@raishin/vanguard-frontier-agentic 2.3.0 → 2.6.0

package/skills/cross-functional/salesforce-data-exposure-escalation-protocol/SKILL.md

@@ -0,0 +1,165 @@
+---
+name: salesforce-data-exposure-escalation-protocol
+description: Use this skill when a Salesforce data exposure event has been detected or is strongly suspected. Triggers include: guest-user data exposure via Experience Cloud, cross-org data sync without a Data Processing Agreement, regulated-data sync in Marketing Cloud without a consent map, Experience Cloud sharing-set widening affecting personal data, and Data Cloud cross-org sharing without appropriate controls. Trigger phrases: "guest user can see records they should not", "data syncing across orgs without DPA", "sharing set was widened in production", "marketing data sync without consent", "Data Cloud sharing concern". Do not use for routine permission reviews (use salesforce-permission-model-review-skill), for pre-change risk assessment (use salesforce-live-change-approval-protocol), or for general security questions not involving a suspected exposure event. This skill governs the immediate escalation response path: pause, preserve evidence, name controllers and processors, escalate to privacy counsel and security, and document.
+allowed-tools: Read Grep Glob
+metadata:
+  author: "github: Raishin"
+  version: "0.1.0"
+  updated: "2026-05-20"
+  category: security
+  lifecycle: experimental
+---
+
+# Salesforce Data Exposure Escalation Protocol
+
+## Purpose
+This skill defines the immediate escalation response path when a Salesforce
+data exposure event has been detected or is strongly suspected. It exists
+because exposure events in Salesforce orgs — particularly involving guest
+users, Experience Cloud, Data Cloud, or Marketing Cloud — can involve
+regulated personal data that triggers legal notification obligations. The
+protocol must be followed immediately; it is not advisory after a trigger fires.
+
+## When to use
+- Guest-user data exposure: an Experience Cloud site's guest-user profile or
+  OWD gives unauthenticated access to records that should be restricted.
+- Cross-org data sync without DPA: data is flowing between Salesforce orgs or
+  to external systems without a documented Data Processing Agreement.
+- Regulated-data Marketing Cloud sync without consent map: personal or
+  regulated data is syncing to Marketing Cloud
+  without a documented lawful basis and consent map.
+- Experience Cloud sharing-set widening: a sharing set or sharing rule change
+  has expanded access to personal data beyond the intended scope.
+- Data Cloud cross-org sharing: Data Cloud
+  is sharing data across orgs without documented controls, purpose limitation,
+  or appropriate consent.
+
+## When not to use
+- Routine permission review with no suspected exposure — use `salesforce-permission-model-review-skill`.
+- Pre-change risk assessment before a deployment — use `salesforce-live-change-approval-protocol`.
+- General security audit without a specific exposure event — use `salesforce-org-assessment-skill`.
+- The trigger is hypothetical or a design question, not an observed event.
+
+## Minimum payload (required inputs)
+- Description of the suspected exposure event (sanitized, no credentials or PII beyond what is strictly necessary to describe the exposure type).
+- Trigger type (one or more from the trigger list above).
+- Environment: is the exposure in a production org? (If unknown, assume yes.)
+- Approximate time of discovery.
+- Who discovered it and how.
+
+## Escalation response path
+
+**Step 1 — Pause**
+Immediately recommend pausing any ongoing data sync, automation, or
+configuration change that is contributing to the exposure. Do not delete
+evidence. Do not attempt to silently fix the configuration without escalation.
+
+**Step 2 — Preserve evidence**
+Recommend capturing and preserving:
+- Sanitized description of the configuration state (sharing rules, sharing
+  sets, OWD, guest-user profile permissions, Data Cloud segment definitions,
+  Marketing Cloud
+data extension scope).
+- Approximate time window of exposure.
+- System or event logs if accessible (do not request log content containing PII).
+- Change history (who changed what, when — from audit trail if available).
+
+Evidence must not be modified or deleted. If litigation hold risk exists,
+flag it immediately.
+
+**Step 3 — Name controllers and processors**
+Identify (using role/placeholder references, not real names):
+- Which Salesforce org is the data controller.
+- Which system (org, cloud, middleware) is acting as a data processor.
+- Which third parties received or may have received the data.
