@raishin/vanguard-frontier-agentic 2.11.0 → 2.13.0

package/skills/sap/sap-live-readonly-landscape-discovery/references/live-environment-access.md

@@ -0,0 +1,111 @@
+# Live environment access — SAP Live Read-Only Landscape Discovery
+
+This reference defines credential setup, role requirements, audit log format, and redaction rules for live read-only landscape discovery sessions.
+
+## Credential setup
+
+### BTP CLI
+
+Required role: **Subaccount Viewer** (for subaccount-scope discovery) or **Global Account Viewer** (for cross-subaccount enumeration).
+
+Never use: Global Account Administrator, Subaccount Administrator, or any role with entitlement assignment or service instance management rights.
+
+Setup:
+```bash
+# Login to BTP CLI (non-interactive with service key or browser flow)
+btp login --url https://cli.btp.cloud.sap
+
+# Verify current active context
+btp --info
+
+# Confirm role assignment before discovery
+btp get security/role-collection <ROLE_COLLECTION_NAME> --subaccount <SUBACCOUNT_ID>
+```
+
+### Cloud Foundry CLI
+
+Required role: **OrgAuditor** (org-scope) or **SpaceAuditor** (space-scope).
+
+Never use: OrgManager, SpaceManager, SpaceDeveloper — these carry write permissions.
+
+```bash
+# Login (use SSO where possible to avoid password in shell history)
+cf login --sso
+
+# Confirm current user and role
+cf org-users <ORG_NAME>
+cf space-users <ORG_NAME> <SPACE_NAME>
+```
+
+### kubectl / Kyma
+
+Required: A ClusterRoleBinding or RoleBinding that grants only `get`, `list`, `watch` verbs. No `create`, `update`, `delete`, `patch`.
+
+```bash
+# Verify current permissions before discovery
+kubectl auth can-i --list --namespace <NAMESPACE>
+```
+
+Confirm the output contains only read verbs before proceeding.
+
+### ABAP RFC/HTTP display user
+
+Required: A user with display-only authorization profiles. Recommended authorization objects:
+- `S_TCODE`: display transactions only (SE10, SE16N display, STMS_IMPORT display)
+- `S_DEVELOP`: display only (`ACTVT = 03`), no create/change/delete
+- `S_TRANSPRT`: display only (`ACTVT = 03`)
+- No `S_DATASET` write, no `S_PROGRAM` execute
+
+Never use an ABAP user with BASIS administrator profile or emergency user (`SAP_ALL`).
+
+## Redaction rules
+
+The following must be redacted from all output, logs, and evidence before returning to the user:
+
+| Data type | Redaction token |
+|-----------|----------------|
+| OAuth client_secret | `[REDACTED:CLIENT_SECRET]` |
+| Service key credentials block | `[REDACTED:SERVICE_KEY]` |
+| VCAP_SERVICES credential values | `[REDACTED:VCAP_CREDENTIALS]` |
+| ABAP logon password | `[REDACTED:ABAP_PASSWORD]` |
+| API key or token value | `[REDACTED:API_KEY]` |
+| BTP platform user password | `[REDACTED:BTP_PASSWORD]` |
+
+The data type name and structure (key names, service names) may remain in output. Only the value is redacted.
+
+## Audit log format
+
+Every command executed must be logged in the following format before the command is run:
+
+```
+[COMMAND_LOG]
+Timestamp (UTC): <ISO 8601>
+System: <BTP global account ID or CF org/space or ABAP SID>
+Tool: <btp | cf | kubectl | abap-http>
+Command: <exact command with arguments, credentials omitted>
+Output summary: <one-line summary of what was returned>
+Redactions applied: <yes/no — what was redacted>
+```
+
+Example:
+```
+[COMMAND_LOG]
+Timestamp (UTC): 2026-06-19T10:32:00Z
+System: BTP Global Account abc-123
+Tool: btp
+Command: btp list accounts/subaccount --global-account abc-123
+Output summary: 4 subaccounts enumerated (dev, test, staging, prod)
+Redactions applied: no
+```
+
+## Approval gate checklist
+
+Before the first live command in any session:
+
+- [ ] User has confirmed live read-only access is authorized for this session
+- [ ] Credential scope confirmed as viewer/auditor (role name or profile confirmed)
+- [ ] Target system and scope explicitly identified by user (global account ID / ABAP SID / CF org)
+- [ ] Redaction rules reviewed and applied to output
+- [ ] User understands no state changes will be made
+
+Do not proceed until all five items are confirmed.

