@raishin/vanguard-frontier-agentic 2.10.0 → 2.13.0

package/skills/sap/sap-data-privacy-analytics-ai-protocol/SKILL.md

@@ -0,0 +1,162 @@
+---
+name: sap-data-privacy-analytics-ai-protocol
+description: Cross-functional coordination protocol governing data privacy, analytics, and AI governance handoffs in SAP landscapes. Covers PII in Datasphere data products, analytics exports from SAP Analytics Cloud, RAG pipelines, embedding generation, prompt logging, SAP Generative AI Hub usage, Joule adoption, and model outputs that could leak sensitive business data. Defines data classification, consent and purpose limitation gates, approval requirements, and audit packaging. Never mutates live systems and never bypasses any guarded-mutating gate.
+allowed-tools: Read Grep Glob WebSearch WebFetch
+metadata:
+  author: "github: Raishin"
+  version: "0.1.0"
+  updated: "2026-06-19"
+  category: compliance
+  lifecycle: experimental
+---
+
+# SAP Data Privacy Analytics AI Protocol
+
+## Purpose
+
+This skill defines the cross-functional coordination and handoff contract between Data Privacy, Analytics, and AI Governance functions when SAP analytics or AI workloads involve personal data, sensitive business data, or outputs that could leak confidential information. It establishes data classification gates, consent and purpose limitation requirements, approved data flow paths for AI pipelines, logging and retention obligations for AI interactions, and approval requirements before sensitive data enters any analytics export, embedding pipeline, RAG index, or generative AI prompt.
+
+This is a governance coordination document. It does not mutate any Datasphere space, SAP Analytics Cloud model, AI Core deployment, Generative AI Hub configuration, or Joule skill. All mutations remain gated behind the relevant guarded-mutating operator agents. No guarded gate is bypassed.
+
+## When to use
+
+Invoke this protocol when any of the following trigger conditions apply:
+
+- **PII in analytics pipelines**: A Datasphere data product, SAP Analytics Cloud story, or exported dataset includes columns that contain or could derive personal identifiable information (name, email, employee ID, national ID, HR attributes, financial account identifiers, or any combination that constitutes indirect PII).
+- **Sensitive business data in analytics exports**: Analytics exports from SAP Analytics Cloud, Datasphere, or integrated data sources include revenue forecasts, M&A pipeline data, margin data, compensation benchmarks, or any data classified as confidential or restricted under the organization's data classification policy.
+- **RAG pipeline data ingestion**: A retrieval-augmented generation pipeline is being designed or reviewed that would index Datasphere data products, SAP Analytics Cloud content, ERP master data, or HR data for use as grounding context for a large language model.
+- **Embedding generation from SAP data**: SAP data (ERP documents, HR records, financial statements, customer records) is being converted into vector embeddings for storage in a vector database used by an AI model.
+- **Prompt logging and retention**: An AI system built on SAP Generative AI Hub, Joule, or a custom AI Core deployment logs user prompts or model responses in a way that may capture sensitive data from system context, few-shot examples, or user input.
+- **Generative AI Hub usage review**: A team is deploying or expanding usage of SAP Generative AI Hub and the prompt designs, model configurations, or orchestration flows have not been reviewed for data privacy compliance or confidential data leakage risk.
+- **Joule adoption governance**: Joule is being activated for a new business function (HR, Finance, Procurement, Sales) and the grounding data, user context injection, and output logging approach have not been assessed for PII handling, data minimization, and consent requirements.
+- **Model outputs that could leak data**: A generative AI model produces outputs (generated text, structured responses, summaries, recommendations) that may contain or reconstruct sensitive data from training or retrieval context.
+
+## Participating agents
+
+The following agents operate within this protocol. Each holds a defined role and must not act outside its lane without explicit cross-function approval per the decision rights table.
+
+- `sap-datasphere-data-product-architect-agent` — Specialist for Datasphere space governance, data product design, data access control layer (DAC), data sensitivity classification within Datasphere, and approved data flow paths. Does not modify live Datasphere spaces or data products.
+- `sap-analytics-cloud-planning-governance-agent` — Specialist for SAP Analytics Cloud model governance, story-level data access controls, export policy compliance, and business intelligence data classification. Does not modify live SAC models or export configurations.
+- `sap-ai-core-genai-hub-governance-reviewer-agent` — Specialist for SAP AI Core deployment governance, Generative AI Hub model and orchestration review, prompt design compliance, logging configuration assessment, and AI model output risk review. Does not modify live AI Core deployments or Generative AI Hub configurations.
+- `sap-joule-governance-adoption-agent` — Specialist for Joule skill activation governance, grounding data scope review, user context injection assessment, PII handling in Joule interactions, and adoption readiness for business functions. Does not modify live Joule configurations.
+
+## Required evidence
+
+Before any cross-function handoff is initiated, the following evidence must be assembled by the requesting function:
+
+1. **Data source inventory**: List of all SAP data sources involved (Datasphere spaces and data products, SAP Analytics Cloud models and stories, ERP datasets, HR datasets), with system identifiers, data classification level, and data owner.
+2. **Data flow diagram**: Documented path from source data through any transformation, export, embedding generation, RAG indexing, or AI prompt construction step to the final output or consumer.
+3. **PII and sensitivity assessment**: Identification of all personal data elements (direct and indirect PII) and sensitive business data elements present in each data source in scope. Assessment performed against the organization's data classification policy.
+4. **Consent and purpose documentation**: Legal basis for processing personal data in analytics or AI workloads (consent records, legitimate interest assessment, or contractual basis), and confirmation that the stated AI or analytics purpose is within the scope of the original collection purpose.
+5. **Logging and retention configuration**: Description of what is logged by the AI system (prompts, responses, retrieved context, user identifiers), where logs are stored, retention period, and access controls on log storage.
+6. **Model and orchestration configuration**: For AI Core or Generative AI Hub deployments, the model ID(s) in use, orchestration flow, grounding data scope, system prompt content (with any sensitive business context highlighted), and whether model output is cached or stored.
+7. **Export and sharing controls**: For analytics exports, the destination (external system, third party, public dataset), the access control applied to the export, and whether re-identification risk has been assessed.
+
+## Redaction policy
+
+Sensitive data elements must be treated as follows before evidence is shared across functions or included in audit documentation:
+
+- **PII in data flow diagrams**: Replace specific field values with data type labels (e.g., `[EMAIL]`, `[EMPLOYEE_ID]`, `[SALARY]`). Include field names and classification levels, not actual data values.
+- **System prompt content**: If the system prompt for a Generative AI Hub or AI Core deployment includes confidential business data (pricing rules, M&A targets, compensation bands), the prompt is shared only with the AI governance function and Legal. Operational teams see a classification summary, not the full prompt text.
+- **Prompt logs**: Prompt log samples shared for governance review must be anonymized — remove user-identifying information, replace actual data values with type labels, and limit samples to the minimum needed for the review.
+- **Embedding vectors**: Raw vector data is not shared outside the AI engineering function. The governance review covers the source data classification and the access control on the vector store, not the vectors themselves.
+- **SAC story data**: Live story data containing confidential business metrics is not included in audit packages. The audit package references the story ID, data model, and classification level.
+
+## Decision rights
+
+| Decision | Primary authority | Secondary approval required | Notes |
+|---|---|---|---|
+| Classify a data element as PII or sensitive | Data Privacy Officer (or delegate) | Data Owner confirmation | Classification is advisory input to all downstream decisions |
+| Approve a data product for AI/analytics use with PII | Data Privacy Officer | Legal (if cross-border data transfer involved) | Required before data enters any RAG index or embedding pipeline |
+| Approve a Generative AI Hub deployment for a business function | AI Governance lead | Data Privacy Officer review | Required before production deployment |
