60s sync / still constrained async).
The Flow has > 5 DML operations in a loop.
The job needs to run more than daily to stay current.
Apex invocables are doing most of the real work (then just make it Batch Apex).
Common Patterns
Pattern 1: Narrow Nightly Follow-Up Flow
When to use: Recurring reminder or status update on a bounded slice of records.
Structure (object-selected mode — the platform iterates, so author for ONE record):
Start: Daily at 2am
Object: Case
Criteria: Status = 'Pending' AND CreatedDate < TODAY - 7 AND Last_Reminded_Date__c != TODAY
Plan (runs once per matching record, $Record is that record):
└── [Send Custom Notification OR Email Alert] → addressed from $Record
└── [Update Records: $Record.Last_Reminded_Date__c = TODAY]
└── [Fault path: log + notify admin]
No Get Records. No Loop. The Start element's criteria are the record selection; adding a Get + Loop here would re-query every matching Case inside every interview.
Object-less variant: if you leave the object blank (say, to build a daily digest across several objects), the flow runs as one interview and you do need Get Records + Loop — plus an explicit LIMIT and a bulk Update after the loop, since nothing bounds the set for you.
Why not the alternative: A broad nightly sweep without clear selection becomes unpredictable and expensive.
Pattern 2: Scheduled Flow As Lightweight Orchestrator
When to use: The timing belongs in Flow, but the heavy work belongs elsewhere.
Structure:
Schedule: Weekly on Monday 3am
Plan:
└── [Get Records: this week's candidates (narrow filter)]
└── [Decision: record_count > threshold?]
└── Yes → [Invocable Apex: kick off Batch Apex job with candidate IDs]
└── No → [Process inline via Loop + subflow]
Flow owns the timing and the "is this worth running today" check. Apex owns the work.
Pattern 3: Replace Time-Based Workflow With Explicit Criteria
When to use: Legacy time-based Workflow Rule being migrated to Flow.
Structure: The Workflow Rule's "Evaluate every time a record is created or edited" + time-based action becomes a record-triggered flow's Scheduled Path (NOT a schedule-triggered flow) — because the behavior is record-event-driven, not portfolio-scan-driven.
For ACTUAL portfolio-scan automation being migrated from custom Apex scheduled jobs: evaluate whether it's really Flow-shaped first. Many scheduled Apex jobs exist because there was no good Flow primitive; now that scheduled flows + scheduled paths exist, some can migrate, but not all.
Pattern 4: Run Log + Monitoring
When to use: Any production scheduled flow — this is not optional.
Structure:
Start of run → [Create Records: Scheduled_Flow_Run__c with start_time + run_id]
... main work ...
End of run → [Update Records: the same log record with end_time + record_count + status]
Fault path → [Update Records: log record with status='failed' + fault_message]
A Scheduled_Flow_Run__c object (or equivalent) gives operations:
Run history — did the 3am job actually run last night?
Duration trends — is the nightly job taking longer over time?
Failure visibility — alert when status='failed' appears without alerting on every successful run.
Decision Guidance
Situation
Recommended Approach
Reason
Recurring job scans a bounded set of records daily or weekly
Schedule-triggered flow (Pattern 1)
Good fit for predictable recurring automation
Follow-up tied to a specific record save event
Scheduled path on record-triggered flow
Event-based, not portfolio-wide
Job volume or processing complexity high
Batch Apex or async code boundary
Flow is not the best execution engine for heavy batch work
Repeated reminders need duplicate prevention
Add idempotent markers or date fields
Recurring runs must know what's already processed
Selection criteria vague or unbounded
Redesign before scheduling
Schedule will amplify poor scope decisions
Job needs to run multiple times per day
Consider sub-hourly schedule (flow limits) or move to Batch Apex
Flow scheduling granularity may not fit
Requires external callout per record
HTTP Action inline for small sets; async via Platform Events for large
Per-record callouts exhaust callout limits
Review Checklist
Schedule primitive is correct (scheduled flow vs scheduled path vs Batch Apex).