package/skills/sap/sap-live-readonly-landscape-discovery/references/official-sources.md

@@ -0,0 +1,83 @@
+# Official sources — SAP Live Read-Only Landscape Discovery
+
+Use this reference when grounding BTP CLI commands, CF CLI commands, destination service access, and ABAP landscape object enumeration.
+
+**Evidence level**: documentation-based (SAP Help Portal). Live evidence gathered during sessions is labeled separately as `live evidence` per the audit format in `references/live-environment-access.md`.
+
+## SAP BTP CLI
+
+- BTP CLI command reference
+  https://help.sap.com/docs/btp/btp-cli-command-reference/btp-cli-command-reference
+  source_owner: SAP SE
+  topic_supported: Full BTP CLI command catalog — used to verify which commands are list/get/describe (allowed) vs create/update/delete (forbidden)
+  why_needed: Authoritative source for BTP CLI read-only command identification
+  evidence_level: primary
+  last_verified: 2026-06-19
+
+- BTP account administration using BTP CLI
+  https://help.sap.com/docs/btp/sap-business-technology-platform/account-administration-using-sap-btp-command-line-interface-btp-cli
+  source_owner: SAP SE
+  topic_supported: BTP CLI setup, login, context management, subaccount navigation
+  why_needed: Required to establish read-only CLI session and navigate subaccount hierarchy
+  evidence_level: primary
+  last_verified: 2026-06-19
+
+- Managing entitlements and quotas using BTP CLI
+  https://help.sap.com/docs/btp/sap-business-technology-platform/managing-entitlements-and-quotas-using-the-btp-cli
+  source_owner: SAP SE
+  topic_supported: Entitlement and quota read operations via BTP CLI
+  why_needed: Defines read-only entitlement enumeration commands (btp list accounts/entitlements)
+  evidence_level: primary
+  last_verified: 2026-06-19
+
+## SAP Cloud Foundry CLI
+
+- CF CLI command reference for SAP BTP
+  https://help.sap.com/docs/cloud-foundry/sap-cloud-foundry-command-line-interface/cf-command-reference
+  source_owner: SAP SE
+  topic_supported: CF CLI read-only commands (cf apps, cf services, cf service-instances, cf env) and role requirements
+  why_needed: Authoritative source for CF read-only command identification and SpaceAuditor/OrgAuditor role mapping
+  evidence_level: primary
+  last_verified: 2026-06-19
+
+## Destinations
+
+- Destinations on SAP BTP
+  https://help.sap.com/docs/btp/sap-business-technology-platform/destinations
+  source_owner: SAP SE
+  topic_supported: Destination service overview, destination types (HTTP, RFC, Mail), destination properties
+  why_needed: Context for read-only destination enumeration and interpreting destination configuration output
+  evidence_level: primary
+  last_verified: 2026-06-19
+
+- Destination Service REST API
+  https://help.sap.com/docs/connectivity/sap-btp-connectivity-cf/destination-service-rest-api
+  source_owner: SAP SE
+  topic_supported: REST API for reading destination configurations — GET operations only for this skill
+  why_needed: Defines allowed GET endpoints for destination enumeration when using REST API access
+  evidence_level: primary
+  last_verified: 2026-06-19
+
+## Authorization and Trust Management
+
+- SAP Authorization and Trust Management Service (XSUAA)
+  https://help.sap.com/docs/btp/sap-business-technology-platform/sap-authorization-and-trust-management-service-in-the-cloud-foundry-environment
+  source_owner: SAP SE
+  topic_supported: Role collections, trust configurations, identity provider setup — read-only enumeration context
+  why_needed: Required for read-only enumeration of trust configurations and role collection assignments
+  evidence_level: primary
+  last_verified: 2026-06-19
+
+## ABAP System Landscape
+
+- S/4HANA Cloud landscape and system administration
+  https://help.sap.com/docs/SAP_S4HANA_CLOUD/e5522a8a7b174979/3e56cde2e1044df5b5f3a5ecc8e59d27.html
+  source_owner: SAP SE
+  topic_supported: ABAP system landscape overview, transport routes, system definitions — read-only context
+  why_needed: Defines ABAP landscape object types this skill can enumerate in read-only mode
+  evidence_level: primary
+  last_verified: 2026-06-19
+
+## Grounding rule
+
+SAP Help Portal documentation describes available CLI commands and API endpoints. It does not prove which commands are available in the user's installed CLI version, which entitlements exist in their account, or whether their assigned role actually restricts write access. Users must confirm credential scope before any live command is executed.

