Salesforce risk taxonomy

Salesforce Risk Taxonomy

Purpose

This skill provides the shared vocabulary for classifying Salesforce matters by type, risk tier, and escalation gate. It exists so that all Salesforce specialist agents use consistent terminology when they classify, route, hand off, and escalate matters. It does not assess compliance, give Salesforce advice, or authorize any action.

When to use

• A matter must be assigned a standardized type before routing or capsule creation.
• A risk tier must be determined for a proposed change, finding, or event.
• An escalation gate must be checked before a handoff or approval proceeds.
• Agents need shared vocabulary to avoid classification drift across sessions.

When not to use

• You need to route the matter to an agent — use salesforce-routing-protocol.
• You need a structured handoff record — use salesforce-case-capsule.
• You need to approve or refuse a live org mutation — use salesforce-live-change-approval-protocol.
• You need to respond to a live data exposure event — use salesforce-data-exposure-escalation-protocol.

Minimum payload (required inputs)

• Description of the Salesforce matter (sanitized, no credentials or PII).
• Available context: environment type, change scope, data sensitivity.

Matter types

Matter type	Description
org-config	Org settings, custom settings, custom metadata, sandboxes, licenses
automation	Flow, approval processes, validation rules, Process Builder
, assignment rules
code	Apex, LWC, triggers, async jobs, static resources, packages
integration	REST/SOAP/Bulk/Streaming APIs, Platform Events
, CDC, MuleSoft
, middleware
security-iam	Permission sets, profiles, sharing rules, OWD, role hierarchy, guest users, OAuth, Shield
data	Data quality, data migration, data archival, retention policies, data classification
sales-cpq	CPQ
pricing, quoting, product catalog, Revenue Cloud
, order management
service-sla	Service Cloud
entitlements, SLAs, case management, omni-channel
experience-cloud	Experience Cloud
sites, guest-user access, sharing sets, digital experiences
marketing-consent	Marketing Cloud
, Account Engagement
, consent capture, preference centers, Data Cloud
agentforce-ai	Agentforce
configuration, Einstein
features, AI grounding, autonomous actions
analytics-metrics	Reports, dashboards, CRM Analytics
, Tableau
, metric definitions
slack-governance	Slack
integration, workspace governance, Slack-Salesforce data flows
industry-vertical	Health Cloud
, Financial Services Cloud
, Government Cloud
, regulated-vertical configuration
architecture	Org strategy, multi-org topology, data model design, platform limits
compliance	Audit requirements, BAA/DPA obligations, regulatory mapping, evidence collection
release-deploy	Change sets, DX packages, CI/CD pipelines, destructive changes, sandbox refresh
live-mutation	Any proposed change to a production org; always triggers live-change-approval-protocol

Risk tiers

Tier	Definition	Default action
Critical	Production data at imminent risk; irreversible action without approval; regulatory breach likely; autonomous AI action beyond boundary	Immediate stop; escalate; do not proceed without human authorization
High	Significant production impact; non-trivial rollback; affects regulated data, revenue logic, or broad permissions	Pause; require documented approval; dual-agent review recommended
Medium	Controlled-environment risk; reversible; affects limited scope; no regulated data	Proceed with review; log rationale; single specialist sufficient
Low	Sandboxed, non-production, or purely advisory; no data exposure; fully reversible	Proceed with standard review; document findings
Unknown	Jurisdiction, scope, data sensitivity, or counterparty identity is missing or ambiguous	Treat as High; gather evidence before reclassifying

Rule: Unknown must never be reclassified to Low without documented evidence. When in doubt, escalate.

Escalation gates

An escalation gate, when fired, requires the matter to pause and be reviewed by a named human owner before any agent takes further action. Gates are not advisory — they are hard stops.

Gate	Fires when
production-data-exposure	Any configuration, automation, or code change that could expose production records to unauthorized parties; guest-user OWD widening; sharing rule expansion on PII objects
guest-user-expansion	Experience Cloud
guest-user profile changes; sharing set modifications; public-site access widening
autonomous-ai-action	Agentforce
agent action allowlist expansion; unsupervised AI write-back to production records; AI model scope exceeding review boundary
finance-revenue-logic	Changes to CPQ
pricing rules, revenue recognition logic, billing integration, order management, or financial reporting fields
regulated-vertical	Changes to Health Cloud
, Financial Services Cloud
, or Government Cloud
orgs; HIPAA, PCI, FINRA, or FedRAMP scope
mass-change	Bulk permission assignment; data loader operations on > configurable threshold records; mass automation activation; org-wide setting change affecting all users
irreversible-deploy	Deployment of destructive changes (destructiveChanges.xml); field or object deletion; package uninstall; irreversible metadata removal

Workflow

1. Read the sanitized matter description.
2. Assign one or more matter_type values from the table above.
3. Assign risk_tier using the tier definitions; default Unknown if evidence is missing.
4. Check each escalation gate: does the matter description meet any gate condition?
5. List all fired gates.
6. Output classification: matter_type(s), risk_tier, escalation_gates_fired, rationale.

Evidence requirements

• Sanitized matter description with scope, environment, and data sensitivity stated.
• Sufficient context to determine whether regulated data is in scope.
• If environment type is unknown, treat as production.

Output format

matter_type: [one or more from the table]
risk_tier: Critical | High | Medium | Low | Unknown
escalation_gates_fired: [list, or "none"]
rationale: [one paragraph, evidence-based]
missing_evidence: [what would change the classification]

Redaction rules

• Never request secrets, credentials, OAuth tokens, refresh tokens, session IDs, MFA seeds, customer PII.
• Sanitize org IDs, user IDs (replace with placeholders) before sharing in outputs.

Privilege / data handling rules

• Classification uses sanitized descriptions only.
• If the description contains what appears to be production data, credentials, or PII, decline and ask for a redacted version.
• Regulated-vertical matter types always require privacy_sensitivity = Regulated-Data.

Handoff rules

• Classification output is consumed by salesforce-case-capsule (capsule population) and salesforce-routing-protocol (matter routing).
• Required handoff fields: matter_type, risk_tier, escalation_gates_fired, rationale.

Audit log fields

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

Stop conditions

• Description contains live credentials, session tokens, or unredacted PII — stop and refuse.
• Matter type cannot be determined from available context — output Unknown tier and request clarification.
• Regulated-vertical scope is asserted but jurisdiction is unknown — escalate before classifying.

Security notes

• This skill produces classification labels only; it does not authorize any change.
• Risk tier Unknown is a safe default; never suppress it to appear more conclusive.
• Escalation gates are hard stops, not suggestions. A gate that fires must pause the matter.