+| Approve Joule activation for a new business function | AI Governance lead + Business Function owner | Data Privacy Officer (if HR or payroll data in scope) | Required before Joule grounding data scope is expanded |
+| Approve an analytics export containing sensitive business data | Data owner | Business unit CISO review | Required for external or third-party destinations |
+| Approve prompt logging that captures user context with PII | Data Privacy Officer | AI Governance lead | Logging configuration must meet retention and minimization requirements |
+| Suspend or restrict a running AI deployment | AI Governance lead | CISO (if security risk) | Routes through sap-ai-core-genai-hub-governance-reviewer-agent advisory output; execution is separate |
+| Approve a cross-border data transfer for AI training or inference | Legal + Data Privacy Officer jointly | — | Requires transfer impact assessment for GDPR-regulated data |
+
+## Escalation owners
+
+| Scenario | First escalation owner | Second escalation owner |
+|---|---|---|
+| PII discovered in production AI prompt logs without approved legal basis | Data Privacy Officer | Chief Privacy Officer / General Counsel |
+| Sensitive business data (M&A, pricing) found in RAG index without data owner approval | Chief Information Security Officer | General Counsel |
+| Generative AI model output reconstructs confidential data | AI Governance lead | CISO + Chief Privacy Officer |
+| Cross-border data transfer for AI inference without transfer mechanism | Legal | Chief Privacy Officer |
+| Joule grounding data includes HR compensation or performance data without HR and DPO approval | Data Privacy Officer | Chief People Officer |
+| Prompt logs retained beyond approved retention period | Data Privacy Officer | Legal |
+| Third-party analytics export containing undisclosed PII | Data Privacy Officer + Legal jointly | — |
+
+## Irreversible-action gate
+
+The following actions are classified as irreversible or high-consequence and must not proceed without the approvals listed:
+
+- **Indexing PII-containing data into a RAG vector store in production**: Requires Data Privacy Officer approval confirming legal basis, data minimization review, and access control validation on the vector store. Once indexed, retrieval paths are difficult to audit fully; approval must precede indexing.
+- **Activating Joule for HR or payroll business functions**: Requires Data Privacy Officer review of the grounding data scope and user context injection design, plus HR lead confirmation that Joule responses will not expose compensation or performance data to unauthorized users.
+- **Publishing a Datasphere data product containing PII as a federated source for AI pipelines**: Requires data owner sign-off and Data Privacy Officer approval. Publishing cannot be reversed without understanding all downstream consumers already connected.
+- **Deploying a Generative AI Hub orchestration flow with a system prompt containing confidential business data**: Requires AI Governance lead approval and Data Privacy Officer review of the system prompt. Prompt content may be cached by intermediate infrastructure; deployment is effectively irreversible without a full redeployment.
+- **Exporting an analytics dataset containing sensitive business data to an external or third-party destination**: Requires data owner and CISO approval. Data shared externally cannot be recalled; re-identification risk assessment must be completed before export.
+
+This protocol does not execute any of these actions. It surfaces the required approvals and routes the action to the appropriate advisory output; execution remains with the relevant operator agent or system administrator.
+
+## Approval requirements
+
+All cross-function data governance actions under this protocol require written approval before execution. The audit package must include the approval record for every gate cleared.
+
+Minimum approval quorum per action class:
+
+- **Data classification decisions**: Data Privacy Officer sign-off with data owner confirmation.
+- **AI pipeline approvals involving PII**: Data Privacy Officer + AI Governance lead.
+- **Analytics export to external/third-party**: Data owner + CISO.
+- **Cross-border data transfer for AI**: Legal + Data Privacy Officer jointly.
+- **Joule or Generative AI Hub production deployment with sensitive grounding data**: AI Governance lead + Data Privacy Officer + (if HR data) HR lead.
+
+## Audit package
+
+The audit package for every protocol invocation must contain:
+
+1. **Scope summary**: Business function, data sources in scope, AI or analytics workload description, and trigger condition.
+2. **Data inventory**: All data elements assessed, classification level per element, data owner, and legal basis for processing (for PII).
+3. **Data flow diagram**: Documented and reviewed path from source to consumer, with PII and sensitive data elements labeled per the redaction policy.
+4. **PII and sensitivity assessment**: Results of the PII classification exercise, indirect PII combinations identified, and sensitive business data categories flagged.
+5. **Consent and purpose documentation**: Legal basis records and purpose limitation confirmation.
+6. **Logging and retention configuration review**: What is logged, retention period, access controls, and Data Privacy Officer confirmation of compliance.
+7. **Approval records**: Written approvals for each gate cleared, with approver identity, date, and scope of approval.
+8. **Irreversible-action gate log**: For each irreversible action, the gate status (cleared / blocked / pending) and the approval documentation reference.
+9. **Escalation log**: Each escalation event, who was notified, when, and the response received.
+10. **Residual risk statement**: Any data privacy or confidentiality risks not fully mitigated, the rationale for acceptance, and the compensating controls in place.
+
+## Refusal conditions
+
+This protocol and all participating agents must refuse the following requests:
+
+- Approving the indexing of PII or sensitive business data into a RAG pipeline without documented Data Privacy Officer approval and confirmed legal basis.
+- Sharing unredacted system prompt content containing confidential business data with operational teams.
+- Confirming that an AI deployment is "privacy compliant" without a completed data flow assessment and Data Privacy Officer sign-off.
+- Bypassing any irreversible-action gate listed in this protocol.
+- Accepting verbal approval for any cross-function data governance action.
+- Advising that prompt logs containing PII may be retained indefinitely — all prompt log retention must be bounded and consistent with the approved purpose and applicable data protection law.
+- Recommending that a cross-border data transfer for AI inference proceed without a completed transfer impact assessment where GDPR or equivalent regulation applies.
+
+## Evidence rules
+
+Label all claims with one of:
+
+- `documentation-based` — grounded in SAP Datasphere, SAP Analytics Cloud, SAP AI Core, SAP Generative AI Hub, or SAP Joule documentation; or in GDPR, NIST AI RMF, ISO/IEC 42001, or equivalent regulatory guidance
+- `user-provided evidence` — data flow diagrams, data classification outputs, system prompt excerpts (redacted), prompt log samples, deployment configurations, or written descriptions provided by the requesting function
+- `inference` — derived reasoning not directly confirmed by official documentation or user-provided evidence; must always be labeled as such
+
+## Response minimum
+
+Every protocol invocation must return, at minimum:
+
+- **Trigger classification**: Which trigger condition(s) apply and which participating agents are activated.
+- **Data inventory status**: Which data sources have been assessed, which PII and sensitive data elements have been identified, and which assessments are outstanding.
+- **Redaction confirmation**: Confirmation that sensitive data elements have been identified and the redaction policy has been applied to shared evidence.
+- **Legal basis status**: Whether a legal basis for PII processing in the analytics or AI workload has been confirmed, is pending, or is absent.
+- **Approval gate status**: For each required approval, who holds authority, what documentation is required, and whether the gate is cleared or blocked.
+- **Irreversible-action gate status**: Whether any irreversible actions are pending, what approvals are outstanding, and whether the gate is cleared to proceed.
+- **Logging and retention compliance status**: Whether the AI system's logging configuration has been reviewed and confirmed compliant with approved retention periods and data minimization requirements.
+- **Escalation notice**: If PII is present in production AI logs without legal basis, if sensitive business data is in a RAG index without approval, or if a cross-border transfer is occurring without a transfer mechanism — the escalation notice must appear before any other recommendation, naming the escalation owners.
+- **Audit package status**: Which audit package elements are populated and which are outstanding.
+- **Next step**: The single next action the requesting function must take, with the responsible owner named.