package/skills/sap/sap-live-readonly-landscape-discovery/references/safety-checklist.md

@@ -0,0 +1,45 @@
+# Safety checklist — SAP Live Read-Only Landscape Discovery
+
+Use before every live discovery session and before every command execution.
+
+## Non-negotiables
+
+- Do not execute any command that creates, updates, deletes, deploys, assigns, rotates, imports, or triggers a state change — even if the user explicitly requests it.
+- Do not request, accept, log, echo, or include in any output: SAP system credentials, BTP service keys, OAuth client secrets, API tokens, ABAP logon passwords, or RFC user passwords.
+- Do not proceed with live commands if the user has not confirmed that the credential in scope is read-only (viewer/auditor role or equivalent).
+- Do not expand scope beyond what the user has explicitly authorized — if authorized for one subaccount, do not enumerate sibling subaccounts.
+- Do not cache, store, or transmit live evidence outside the current session context.
+- Redact all credential values from CF env output, service binding details, and destination configuration before logging.
+
+## Forbidden action verification
+
+Before executing any CLI command, verify the verb is on the allowed list:
+
+| Allowed verbs | Forbidden verbs |
+|--------------|----------------|
+| list, get, describe, export, status, show, read, view | create, update, delete, set, push, deploy, assign, bind, entitle, rotate, regenerate, import, trigger, run, execute, apply, patch, approve |
+
+If the command verb is not on the allowed list, refuse and explain which downstream skill handles that action.
+
+## What people get wrong
+
+- **Using CF SpaceDeveloper for "just reading"**: SpaceDeveloper has write access. Always use SpaceAuditor or OrgAuditor for discovery.
+- **Including service key values in output**: `cf env` output includes VCAP_SERVICES with bound service credentials. Redact all credential blocks before including in output.
+- **Treating `cf restage` or `cf restart` as read-only**: These are state-transition commands. Forbidden.
+- **Assuming `btp list` can never cause side effects**: Some CLI versions may trigger audit log entries. This is acceptable; the key constraint is no state mutation.
+- **Scope creep on ABAP RFC enumeration**: A display RFC user may have access to more tables than needed. Query only the specific landscape objects in scope.
+
+## When to push back
+
+- Push back when the user asks to combine a read step with a write step in a single command or sequence.
+- Push back when the user asks to "quickly create" a test destination or service instance "to check something."
+- Push back when credentials provided appear to have write scope (e.g., SpaceDeveloper, subaccount administrator).
+- Push back when the user asks to enumerate systems outside the agreed scope.
+- Push back when the user asks to export data to an external system or share it outside the session.
+
+## Evidence labels
+
+- `live evidence` — directly observed from a live SAP system; include command, timestamp, system identifier
+- `documentation-based` — grounded in SAP official docs; no live access
+- `user-provided evidence` — stated or supplied by the user
+- `inference` — derived reasoning; must always be labeled as such

package/skills/sap/sap-live-readonly-landscape-discovery/references/workflow-and-output.md

