Agent Skills: sf-flow: Salesforce Flow Creation and Validation

sf-flow: Salesforce Flow Creation and Validation

Use this skill when the user needs Flow design or Flow XML work: record-triggered, screen, autolaunched, scheduled, or platform-event Flows, including validation, architecture choices, and safe deployment sequencing.

When This Skill Owns the Task

Use sf-flow when the work involves:

• .flow-meta.xml files
• Flow Builder architecture and XML generation
• record-triggered, screen, scheduled, autolaunched, or platform-event flows
• Flow-specific bulk safety, fault paths, and subflow orchestration

Delegate elsewhere when the user is:

• writing Apex-first automation → sf-apex
• creating objects / fields first → sf-metadata
• deploying metadata → sf-deploy
• seeding post-deploy test data → sf-data

Required Context to Gather First

Ask for or infer:

• flow type
• trigger object / entry conditions
• core business goal
• whether this is new, refactor, or repair
• target org alias if deployment or validation is needed
• whether related objects / fields already exist

Recommended Workflow

1. Choose the right automation tool

Before building, confirm Flow is the right answer rather than:

• formula field
• validation rule
• roll-up summary
• Apex

2. Choose the right Flow type

| Need | Default flow type | |---|---| | same-record update before save | before-save record-triggered | | related-record work / emails / callouts | after-save record-triggered | | guided UI | screen flow | | reusable background logic | autolaunched / subflow | | scheduled processing | scheduled flow | | event-driven declarative response | platform-event flow | | AI-evaluated routing (sentiment, intent, tone) | autolaunched with AI Decision element |

3. Start from a template

Prefer the provided assets:

• assets/record-triggered-before-save.xml
• assets/record-triggered-after-save.xml
• assets/screen-flow-template.xml
• assets/autolaunched-flow-template.xml
• assets/scheduled-flow-template.xml
• assets/platform-event-flow-template.xml
• assets/ai-decision-template.xml
• assets/subflows/

4. Validate against Flow guardrails

Focus on:

• no DML in loops
• no Get Records inside loops
• proper fault paths
• correct trigger conditions
• safe subflow composition
• AI Decision elements not placed inside loops (credit cost per iteration)
• AI Decision prompts include merge field references for data context

5. Hand off deployment and testing

Use:

• sf-deploy for deploy / dry-run
• sf-data for high-volume test data

High-Signal Rules

Flow architecture

• before-save for same-record field updates
• after-save for related records, emails, and callouts
• do not loop over $Record
• use subflows when logic becomes wide or repetitive

Bulk safety

• no DML in loops
• no Get Records in loops
• test with 251+ records when bulk behavior matters
• prefer Transform when the job is shaping data, not per-record branching

Error handling

• every data-changing path should have fault handling
• avoid self-referencing fault connectors
• deploy Flows as Draft first when activation risk is non-trivial

Output Format

When finishing, report in this order:

1. Flow type and goal
2. Files created or updated
3. Architecture choices
4. Bulk/error-handling notes
5. Deploy/testing next steps

Suggested shape:

Flow: <name>
Type: <flow type>
Files: <paths>
Design: <trigger choice, subflows, key decisions>
Risks: <bulk safety, fault paths, dependencies>
Next step: <dry-run deploy, activate, or test>

Flow Testing (CLI)

Run Flow tests from the command line without VS Code:

# Run all flow tests
sf flow run test --target-org <alias> --json

# Run tests for a specific flow
sf flow run test --class-names MyFlow --target-org <alias> --json

# Get results for an asynchronous run
sf flow get test --test-run-id <id> --target-org <alias> --json

Flow tests execute in the org and can take 1-5 minutes. sf flow run test returns a test run ID for asynchronous runs; use sf flow get test to retrieve results later. Always run with --json and use background execution for longer runs.

Cross-Skill Integration

| Need | Delegate to | Reason | |---|---|---| | create objects / fields first | sf-metadata | schema readiness | | deploy / activate flow | sf-deploy | safe deployment sequence | | create realistic bulk test data | sf-data | post-deploy verification | | create Apex actions / invocables | sf-apex | imperative logic | | embed LWC in a screen flow | sf-lwc | custom UI components | | expose Flow to Agentforce | sf-ai-agentscript | agent action orchestration |

Reference Map

Start here

• references/flow-best-practices.md
• references/flow-quick-reference.md
• references/orchestration.md
• references/testing-guide.md

Design / orchestration

• references/subflow-library.md
• references/governance-checklist.md
• references/transform-vs-loop-guide.md
• references/orchestration-guide.md
• references/orchestration-parent-child.md
• references/orchestration-sequential.md
• references/orchestration-conditional.md

AI Decision

• references/ai-decision-guide.md

Screen / integration / troubleshooting

• references/form-building-guide.md
• references/integration-patterns.md
• references/lwc-integration-guide.md
• references/agentforce-flow-integration.md
• references/xml-gotchas.md
• references/testing-checklist.md
• references/wait-patterns.md
• assets/

Score Guide

| Score | Meaning | |---|---| | 88+ | production-ready Flow | | 75–87 | good Flow with some review items | | 60–74 | functional but needs stronger guardrails | | < 60 | unsafe / incomplete for deployment |