package/skills/sap/sap-data-privacy-analytics-ai-protocol/agents/openai.yaml

@@ -0,0 +1,11 @@
+interface:
+  display_name: "SAP Data Privacy Analytics AI Protocol"
+  short_description: "Cross-functional coordination protocol for data privacy, analytics, and AI governance in SAP landscapes. Covers PII in Datasphere data products, SAC exports, RAG pipelines, embeddings, prompt logging, Generative AI Hub, Joule adoption, and model outputs that could leak sensitive data. Defines classification gates, consent requirements, approval flows, and audit packaging. Never mutates live systems."
+  default_prompt: "Use $sap-data-privacy-analytics-ai-protocol to coordinate a cross-functional data privacy and AI governance review. Classify the trigger condition, identify participating agents, assess PII and sensitive data presence, confirm legal basis status, map approval gates, check the irreversible-action gate, and state the next action with the responsible owner named. Flag any PII in production AI logs without legal basis, sensitive data in a RAG index without approval, or cross-border transfer without a transfer mechanism as an immediate escalation before any other recommendation."
+dependencies:
+  tools:
+    - type: "web"
+      value: "https://help.sap.com/docs/sap-ai-core/sap-ai-core/generative-ai-hub-in-sap-ai-core"
+      description: "SAP Generative AI Hub documentation — primary reference for AI Core and Generative AI Hub governance review"
+policy:
+  allow_implicit_invocation: true