@@ -0,0 +1,102 @@
+# Workflow and output contract — SAP Live Read-Only Landscape Discovery
+
+Use this reference for all discovery workflows, command selection, and output formatting.
+
+## Allowed command patterns
+
+### BTP CLI (read-only)
+
+```bash
+# List global account entitlements
+btp list accounts/entitlement --global-account <ID>
+
+# List subaccounts in a directory
+btp list accounts/subaccount --directory <ID>
+
+# Get subaccount details
+btp get accounts/subaccount <SUBACCOUNT_ID>
+
+# List service instances in a subaccount
+btp list services/instance --subaccount <SUBACCOUNT_ID>
+
+# List available service plans
+btp list services/plan --subaccount <SUBACCOUNT_ID>
+
+# List destinations in a subaccount
+btp list connectivity/destination --subaccount <SUBACCOUNT_ID>
+
+# Get a specific destination
+btp get connectivity/destination <DESTINATION_NAME> --subaccount <SUBACCOUNT_ID>
+```
+
+### CF CLI (read-only)
+
+```bash
+# List orgs
+cf orgs
+
+# List spaces in org
+cf spaces
+
+# List apps (no modification)
+cf apps
+
+# List service instances
+cf service-instances
+
+# Get environment (redact credentials before logging)
+cf env <APP_NAME>
+
+# List routes
+cf routes
+```
+
+### kubectl (read-only, Kyma)
+
+```bash
+# List namespaces
+kubectl get namespaces
+
+# List service instances (SAP BTP Operator)
+kubectl get serviceinstances -n <NAMESPACE>
+
+# Describe a service instance
+kubectl describe serviceinstance <NAME> -n <NAMESPACE>
+
+# List API rules (read-only)
+kubectl get apirules -n <NAMESPACE>
+```
+
+## Discovery workflow
+
+1. **Confirm authorization gate**
+   - Confirm with the user that live read-only access is authorized for this session.
+   - Confirm credential scope is viewer/auditor role only.
+   - Confirm target system and scope (global account, subaccount, space, ABAP system).
+
+2. **Select discovery scope**
+   - Identify which landscape layer(s) are in scope: BTP account model / CF runtime / Kyma / ABAP system landscape.
+   - Select the minimum command set needed.
+
+3. **Execute read-only commands**
+   - Run only allowed command patterns above.
+   - Log every command with timestamp.
+   - Redact credential values from CF env output before logging.
+
+4. **Structure evidence**
+   - Organize output by resource type.
+   - Label all data as `live evidence` with command and timestamp.
+
+5. **Return output** per the output contract below.
+
+## Output contract
+
+Return:
+
+1. Discovery scope (what was enumerated and why)
+2. Authorization gate confirmation (role confirmed read-only: yes/no)
+3. Command log (every command, timestamp, summary of output)
+4. Structured evidence (resources enumerated, organized by type)
+5. Redaction log (what was redacted and why)
+6. Next step (how this evidence feeds into the downstream advisory: clean core review, transport planning, etc.)
+7. Escalation trigger (any forbidden action encountered or write attempt by the tool)