+- Whether a DPA exists between controller and processor.
+
+**Step 4 — Escalate to privacy counsel and security**
+Recommend immediate escalation to:
+- Privacy counsel or Data Protection Officer (DPO) for assessment of
+  notification obligations under applicable law (GDPR, CCPA, HIPAA, or
+  other jurisdiction-specific requirements).
+- Information security team for technical containment assessment.
+- Salesforce Trust (
+  https://help.salesforce.com/s/trust) if the exposure may involve a
+  Salesforce platform-level issue.
+
+Do not conclude that notification is required or not required — that is a
+legal determination for qualified counsel.
+
+**Step 5 — Document**
+Produce a structured escalation record (salesforce-case-capsule with
+escalation_gate_fired = `production-data-exposure`) containing:
+- Trigger type(s) fired.
+- Evidence preservation status.
+- Controller/processor identification (placeholder references).
+- Escalation recipients (roles, not personal identifiers).
+- Open questions for privacy counsel.
+- Do-not-do list.
+- Decision owner (named human).
+
+## Workflow
+1. Receive sanitized exposure description.
+2. Match against trigger list; identify all triggers that apply.
+3. Output ESCALATE immediately — do not defer.
+4. Execute steps 1–5 in order.
+5. Produce salesforce-case-capsule with escalation_gate_fired = `production-data-exposure`.
+6. List open questions for privacy counsel (do not answer them — they require legal determination).
+7. Remind invoker: no self-remediation without human authorization.
+
+## Evidence requirements
+- Sanitized description of the configuration state at time of discovery.
+- Trigger type(s) clearly identified.
+- Time of discovery and approximate exposure window.
+- Whether regulated data (PII, HIPAA, PCI, financial) is or may be involved.
+
+## Output format
+```
+escalation_verdict: ESCALATE
+triggers_fired: [list]
+environment: production | unknown (treat as production)
+regulated_data_in_scope: yes | no | unknown
+pause_recommendation: [specific actions to pause]
+evidence_preservation_checklist: [items to capture]
+controller_processor_map: [placeholder references]
+escalation_recipients: [roles: privacy counsel, DPO, security team, Salesforce Trust if applicable]
+open_questions_for_counsel: [list — do not answer]
+do_not_do_list: [explicit prohibitions]
+decision_owner: [named human role]
+salesforce_case_capsule_required: true
+```
+
+## 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.
+- Exposure descriptions must use role and system references, not real names or customer identifiers.
+
+## Privilege / data handling rules
+- Escalation records may be subject to legal privilege if prepared in anticipation of litigation.
+  Flag this and recommend handling only through or with counsel.
+- Do not circulate the escalation record beyond the named escalation recipients.
+- The escalation record is not a public incident disclosure; do not draft customer
+  communications without qualified counsel involved.
+
+## Handoff rules
+- Always hands off to salesforce-case-capsule with escalation_gate_fired = `production-data-exposure`.
+- Escalates to privacy counsel (external) and security team (internal) as human recipients.
+- If regulated-vertical is in scope, also escalates to compliance lead.
+- Required handoff fields: trigger_type, environment, regulated_data_in_scope, evidence_preservation_status, decision_owner.
+
+## Audit log fields
+- matter_id, skill_id, skill_version, invoked_by, input_hash, evidence_quality, output_verdict, escalation_fired, timestamp
+
+## Stop conditions
+- Invoker provides real PII, credentials, or customer data in the description — stop and ask for sanitized version.
+- Invoker requests self-remediation without human involvement — stop and refuse; escalation requires human authorization.
+- Notification obligation is asserted or denied without counsel — stop and state that legal determination is required.
+
+## Security notes
+- This protocol never determines whether regulatory notification is required.
+  That is a legal determination for qualified privacy counsel.
+- Pausing is always safer than attempting a silent configuration fix.
+- Evidence must be preserved; do not recommend deletion of logs or configuration
+  snapshots even if they contain evidence of misconfiguration.
+- Salesforce Trust contacts are referenced for platform-level issues only;
+  verify current contact information at https://help.salesforce.com/s/trust before use.

package/skills/cross-functional/salesforce-data-exposure-escalation-protocol/metadata.json