package/skills/sap/sap-data-privacy-analytics-ai-protocol/metadata.json

@@ -0,0 +1,34 @@
+{
+  "id": "sap-data-privacy-analytics-ai-protocol",
+  "name": "SAP Data Privacy Analytics AI Protocol",
+  "type": "skill",
+  "provider": "sap",
+  "harnesses": [
+    "claude-code",
+    "codex",
+    "cursor",
+    "gemini",
+    "kiro",
+    "other"
+  ],
+  "summary": "Cross-functional coordination protocol for data privacy, analytics, and AI governance in SAP landscapes. Covers PII in Datasphere data products, SAC analytics exports, RAG pipeline data ingestion, embedding generation, prompt logging, Generative AI Hub usage, Joule adoption, and model outputs that could leak sensitive data. Defines classification gates, consent requirements, approval flows, and audit packaging. Never mutates live systems.",
+  "source_type": "original",
+  "category": "compliance",
+  "official_docs": [
+    "https://help.sap.com/docs/datasphere/datasphere/introduction-to-data-governance",
+    "https://help.sap.com/docs/datasphere/datasphere/data-access-control",
+    "https://help.sap.com/docs/sap-analytics-cloud/sap-analytics-cloud/security-guide",
+    "https://help.sap.com/docs/sap-ai-core/sap-ai-core/what-is-sap-ai-core",
+    "https://help.sap.com/docs/sap-ai-core/sap-ai-core/generative-ai-hub-in-sap-ai-core",
+    "https://help.sap.com/docs/joule/joule/what-is-joule",
+    "https://gdpr-info.eu/art-5-gdpr/",
+    "https://airc.nist.gov/RMF_Overview",
+    "https://www.iso.org/standard/81230.html"
+  ],
+  "security_notes": "Does not access live Datasphere spaces, SAP Analytics Cloud tenants, AI Core deployments, Generative AI Hub configurations, or Joule instances. Accepts only user-provided data flow diagrams, data classification outputs, system prompt excerpts (redacted), prompt log samples, and deployment configuration descriptions. Never accepts or stores unredacted PII, compensation data, M&A data, or confidential system prompt content. All irreversible actions (RAG indexing, Joule activation for HR, Datasphere data product publishing, Generative AI Hub deployment with sensitive grounding) require documented Data Privacy Officer approval before execution. No guarded-mutating gate is bypassed by this protocol.",
+  "last_verified": "2026-06-19",
+  "path": "skills/sap/sap-data-privacy-analytics-ai-protocol",
+  "author": "github: Raishin",
+  "version": "0.1.0",
+  "lifecycle": "experimental"
+}

package/skills/sap/sap-data-privacy-analytics-ai-protocol/references/official-sources.md

@@ -0,0 +1,93 @@
+# Official sources — SAP Data Privacy Analytics AI Protocol
+
+Use this reference when grounding data classification assessments, Datasphere data product governance, SAC analytics export controls, SAP AI Core and Generative AI Hub configuration review, Joule adoption governance, and regulatory compliance framing for cross-functional data privacy and AI governance escalations.
+
+**Evidence level**: documentation-based (SAP Help Portal, GDPR text, NIST AI RMF, ISO/IEC 42001). No live-system evidence is collected by this skill.
+
+## SAP Datasphere
+
+- Introduction to Data Governance in SAP Datasphere
+  https://help.sap.com/docs/datasphere/datasphere/introduction-to-data-governance
+  source_owner: SAP SE
+  topic_supported: Datasphere governance model, data ownership, data product lifecycle, semantic layer
+  why_needed: Primary reference for Datasphere data governance model — defines the data product, space, and ownership concepts used throughout this protocol
+  evidence_level: primary
+  last_verified: 2026-06-19
+
+- Data Access Control in SAP Datasphere
+  https://help.sap.com/docs/datasphere/datasphere/data-access-control
+  source_owner: SAP SE
+  topic_supported: Row-level and column-level access control, data access control objects, assignment to analytic models
+  why_needed: Required to assess whether PII-containing Datasphere data products have adequate access controls before they are approved for AI or analytics pipelines
+  evidence_level: primary
+  last_verified: 2026-06-19
+
+## SAP Analytics Cloud
+
+- SAP Analytics Cloud Security Guide
+  https://help.sap.com/docs/sap-analytics-cloud/sap-analytics-cloud/security-guide
+  source_owner: SAP SE
+  topic_supported: SAC tenant security, model-level access, story sharing and export controls, data connection trust settings
+  why_needed: Primary reference for SAC data access controls and export governance — required to assess analytics export compliance and story-level access restrictions
+  evidence_level: primary
+  last_verified: 2026-06-19
+
+## SAP AI Core and Generative AI Hub
+
+- What is SAP AI Core
+  https://help.sap.com/docs/sap-ai-core/sap-ai-core/what-is-sap-ai-core
+  source_owner: SAP SE
+  topic_supported: AI Core deployment model, resource groups, AI scenarios, pipeline execution
+  why_needed: Primary reference for AI Core deployment governance — required to assess deployment configuration, resource group isolation, and logging behavior
+  evidence_level: primary
+  last_verified: 2026-06-19
+
+- Generative AI Hub in SAP AI Core
+  https://help.sap.com/docs/sap-ai-core/sap-ai-core/generative-ai-hub-in-sap-ai-core
+  source_owner: SAP SE
+  topic_supported: Generative AI Hub model catalog, orchestration, prompt registry, grounding configuration, token usage, logging
+  why_needed: Primary reference for Generative AI Hub governance review — defines the model deployment, orchestration flow, grounding data, and logging capabilities assessed by this protocol
+  evidence_level: primary
+  last_verified: 2026-06-19
+
+## SAP Joule
+
+- What is Joule
+  https://help.sap.com/docs/joule/joule/what-is-joule
+  source_owner: SAP SE
+  topic_supported: Joule AI assistant capabilities, SAP system integration, skill activation model, user context handling
+  why_needed: Primary reference for Joule governance — defines the skill activation model and system integration scope used to assess Joule adoption readiness and grounding data risk
+  evidence_level: primary
+  last_verified: 2026-06-19
+
+## Data Protection Regulation
+
+- GDPR Article 5 — Principles Relating to Processing of Personal Data
+  https://gdpr-info.eu/art-5-gdpr/
+  source_owner: European Parliament and Council
+  topic_supported: Lawfulness, fairness, transparency; purpose limitation; data minimization; accuracy; storage limitation; integrity and confidentiality; accountability
+  why_needed: Defines the data protection principles that govern PII processing in analytics and AI workloads under GDPR — particularly purpose limitation and data minimization, which are directly applied in the consent and purpose documentation requirements of this protocol
+  evidence_level: primary (regulatory)
+  last_verified: 2026-06-19
+
+## AI Risk Frameworks
+
+- NIST AI Risk Management Framework (AI RMF)
+  https://airc.nist.gov/RMF_Overview
+  source_owner: NIST
+  topic_supported: AI risk identification, assessment, and management; GOVERN, MAP, MEASURE, MANAGE functions; AI trustworthiness properties including privacy, security, and accountability
+  why_needed: Secondary framework for AI governance escalation framing — NIST AI RMF GOVERN and MAP functions directly support the approval gate and audit package requirements for Generative AI Hub and Joule deployments
+  evidence_level: secondary
+  last_verified: 2026-06-19
+
+- ISO/IEC 42001 — Artificial Intelligence Management System
+  https://www.iso.org/standard/81230.html
+  source_owner: ISO/IEC
+  topic_supported: AI management system requirements, risk assessment for AI systems, transparency and explainability obligations, impact assessment
+  why_needed: Secondary framework providing ISO/IEC 42001 context for AI system governance obligations cited in this protocol's approval requirements and audit package structure
+  evidence_level: secondary
+  last_verified: 2026-06-19
+
+## Grounding rule
+
+SAP Help Portal documentation describes the designed capabilities, data models, and governance features of Datasphere, SAP Analytics Cloud, SAP AI Core, Generative AI Hub, and Joule. It does not prove which data products contain PII in the user's landscape, what grounding data is configured in the user's Generative AI Hub orchestration, or whether prompt logging is retaining personal data. Users must supply data flow diagrams, deployment configuration descriptions, system prompt excerpts (redacted), and prompt log samples for concrete protocol execution.