package/skills/sap/sap-maestro/SKILL.md

@@ -0,0 +1,81 @@
+---
+name: sap-maestro
+description: Route SAP AI workflow requests to the correct domain skill using taxonomy classification, a routing table, and dispatch mode selection. Use when an incoming request spans multiple SAP domains (BTP, S/4HANA, SuccessFactors, Ariba, etc.) and the correct downstream skill or agent must be identified before any action is taken. 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: platform
+  lifecycle: experimental
+---
+
+# SAP Maestro
+
+## Purpose
+
+Classify, route, and dispatch SAP AI workflow requests to the correct domain skill or agent without conflating guidance across unrelated SAP domains. Maestro is a pure routing layer — it reads domain taxonomy, applies a routing table, selects a dispatch mode, and passes control to the appropriate downstream skill. It does not execute business logic, mutate live systems, or make architectural recommendations on behalf of domain skills.
+
+## When to use
+
+Use this skill when the request:
+
+- spans or may span multiple SAP product domains (BTP, S/4HANA, SuccessFactors, Integration Suite, Ariba, Datasphere, GRC, Transport Management, etc.),
+- is ambiguous about which SAP skill or agent should handle it,
+- needs an explicit routing decision logged before dispatch (compliance or traceability requirement),
+- involves maestro-gated dispatch — i.e., downstream skills or agents require an upstream classification before they will accept a request.
+
+## When not to use
+
+- When the domain is already known and a specific skill (e.g., `sap-clean-core-debt-review`, `sap-live-readonly-landscape-discovery`, `sap-guarded-transport-import`) should be invoked directly.
+- When the request is purely informational and no routing or dispatch is needed.
+- When the downstream skill is already loaded and active in the current session.
+
+## Does not touch live systems
+
+This skill is a pure routing and classification layer. It does not connect to any SAP system, issue API calls, read from BTP cockpit, query ABAP systems, or invoke transport management services. All live-system interaction is delegated to downstream skills that carry their own live-environment access rules.
+
+## Lean operating rules
+
+- Classify first. Every request must receive a domain classification before any dispatch action is proposed.
+- Use the routing table verbatim. Do not improvise skill-to-domain mappings from memory; use the documented routing table in `references/workflow-and-output.md`.
+- One domain per dispatch. If the request genuinely crosses domains, split it into domain-scoped sub-requests and route each separately.
+- Surface routing rationale. Always state which taxonomy bucket was matched, which routing table entry fired, and which dispatch mode was selected.
+- Live-guard gate. If the routed skill carries a live-environment tier (`read-only-runtime` or `mutating-runtime`), call out the live-guard requirement explicitly before proposing dispatch. Do not bypass the gate.
+- Do not hallucinate skill IDs. Only route to skills and agents that are declared in the catalog. If no matching skill exists, return `unrouted` and explain why.
+- Prefer official SAP documentation for taxonomy clarification; use `references/official-sources.md` for grounding.
+
+## Evidence rules
+
+Label all claims with one of:
+
+- `documentation-based` — grounded in SAP official docs or BTP taxonomy
+- `catalog-evidence` — grounded in a declared skill/agent entry in this repo
+- `user-provided evidence` — stated by the user in this session
+- `inference` — derived reasoning not directly confirmed by docs or catalog
+
+## Live-environment rules
+
+**This skill does not touch live systems.** There is no credential, API call, shell command, or system connection of any kind in this skill's execution path. If a routing decision leads to a downstream skill that does touch live systems, those rules apply to the downstream skill — not here.
+
+## References
+
+Load only when needed:
+
+- [Workflow and output contract](references/workflow-and-output.md) — routing table, domain taxonomy, dispatch modes, output format.
+- [Safety checklist](references/safety-checklist.md) — routing safety, live-guard gate, non-negotiables before dispatch.
+- [Official sources](references/official-sources.md) — SAP BTP taxonomy, product catalog, and domain classification grounding.
+
+## Response minimum
+
+Return, at minimum:
+
+- **Problem classification**: which SAP domain(s) are in scope and why.
+- **Evidence used**: documentation-based / catalog-evidence / user-provided / inference.
+- **Routing table match**: which entry in the routing table was fired.
+- **Dispatch mode**: direct / gated / split / unrouted.
+- **Risk level**: none (pure classification) / low (advisory dispatch) / elevated (dispatch to live-tier skill).
+- **Recommended action**: load the named skill or agent, or explain why routing is unrouted.
+- **Refusal / escalation triggers**: if the request matches a live-tier skill and no live-guard approval is present, escalate before dispatch.
+- **Business impact**: what breaks if routing is wrong (wrong domain skill gives wrong guidance).
+- **Next verification step**: confirm with the user that the classification is correct before dispatching to a mutating-runtime skill.