Start criteria narrow the candidate record set intentionally (no bare Active=true).
Idempotent markers prevent duplicate tasks, emails, or updates across runs.
Failure handling exists for every DML element (Pattern A from flow/fault-handling).
Volume expectations were reviewed for normal and worst-case runs.
Heavy work was escalated out of Flow when appropriate.
Run log exists (custom object or equivalent) for operational monitoring.
Error-email recipient is a monitored mailbox, not a single inactive admin.
Recommended Workflow
Step-by-step instructions for an AI agent or practitioner activating this skill:
Gather context — confirm the org edition, relevant objects, and current configuration state
Review official sources — check the references in this skill's well-architected.md before making changes
Implement or advise — apply the patterns from Common Patterns above
Validate — run the skill's checker script and verify against the Review Checklist above
Document — record any deviations from standard patterns and update the template if needed
Salesforce-Specific Gotchas
A recurring schedule amplifies weak selection logic — broad filters turn into repeated large jobs quickly.
Recurring automation needs duplicate prevention explicitly — without markers, the same records can be processed on every run.
Scheduled flows are not a substitute for true batch architecture — the timing may fit, while the execution model does not.
A record-event requirement may fit scheduled paths better — choosing the wrong time primitive creates unnecessary complexity.
Scheduled-flow errors email the Process Automation user, same as record-triggered flows. Verify the recipient is monitored annually.
Scheduled flows DO NOT retry automatically on failure — a failed run just... doesn't run. The next scheduled run tries fresh. Build retry logic into the flow if the business needs it.
Two schedules overlapping can compete for records — if two scheduled flows both filter "Status = Open", run one, the records transition, then the second flow sees different results than the first. Design mutually-exclusive filters or serialize schedules.
Salesforce limits the number of scheduled jobs per org — check AsyncApexJob + scheduled-flow limits before adding the Nth schedule.
Daily-at-2am is everyone's default — so everyone's org hits Salesforce's scheduler peak load at 2am. Spread actual scheduled times to avoid platform contention.
Scheduled flows can't be paused mid-run — once started, they run to completion or failure. Design the scope accordingly.
Proactive Triggers
Surface these WITHOUT being asked:
Scheduled flow with no idempotency marker → Flag as Critical. Will double-process on retry or next run.
Scheduled flow with unbounded start criteria (e.g. "Active = true" alone) → Flag as Critical. Volume bomb waiting to happen.
Record-event requirement being built as scheduled flow instead of scheduled path → Flag as High. Wrong primitive.
Scheduled flow > 500 records per run with no escalation plan → Flag as High. Probably needs Batch Apex.
No run log / monitoring surface for production scheduled flow → Flag as High. Silent failure invitation.
Multiple scheduled flows on same object with overlapping filters → Flag as Medium. Coordination risk.
Scheduled flow error email recipient is inactive → Flag as Critical. Unseen failure pattern.
All scheduled flows set for 2am → Flag as Low. Platform-peer peak-load contention; spread schedules.
Output Artifacts
Artifact
Description
Scheduling design
Recommendation for schedule-triggered flow, scheduled path, or Apex
Scope and idempotency plan
Rules for bounded selection and duplicate prevention
Volume review findings
Risks around record set size, side effects, execution model choice
Run-log specification
Scheduled_Flow_Run__c schema + dashboard guidance
Related Skills
flow/flow-bulkification — use when the recurring job may still break under shared limit pressure.
flow/fault-handling — use when the main challenge is how background failures should be surfaced.
flow/record-triggered-flow-patterns — use when a scheduled path on a record-triggered flow is the right primitive instead.
apex/batch-apex-patterns — use when the workload has clearly crossed into true batch-processing territory.
standards/decision-trees/async-selection.md — upstream decision tree for scheduled flow vs Batch Apex vs Queueable.