package/skills/sap/sap-data-privacy-analytics-ai-protocol/references/safety-checklist.md

@@ -0,0 +1,39 @@
+# Safety checklist — SAP Data Privacy Analytics AI Protocol
+
+Use before any cross-function handoff, before any approval gate is cleared, and before any evidence is shared outside its originating function.
+
+## Non-negotiables
+
+- Do not approve or recommend approval of any RAG pipeline ingestion, embedding generation, or AI deployment involving PII without confirmed legal basis and Data Privacy Officer sign-off.
+- Do not share unredacted system prompt content containing confidential business data (M&A, pricing, compensation) with operational teams or outside the AI governance and Legal functions.
+- Do not confirm an AI deployment is "privacy compliant" without a completed data flow assessment, legal basis confirmation, and Data Privacy Officer written approval.
+- Do not bypass the irreversible-action gate. Any action listed under that section of SKILL.md requires all listed written approvals before execution.
+- Do not accept verbal approval for any cross-function data governance action. Written approval is mandatory.
+- Do not advise that prompt logs containing PII may be retained indefinitely. All prompt log retention must be bounded by an approved retention period consistent with the processing purpose and applicable data protection law.
+- Do not recommend a cross-border data transfer for AI inference or training without a completed transfer impact assessment where GDPR or equivalent regulation applies.
+- Do not share raw vector embedding data outside the AI engineering function. The governance review covers source data classification and vector store access controls, not the vectors themselves.
+
+## What people get wrong
+
+- **Treating data minimization as optional for AI grounding**: Data minimization is a legal requirement under GDPR and equivalent laws, not an optimization. Grounding a RAG pipeline with a full Datasphere data product that includes PII when only non-personal business attributes are needed is a legal compliance gap, not just a technical choice.
+- **Assuming Generative AI Hub model providers do not retain prompt data**: Model providers may retain prompt and completion data for safety or improvement purposes per their terms. The governance review must confirm the data processing terms for each model in use, especially for models accessed via the Generative AI Hub that are hosted by third parties.
+- **Conflating a SAC export approval with a data privacy assessment**: A business user approving an analytics export does not constitute a data privacy review. Export of PII or sensitive business data to an external destination requires a separate Data Privacy Officer review and, where applicable, a transfer impact assessment.
+- **Forgetting that Joule injects user context from connected SAP systems**: Joule retrieves context from connected SAP systems (HR, Finance, Procurement) to generate responses. If Joule is activated for a business function that has access to HR compensation or performance data, the grounding data scope includes that data — even if the intent was limited to a different use case.
+- **Using analytics exports as a proxy for AI training data without review**: A SAC story or Datasphere export that is acceptable for internal business analytics may not be acceptable as AI training data without additional de-identification, legal basis review, and data owner approval.
+- **Assuming AI output leakage only occurs through direct retrieval**: A generative AI model can reconstruct or infer sensitive data from patterns in its training data or retrieval context even when the original data is not directly quoted. Model output leakage risk assessment must consider indirect reconstruction, not only verbatim retrieval.
+
+## When to push back
+
+- Push back (and escalate immediately) when PII is discovered in production AI prompt logs without a confirmed legal basis for processing — escalate to the Data Privacy Officer before any other action.
+- Push back when asked to approve RAG indexing or embedding generation involving PII without Data Privacy Officer written approval.
+- Push back when asked to confirm that a system prompt containing confidential business data is safe without reviewing the prompt content and AI deployment configuration.
+- Push back when a cross-border data transfer for AI inference or training is proposed without a transfer impact assessment.
+- Push back when asked to clear an irreversible-action gate without all required written approvals documented.
+- Push back when asked to confirm legal basis from memory — require Data Privacy Officer confirmation grounded in a records-of-processing-activities entry or equivalent documentation.
+- Push back when Joule is proposed for activation on an HR or payroll business function without HR lead and Data Privacy Officer review of the grounding data scope.
+
+## Evidence labels
+
+- `documentation-based` — grounded in SAP Datasphere, SAP Analytics Cloud, SAP AI Core, SAP Generative AI Hub, SAP Joule, GDPR, NIST AI RMF, or ISO/IEC 42001 documentation
+- `user-provided evidence` — data flow diagrams, data classification outputs, system prompt excerpts (redacted), prompt log samples, deployment configuration descriptions, or legal basis documentation provided by the requesting function
+- `inference` — derived reasoning not directly confirmed by official documentation or user-provided evidence; must always be labeled as such and must not be used as the sole basis for any approval gate clearance

package/skills/sap/sap-data-privacy-analytics-ai-protocol/references/workflow-and-output.md