@@ -0,0 +1,19 @@
+{
+  "id": "salesforce-data-exposure-escalation-protocol",
+  "name": "Salesforce Data Exposure Escalation Protocol",
+  "type": "skill",
+  "provider": "generic",
+  "harnesses": ["codex", "claude-code", "cursor", "gemini", "kiro", "other"],
+  "summary": "Immediate escalation-response protocol for Salesforce data exposure events — fires on guest-user exposure, cross-org sync without DPA, regulated-data Marketing Cloud sync without consent map, Experience Cloud sharing-set widening, and Data Cloud cross-org sharing. Required path: pause, preserve evidence, name controllers and processors, escalate to privacy counsel and security, and document.",
+  "source_type": "original",
+  "official_docs": [
+    "https://help.salesforce.com/",
+    "https://trailhead.salesforce.com/",
+    "https://developer.salesforce.com/docs"
+  ],
+  "security_notes": "Hard escalation protocol — never determines regulatory notification obligations (legal determination only); never authorizes self-remediation; always requires human authorization. Evidence must be preserved and not deleted. Sanitized inputs only; never accepts real credentials, PII, or customer data.",
+  "last_verified": "2026-05-20",
+  "path": "skills/cross-functional/salesforce-data-exposure-escalation-protocol",
+  "author": "github: Raishin",
+  "version": "0.1.0"
+}

package/skills/cross-functional/salesforce-live-change-approval-protocol/SKILL.md

