Salesforce Routing Protocol

Salesforce Routing Protocol

Purpose

This skill defines how a Salesforce matter is classified, routed to the right specialist agent, and coordinated when it crosses multiple Salesforce domains. It exists so agents never work in silos when risk crosses a domain boundary, and so every handoff is controlled and auditable rather than free-form. It does not give Salesforce or business advice; routing is a recommendation that a human owner confirms.

When to use

• A maestro agent must classify an incoming Salesforce matter and dispatch it.
• A specialist agent finds a matter has crossed into another Salesforce domain.
• A matter triggers an escalation gate and must be paused.
• Specialist agents reach conflicting recommendations.
• A matter is ambiguous and cannot be confidently assigned to a single domain.

When not to use

• The matter is already classified and a specialist agent is actively working it.
• The matter is outside Salesforce scope entirely — route to the appropriate non-Salesforce protocol.
• You need direct live-org mutation advice — that requires the salesforce-live-change-approval-protocol, not this routing skill.

Communication principles

• One accountable human owner per matter. One primary agent per matter.
• Parallel review only when the matter genuinely crosses domains.
• No agent makes a final configuration, deployment, business-process, or compliance decision.
• All high-risk cross-domain matters produce a pause and escalate recommendation unless sufficient documented controls already exist.
• Every handoff uses a salesforce-case-capsule. No free-form agent chatter.
• Every handoff preserves context, uncertainty, evidence quality, and open questions, and carries a do_not_do_list.
• Every handoff labels privilege sensitivity and privacy sensitivity.
• Every agent defaults to least-privilege access and minimum-necessary data.

Routing rules

Admin / org-config

Declarative setup, org settings, sandbox management, permission sets, profiles, sharing, OWD, role hierarchy, IP restrictions, session policy → salesforce-org-assessment-skill or salesforce-permission-model-review-skill.

Developer / code

Apex, LWC, triggers, async jobs, managed packages, ISV, API integration, deployments → salesforce-apex-lwc-code-review-skill.

Automation / process

Flow, validation rules, approval processes, Process Builder , record-triggered automation → salesforce-flow-automation-review-skill.

Security / IAM

Permission topology, toxic combinations, guest-user exposure, data classification, sharing rule widening, field-level security, connected-app OAuth scopes → salesforce-permission-model-review-skill AND salesforce-data-exposure-escalation-protocol (if triggered).

RevOps / CPQ

Pricing rules, product catalog, quote configuration, CPQ custom scripting, contract and order management → salesforce-org-assessment-skill (risk register triage); escalate business-rule decisions to human RevOps owner.

Marketing / consent

Marketing Cloud data flows, consent capture, preference centers, subscriber keys, SFMC send classifications → salesforce-marketing-consent-review-skill.

Compliance / regulated verticals

Health Cloud , Financial Services Cloud , Government Cloud regulated-data requirements, BAA/DPA obligations, audit requirements → salesforce-org-assessment-skill plus external compliance counsel.

AI / Agentforce

Agentforce agent configuration, Einstein feature setup, autonomous action scope, AI grounding and retrieval → salesforce-agentforce-risk-review-skill.

Release / deploy

Change sets, unlocked packages, CI/CD pipelines, sandboxes, destructive changes → salesforce-release-readiness-skill.

Integration / middleware

REST/SOAP/Bulk/Streaming APIs, Platform Events , CDC , MuleSoft , named credentials, external services → salesforce-integration-review-skill.

Metadata

Object model, custom fields, layouts, page layouts, Lightning record pages , deprecated types → salesforce-metadata-review-skill.

Cross-domain overlap matrix

Conflict-resolution protocol

When specialist agents disagree, follow this order and stop at the first step that is unmet:

1. Freeze any irreversible action.
2. Preserve evidence; preserve sandbox or scratch-org state if applicable.
3. State the disagreement precisely — what each agent concluded and why.
4. Separate technical risk from business/operational risk; do not collapse them.
5. Identify the accountable human owner.
6. Escalate to qualified Salesforce architect or technical lead.
7. Document unresolved assumptions and open questions.
8. Produce options, not a single conclusion.
9. Require human approval before any action.
10. Log the decision path in the audit log.

Security notes

• Routing is a recommendation, never an authorization. The protocol never approves, denies, or directs a deployment or configuration action.
• A matter is routed on sanitized signals. Never request org credentials, session IDs, OAuth tokens, customer PII, or production org IDs to classify.
• When classification is ambiguous, route to a maestro agent and mark the matter unclassified rather than guessing a specialist.
• This protocol does not authorize live org mutations; that requires salesforce-live-change-approval-protocol.

Audit log fields

• matter_id, skill_id, skill_version, invoked_by, input_hash, evidence_quality, output_verdict, escalation_fired, timestamp

Stop conditions

• A live-mutation action is requested — stop and invoke salesforce-live-change-approval-protocol.
• Classification requires access to production org credentials — stop and refuse.
• Matter involves regulated personal data and jurisdiction is unknown — stop and escalate.