@@ -0,0 +1,78 @@
+# Workflow and output contract — SAP Data Privacy Analytics AI Protocol
+
+Use this reference for all trigger classification, data assessment sequencing, cross-function handoff coordination, decision rights application, and output formatting.
+
+## Trigger classification table
+
+| Trigger class | Primary signal | Activating agent(s) | Data Privacy involvement | Legal involvement |
+|---|---|---|---|---|
+| `pii-in-analytics-pipeline` | PII columns in Datasphere data product or SAC story | sap-datasphere-data-product-architect-agent + sap-analytics-cloud-planning-governance-agent | Required (legal basis and minimization review) | Conditional (cross-border transfer or breach threshold) |
+| `sensitive-business-data-in-export` | Confidential metrics in analytics export to external destination | sap-analytics-cloud-planning-governance-agent | Required (purpose limitation review) | Required (if third-party or cross-border) |
+| `rag-pipeline-ingestion` | PII or sensitive data being indexed for RAG | sap-datasphere-data-product-architect-agent + sap-ai-core-genai-hub-governance-reviewer-agent | Required (pre-ingestion approval gate) | Conditional (cross-border or special category data) |
+| `embedding-generation` | SAP data converted to embeddings for AI use | sap-ai-core-genai-hub-governance-reviewer-agent | Required (source data classification review) | Conditional (if source includes HR or financial data requiring special treatment) |
+| `prompt-logging-with-pii` | AI system logs capturing user context with PII | sap-ai-core-genai-hub-governance-reviewer-agent | Required (retention and minimization review) | Required if logs retained beyond policy period |
+| `genai-hub-deployment-review` | New or expanding Generative AI Hub deployment | sap-ai-core-genai-hub-governance-reviewer-agent | Required (system prompt and grounding data review) | Conditional (if sensitive business data in system prompt) |
+| `joule-activation-governance` | Joule activation for new business function | sap-joule-governance-adoption-agent | Required (grounding data scope and PII handling review) | Conditional (if HR or payroll data in scope) |
+| `model-output-leakage` | Generative AI output reconstructs or reveals sensitive data | sap-ai-core-genai-hub-governance-reviewer-agent | Required (assess source data and retrieval path) | Required (potential data breach determination) |
+
+## Protocol workflow
+
+### Phase 1 — Triage (AI Governance or Data Privacy lead)
+
+1. Classify the trigger condition using the table above.
+2. Identify which participating agents are activated.
+3. Request the required evidence inventory from the responsible function.
+4. Apply the redaction policy before any cross-function evidence share.
+
+### Phase 2 — Data classification and legal basis assessment
+
+1. `sap-datasphere-data-product-architect-agent` classifies data elements in Datasphere data products and confirms data access control configuration.
+2. `sap-analytics-cloud-planning-governance-agent` assesses SAC model and story data classification and export control settings.
+3. Data Privacy Officer confirms legal basis for each PII processing activity in scope and assesses purpose limitation compliance.
+4. If cross-border data transfer is involved, Legal is notified immediately for transfer impact assessment.
+
+### Phase 3 — AI pipeline and deployment review
+
+1. `sap-ai-core-genai-hub-governance-reviewer-agent` reviews AI Core deployment configuration, Generative AI Hub orchestration, system prompt content, grounding data scope, and logging configuration.
+2. `sap-joule-governance-adoption-agent` reviews Joule grounding data scope, user context injection design, and output handling for the business function in scope.
+3. Logging and retention configuration is compared against approved retention period and data minimization requirements.
+
+### Phase 4 — Approval gate evaluation
+
+1. Each required approval is identified from the decision rights table.
+2. Approval documentation is confirmed present or flagged as outstanding.
+3. Irreversible-action gate is evaluated for each proposed action — no action proceeds without confirmed written approval.
+
+### Phase 5 — Audit package assembly
+
+1. All evidence, data classification results, legal basis confirmations, approval records, redaction logs, and action outcomes are consolidated into the audit package.
+2. Residual risks are assessed and documented with compensating controls.
+
+## Data classification labels
+
+Use these labels consistently across all evidence and output:
+
+| Label | Meaning |
+|---|---|
+| `PII-direct` | Data element that directly identifies a natural person (name, email, employee ID, national ID) |
+| `PII-indirect` | Data element that, combined with other available data, could identify a natural person |
+| `special-category` | Personal data warranting heightened protection under GDPR Art 9 or equivalent (health, ethnic origin, union membership, biometric, etc.) |
+| `sensitive-business` | Confidential business data (revenue forecasts, M&A, margin, compensation benchmarks, pricing strategy) |
+| `restricted` | Data accessible only to named roles under the organization's data classification policy |
+| `internal` | Data for internal use only; not approved for external sharing or AI grounding without review |
+| `public` | Data approved for unrestricted use |
+
+## Output contract
+
+Return, in order:
+
+1. **Trigger classification**: Which trigger class(es) apply; which participating agents are activated.
+2. **Data inventory and classification**: Data sources assessed, classification label per element, and any PII-direct, PII-indirect, special-category, or sensitive-business elements identified.
+3. **Legal basis status**: Legal basis confirmed, pending, or absent for each PII processing activity. If absent, this is a blocking condition — escalate immediately.
+4. **Redaction confirmation**: Explicit statement that sensitive elements have been identified and the redaction policy has been applied.
+5. **AI pipeline assessment** (if applicable): Grounding data scope, logging configuration compliance, system prompt classification, embedding source classification, and retrieval path risk.
+6. **Approval gate status**: For each required approval, primary authority, documentation status (present / outstanding), and gate status (cleared / blocked).
+7. **Irreversible-action gate**: Whether any irreversible actions are pending, approval status, and whether the gate is cleared or blocked.
+8. **Escalation notice**: If PII is present in production AI logs without legal basis, sensitive business data is in a RAG index without data owner approval, or a cross-border transfer is occurring without a transfer mechanism — escalation notice must appear before any other recommendation.
+9. **Audit package status**: Populated items and outstanding items.
+10. **Next step**: Single next action with named responsible owner.