@@ -0,0 +1,118 @@
+---
+name: salesforce-live-change-approval-protocol
+description: Use this skill when any proposed mutation to a live Salesforce production org must be evaluated before execution. This is a refusal-by-default gate: if any required precondition is missing, the skill stops and refuses. Required preconditions are target_org_identity, environment_type, user_identity, permission_scope, change_ticket, approval_state, dry_run_or_deployment_preview, backup_rollback_plan, test_evidence, and post_change_verification_plan. Trigger phrases: "approve this Salesforce change", "can we deploy to production", "review this org mutation", "is this change safe to push live", "run this in production". Do not use for sandbox or scratch-org changes that have no production impact, for read-only reviews of exported metadata (use domain review skills), or for classifying matter types (use salesforce-risk-taxonomy). Note: this repo is a markdown marketplace; this protocol governs advisory checklists only, not real org executions. All live mutation decisions require human authorization.
+allowed-tools: Read Grep Glob
+metadata:
+  author: "github: Raishin"
+  version: "0.1.0"
+  updated: "2026-05-20"
+  category: delivery
+  lifecycle: experimental
+---
+
+# Salesforce Live Change Approval Protocol
+
+## Purpose
+This skill is a refusal-by-default gate for any proposed mutation to a live
+Salesforce production org. It exists because production org changes carry
+irreversible risk — data exposure, broken automation, permission widening,
+and revenue-logic corruption can all result from unreviewed deployments.
+No live-mutation advisory may proceed until all ten required preconditions
+are confirmed present and documented.
+
+**Important:** This repository is a markdown marketplace for advisory
+workflows. This protocol governs checklist-based advisory review, not real
+org executions. Live org mutations require human authorization through your
+organization's actual change management process.
+
+## When to use
+- Any proposed change to a Salesforce production org is under discussion.
+- A deployment is being planned and must be evaluated for approval-readiness.
+- A change request has been submitted and needs precondition verification.
+- An escalation gate (`live-mutation` from salesforce-risk-taxonomy) has fired.
+
+## When not to use
+- The change targets only sandboxes or scratch orgs with no production impact.
+- You need read-only review of exported metadata — use the appropriate domain
+  review skill (salesforce-metadata-review-skill, salesforce-flow-automation-review-skill, etc.).
+- You need to classify matter types — use `salesforce-risk-taxonomy`.
+- You need a structured handoff — use `salesforce-case-capsule`.
+
+## Minimum payload (required inputs)
+The following ten preconditions must ALL be present. If any is missing, the
+skill outputs STOP and lists the missing items.
+
+1. **target_org_identity** — A placeholder identifier for the target org (never a real org ID or credentials). Environment type must be confirmed as production.
+2. **environment_type** — Must be `production`. If sandbox or scratch, this protocol does not apply.
+3. **user_identity** — The role or placeholder identity of the person authorizing the change (never a real username, email, or SSO ID).
+4. **permission_scope** — The permissions held by the deploying identity. Must be documented; "admin" alone is insufficient.
+5. **change_ticket** — A reference to an approved change management ticket (e.g., Jira, ServiceNow, Salesforce Cases). Must exist and be in approved state.
+6. **approval_state** — The formal approval status. Must be `approved` by a named human owner. `pending` or `draft` → STOP.
+7. **dry_run_or_deployment_preview** — Evidence that the change was previewed in a comparable environment (check-only deploy output, sandbox result, or equivalent). Must be present.
+8. **backup_rollback_plan** — A documented plan for reversing the change if it fails. Must name the rollback mechanism and estimated recovery time.
+9. **test_evidence** — Test results demonstrating the change is safe. Must include test class coverage (Apex) or equivalent automated evidence. Must meet org threshold.
+10. **post_change_verification_plan** — Steps to verify the change is working correctly after deployment. Must be documented before deployment begins.
+
+## Workflow
+1. Receive the proposed change description (sanitized, no credentials or PII).
+2. Check each precondition in order.
+3. If ANY precondition is missing or incomplete → output STOP with missing items listed.
+4. If all ten preconditions are present → output PROCEED-WITH-HUMAN-AUTHORIZATION, listing each precondition's confirmed state.
+5. Check salesforce-risk-taxonomy escalation gates. If any gate fires → output ESCALATE regardless of precondition state.
+6. Produce the advisory checklist output.
+7. Remind the invoker that human authorization is required before any real deployment.
+
+## Evidence requirements
+- All ten preconditions must be documented by the invoker.
+- Evidence must be sanitized: no real org IDs, no credentials, no customer PII.
+- Dry-run or deployment preview output must be present as text (not a promise to run it later).
+
+## Output format
+```
+advisory_verdict: STOP | PROCEED-WITH-HUMAN-AUTHORIZATION | ESCALATE
+missing_preconditions: [list, or "none"]
+escalation_gates_fired: [list from salesforce-risk-taxonomy, or "none"]
+precondition_check:
+  target_org_identity: confirmed | missing | incomplete
+  environment_type: confirmed | missing | incomplete
+  user_identity: confirmed | missing | incomplete
+  permission_scope: confirmed | missing | incomplete
+  change_ticket: confirmed | missing | incomplete
+  approval_state: confirmed | missing | incomplete
+  dry_run_or_deployment_preview: confirmed | missing | incomplete
+  backup_rollback_plan: confirmed | missing | incomplete
+  test_evidence: confirmed | missing | incomplete
+  post_change_verification_plan: confirmed | missing | incomplete
+advisory_notes: [risk observations, not authorizations]
+human_authorization_reminder: "All live org mutations require human authorization through your organization's change management process. This advisory checklist does not constitute approval."
+```
+
+## 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.
+- If the invoker provides real credentials or org IDs, decline and ask for sanitized placeholders.
+
+## Privilege / data handling rules
+- This skill never stores, logs, or repeats credentials or session tokens.
+- Production org identifiers must be replaced with placeholders in all outputs.
+- Change descriptions must not carry customer data samples.
+
+## Handoff rules
+- STOP verdict: handoff to salesforce-case-capsule with missing preconditions as blockers.
+- ESCALATE verdict: handoff to salesforce-data-exposure-escalation-protocol if data-exposure gate fired; otherwise to salesforce-case-capsule with escalation_required = true.
+- PROCEED-WITH-HUMAN-AUTHORIZATION: advisory output is presented to the human decision owner. No agent takes further action without explicit human authorization.
+
+## Audit log fields
+- matter_id, skill_id, skill_version, invoked_by, input_hash, evidence_quality, output_verdict, escalation_fired, timestamp
+
+## Stop conditions
+- Any of the ten preconditions is missing or incomplete → output STOP immediately.
+- An escalation gate fires → output ESCALATE regardless of precondition state.
+- The invoker provides real credentials or production org IDs → stop and refuse.
+- The proposed change is described as a live production mutation but environment_type is not confirmed → output STOP.
+
+## Security notes
+- Refusal is the default. The burden is on the invoker to supply all ten preconditions.
+- This protocol never issues authorization. The output is advisory only.
+- Production org IDs, credentials, and session tokens must never appear in any field.
+- This repo is a markdown marketplace; no code in this skill executes real Salesforce API calls.