package/skills/sap/sap-datasphere-data-product-architecture/SKILL.md

@@ -0,0 +1,86 @@
+---
+name: sap-datasphere-data-product-architecture
+description: Review SAP Datasphere architecture: spaces, data flows and replication, semantic and analytic models, data products and sharing, data access controls, and integration with SAP Analytics Cloud and SAP HANA Cloud. Flags structural design gaps, access control weaknesses, and data product sharing anti-patterns. Does not touch live systems.
+allowed-tools: Read Grep Glob WebSearch WebFetch
+metadata:
+  author: "github: Raishin"
+  version: "0.1.0"
+  updated: "2026-06-19"
+  category: data
+  lifecycle: experimental
+---
+
+# SAP Datasphere Data Product Architecture Review
+
+## Purpose
+
+Assess the architecture and governance posture of an SAP Datasphere implementation. Review space design for isolation and purpose separation, data flow and replication flow design for completeness and error handling, semantic layer modeling (views, analytical datasets, dimensions, facts) for correctness and reuse, data product definition and cross-space or external sharing configuration, data access control (DAC) and row-level security design, and integration with SAP Analytics Cloud (SAC) and SAP HANA Cloud. Surface structural design gaps, over-permissive sharing configurations, missing access controls, and modeling anti-patterns. Does not connect to or mutate any live Datasphere tenant.
+
+## When to use
+
+Use this skill when the user asks to:
+
+- review the Datasphere space structure for isolation, purpose separation, naming conventions, and governance boundary alignment,
+- assess data flow and replication flow design for error handling coverage, delta-load correctness, scheduling, and target table configuration,
+- evaluate semantic models (graphical views, SQL views, analytical datasets, dimensions, fact views, hierarchies) for correctness, reuse, and star schema completeness,
+- review data product definitions, output ports, and cross-space or external sharing configurations for governance and access appropriateness,
+- audit data access control (DAC) rules and row-level security configurations for correctness and completeness,
+- assess integration patterns between Datasphere and SAP Analytics Cloud: live connections vs. import connections, model type compatibility, and refresh scheduling,
+- review Datasphere integration with SAP HANA Cloud for virtual access, remote table replication, and federation correctness,
+- flag data architecture anti-patterns such as redundant replicated tables, missing semantic annotations, overly broad space access, or unmanaged data product output ports.
+
+## When not to use
+
+- When the user needs live inspection of a running Datasphere tenant, space, or data flow execution log — this skill accepts only user-provided descriptions, space exports, model exports, or architecture documents.
+- When the request is about SAP Analytics Cloud stories, planning models, or SAC-specific governance — use `sap-analytics-cloud-planning-governance`.
+- When the request is specifically about HANA Cloud performance tuning, cost management, or sizing — use `sap-hana-cloud-performance-cost`.
+- When the request concerns BTP-level governance (entitlements, subaccounts, role collections) — use `sap-btp-governance-review`.
+
+## Does not touch live systems
+
+This skill operates on user-provided descriptions, architecture diagrams, space configuration exports, data flow screenshots, semantic model exports, data product definitions, or written descriptions of the Datasphere implementation. It does not connect to any Datasphere tenant, invoke Datasphere APIs, access space files, read data flow run logs, or interact with any connected source system. All live inspection is out of scope.
+
+## Lean operating rules
+
+- Classify findings before recommending. Every finding must be assigned to a design domain (space design, data flows, semantic models, data products, data access controls, SAC integration, HANA Cloud integration) before remediation is proposed.
+- Spaces are the primary isolation boundary. Each Datasphere space has its own connections, storage quota, users, and data sharing scope. Cross-space data access must go through explicitly defined data product output ports — direct cross-space table access is an architecture violation.
+- Semantic layer completeness is required for trusted analytics. Views exposed to SAC must have correct semantic usage set (dimension, fact, analytical dataset, or cube). Views without a semantic usage annotation cannot be used reliably in SAC Live Connection models.
+- Row-level data access control (DAC) is not optional for multi-tenant or shared data. Any Datasphere space that contains data from multiple business units or customers without DAC enforcement is a data isolation risk.
+- Data products require defined output ports. A data product definition without a configured output port (table sharing, data sharing service, or Open SQL schema) is incomplete and cannot be consumed by other spaces or external consumers.
+- Replication flows and data flows handle error differently. Replication flows are designed for bulk and delta replication with automatic restart; data flows are designed for transformation pipelines with explicit scheduling. Confusing the two leads to incorrect delta-load behavior and missing transformation error handling.
+- SAC live connections require stable semantic models. Changing the key structure or removing measures from a Datasphere view consumed via SAC live connection breaks the SAC story without warning. Live connection model dependencies must be tracked before structural model changes.
+- Data access controls must be tested against real user assignments. DAC configurations that look correct in isolation may fail silently if the DAC combination method (AND/OR) is misconfigured for multi-rule scenarios.
+- Evidence from user-provided artifacts or official SAP Datasphere documentation takes precedence over inference.
+- Load only the reference needed for the component under review.
+
+## Evidence rules
+
+Label all claims with one of:
+
+- `documentation-based` — grounded in SAP Datasphere Help Portal documentation, SAP Help Portal data product guides, or official SAP Datasphere modeling docs
+- `user-provided evidence` — space configuration exports, data flow screenshots, semantic model exports, data product definitions, or architecture documents provided by the user
+- `inference` — derived reasoning not directly confirmed by official docs or user evidence
+
+## Live-environment rules
+
+**This skill does not touch live systems.** There is no Datasphere API call, tenant login, space file access, data flow execution, or connection to any source or target system in this skill's execution path. Users must supply space configuration exports, model exports, data product definitions, architecture documents, or written descriptions of their Datasphere implementation for this skill to review.
+
+## References
+
+Load only when needed:
+
+- [Workflow and output contract](references/workflow-and-output.md) — design domain taxonomy, finding severity, output format.
+- [Safety checklist](references/safety-checklist.md) — non-negotiables, common mistakes, when to push back.
+- [Official sources](references/official-sources.md) — SAP Datasphere space design, data flows, semantic modeling, data products, data access controls, SAC and HANA Cloud integration docs.
+
+## Response minimum
+
+Return, at minimum:
+
+- **Problem classification**: design domain(s) affected (space design / data flows / semantic models / data products / data access controls / SAC integration / HANA Cloud integration) and specific finding(s).
+- **Evidence used**: documentation-based / user-provided evidence / inference.
+- **Risk level**: critical (data isolation breach or access control failure) / high (data architecture failure or pipeline reliability risk) / medium (governance gap or modeling quality gap) / low (best practice deviation).
+- **Recommended action**: specific remediation per finding (space restructure, DAC rule addition, output port configuration, semantic usage annotation, model redesign, or connection type change).
+- **Refusal / escalation triggers**: if live Datasphere tenant access, space file inspection, or data flow execution log access is required to complete the review, state that clearly and do not proceed.
+- **Business impact**: data isolation risk, analytics quality risk, pipeline reliability risk, cost overrun, or compliance gap.
+- **Next verification step**: validate recommended changes against the current Datasphere tenant configuration before applying.

package/skills/sap/sap-datasphere-data-product-architecture/agents/openai.yaml

@@ -0,0 +1,11 @@
+interface:
+  display_name: "SAP Datasphere Data Product Architecture Review"
+  short_description: "Review SAP Datasphere architecture: spaces, data flows, replication flows, semantic and analytic models, data products and output ports, data access controls, and SAC/HANA Cloud integration. Flags design gaps and access control weaknesses."
+  default_prompt: "Use $sap-datasphere-data-product-architecture to assess the provided SAP Datasphere implementation. Classify each finding by design domain (space design, data flows, semantic models, data products, data access controls, SAC integration, HANA Cloud integration), assign a risk level (critical/high/medium/low), and recommend specific remediation actions. State evidence level (documentation-based, user-provided evidence, inference) and whether live tenant access is required to complete the review."
+dependencies:
+  tools:
+    - type: "web"
+      value: "https://help.sap.com/docs/SAP_DATASPHERE/c8a54ee704e94e15926551293243fd1d/20828efb80714786b2cf91e39b9706dd.html"
+      description: "SAP Datasphere spaces documentation — primary reference for space design, isolation boundaries, and data sharing governance"
+policy:
+  allow_implicit_invocation: true

package/skills/sap/sap-datasphere-data-product-architecture/metadata.json

@@ -0,0 +1,33 @@
+{
+  "id": "sap-datasphere-data-product-architecture",
+  "name": "SAP Datasphere Data Product Architecture Review",
+  "type": "skill",
+  "provider": "sap",
+  "harnesses": [
+    "claude-code",
+    "codex",
+    "cursor",
+    "gemini",
+    "kiro",
+    "other"
+  ],
+  "summary": "Review SAP Datasphere architecture: spaces, data flows, replication flows, semantic and analytic models, data products and output ports, data access controls, and integration with SAP Analytics Cloud and SAP HANA Cloud. Flags design gaps, access control weaknesses, and data product sharing anti-patterns. Does not access live systems.",
+  "source_type": "original",
+  "category": "data",
+  "official_docs": [
+    "https://help.sap.com/docs/sap-datasphere/sap-datasphere/spaces",
+    "https://help.sap.com/docs/sap-datasphere/sap-datasphere/data-flows",
+    "https://help.sap.com/docs/sap-datasphere/sap-datasphere/replication-flows",
+    "https://help.sap.com/docs/sap-datasphere/sap-datasphere/semantic-modeling",
+    "https://help.sap.com/docs/sap-datasphere/sap-datasphere/data-products",
+    "https://help.sap.com/docs/sap-datasphere/sap-datasphere/cross-space-sharing",
+    "https://help.sap.com/docs/sap-datasphere/sap-datasphere/data-access-controls",
+    "https://help.sap.com/docs/sap-datasphere/sap-datasphere/what-is-sap-datasphere"
+  ],
+  "security_notes": "Does not access live SAP Datasphere tenants, space APIs, data flow execution APIs, or connected source systems. Accepts only user-provided space configuration exports, model exports, data product definitions, architecture documents, or written descriptions. Never request or accept Datasphere tenant credentials, space user assignments, OAuth tokens, or connection credentials. Governance recommendations must be validated against the live Datasphere tenant configuration before applying.",
+  "last_verified": "2026-06-19",
+  "path": "skills/sap/sap-datasphere-data-product-architecture",
+  "author": "github: Raishin",
+  "version": "0.1.0",
+  "lifecycle": "experimental"
+}