package/skills/cross-functional/salesforce-live-change-approval-protocol/metadata.json

@@ -0,0 +1,19 @@
+{
+  "id": "salesforce-live-change-approval-protocol",
+  "name": "Salesforce Live Change Approval Protocol",
+  "type": "skill",
+  "provider": "generic",
+  "harnesses": ["codex", "claude-code", "cursor", "gemini", "kiro", "other"],
+  "summary": "Refusal-by-default advisory gate for proposed live Salesforce production org mutations — verifies ten required preconditions (org identity, environment type, user identity, permission scope, change ticket, approval state, dry-run preview, rollback plan, test evidence, post-change verification) and outputs STOP if any is missing. Advisory only; all live mutations require human authorization.",
+  "source_type": "original",
+  "official_docs": [
+    "https://help.salesforce.com/",
+    "https://trailhead.salesforce.com/",
+    "https://developer.salesforce.com/docs"
+  ],
+  "security_notes": "Refusal-by-default gate; never issues authorization for live org mutations; advisory checklist only. Never accepts or repeats real credentials, session tokens, or production org IDs. All live mutation decisions require human authorization through the organization's change management process.",
+  "last_verified": "2026-05-20",
+  "path": "skills/cross-functional/salesforce-live-change-approval-protocol",
+  "author": "github: Raishin",
+  "version": "0.1.0"
+}

package/skills/cross-functional/salesforce-risk-taxonomy/SKILL.md

@@ -0,0 +1,162 @@
+---
+name: salesforce-risk-taxonomy
+description: Use this skill when a Salesforce matter must be assigned a standardized matter type, risk tier, or escalation gate before routing or handoff. Defines all matter types (org-config, automation, code, integration, security/IAM, data, sales/CPQ, service/SLA, experience-cloud, marketing/consent, agentforce-AI, analytics-metrics, slack-governance, industry-vertical, architecture, compliance, release-deploy, live-mutation), risk tiers (Critical/High/Medium/Low/Unknown), and escalation gates (production data exposure, guest-user expansion, autonomous AI action, finance/revenue logic, regulated-vertical, mass change, irreversible deploy). Trigger phrases: "classify this Salesforce risk", "what tier is this change", "does this trigger an escalation gate", "assign a risk rating to this Salesforce matter". Do not use when you need to route a matter to an agent (use salesforce-routing-protocol), when you need a structured handoff (use salesforce-case-capsule), or when you need to approve a live change (use salesforce-live-change-approval-protocol). This skill classifies risk; it does not assess compliance or give Salesforce advice.
+allowed-tools: Read Grep Glob
+metadata:
+  author: "github: Raishin"
+  version: "0.1.0"
+  updated: "2026-05-20"
+  category: compliance
+  lifecycle: experimental
+---
+
+# 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.

package/skills/cross-functional/salesforce-risk-taxonomy/metadata.json

@@ -0,0 +1,19 @@
+{
+  "id": "salesforce-risk-taxonomy",
+  "name": "Salesforce Risk Taxonomy",
+  "type": "skill",
+  "provider": "generic",
+  "harnesses": ["codex", "claude-code", "cursor", "gemini", "kiro", "other"],
+  "summary": "Shared classification vocabulary for Salesforce matters — matter types, risk tiers (Critical/High/Medium/Low/Unknown), and escalation gates (production data exposure, guest-user expansion, autonomous AI action, finance/revenue logic, regulated-vertical, mass change, irreversible deploy). Does not assess compliance or authorize any action.",
+  "source_type": "original",
+  "official_docs": [
+    "https://help.salesforce.com/",
+    "https://trailhead.salesforce.com/",
+    "https://developer.salesforce.com/docs"
+  ],
+  "security_notes": "Classification only — never authorizes action; escalation gates are hard stops requiring human review. Classifies from sanitized descriptions only and refuses inputs containing live credentials, session tokens, or unredacted PII.",
+  "last_verified": "2026-05-20",
+  "path": "skills/cross-functional/salesforce-risk-taxonomy",
+  "author": "github: Raishin",
+  "version": "0.1.0"
+}