package/skills/sap/sap-maestro/agents/openai.yaml

@@ -0,0 +1,11 @@
+interface:
+  display_name: "SAP Maestro"
+  short_description: "Route SAP AI workflow requests to the correct domain skill via taxonomy classification and a routing table"
+  default_prompt: "Use $sap-maestro to classify the incoming SAP request, match it to the routing table, select a dispatch mode, and identify the correct downstream skill or agent. State domain classification, routing table entry, dispatch mode, and live-guard gate status before any dispatch is proposed."
+dependencies:
+  tools:
+    - type: "catalog"
+      value: "sap-skill-catalog"
+      description: "Optional catalog reference for declared SAP skills and agents; do not assume it exists in every environment — fall back to documented routing table in references/workflow-and-output.md"
+policy:
+  allow_implicit_invocation: true

package/skills/sap/sap-maestro/metadata.json

@@ -0,0 +1,32 @@
+{
+  "id": "sap-maestro",
+  "name": "SAP Maestro",
+  "type": "skill",
+  "provider": "sap",
+  "harnesses": [
+    "claude-code",
+    "codex",
+    "cursor",
+    "gemini",
+    "kiro",
+    "other"
+  ],
+  "summary": "Route SAP AI workflow requests to the correct domain skill or agent via taxonomy classification, a routing table, and dispatch mode selection. Pure routing layer — does not touch live systems.",
+  "source_type": "original",
+  "category": "platform",
+  "official_docs": [
+    "https://help.sap.com/docs/btp/sap-business-technology-platform/sap-business-technology-platform",
+    "https://help.sap.com/docs/btp/sap-business-technology-platform/what-is-sap-btp",
+    "https://help.sap.com/docs/btp/sap-business-technology-platform/regions",
+    "https://help.sap.com/docs/btp/sap-business-technology-platform/account-model",
+    "https://help.sap.com/docs/btp/sap-business-technology-platform/entitlements-and-quotas",
+    "https://help.sap.com/docs/btp/sap-business-technology-platform/sap-btp-cockpit",
+    "https://help.sap.com/docs/btp/sap-business-technology-platform/cloud-management-tools-feature-set-b"
+  ],
+  "security_notes": "This skill is a pure routing layer and does not access any live SAP system. It must not accept or propagate credentials, tenant IDs, system URLs, or user tokens. Routing decisions that target live-tier skills must pass through the live-guard gate before dispatch is proposed.",
+  "last_verified": "2026-06-19",
+  "path": "skills/sap/sap-maestro",
+  "author": "github: Raishin",
+  "version": "0.1.0",
+  "lifecycle": "experimental"
+}

package/skills/sap/sap-maestro/references/official-sources.md

@@ -0,0 +1,53 @@
+# Official sources — SAP Maestro
+
+Use this reference when grounding SAP domain taxonomy and BTP product classification.
+
+**Evidence level**: documentation-based (official SAP Help Portal). No live-system evidence is collected by this skill.
+
+## SAP BTP taxonomy and account model
+
+- SAP Business Technology Platform — product home
+  https://help.sap.com/docs/btp/sap-business-technology-platform/sap-business-technology-platform
+  source_owner: SAP SE
+  topic_supported: BTP product taxonomy, service portfolio overview
+  why_needed: Authoritative domain classification for all BTP-native services
+
+- What is SAP BTP
+  https://help.sap.com/docs/btp/sap-business-technology-platform/what-is-sap-btp
+  source_owner: SAP SE
+  topic_supported: Platform overview, pillars (Database & Data Management, Analytics, Application Development, Integration, AI)
+  why_needed: Top-level domain taxonomy used by the routing table
+
+- BTP Regions
+  https://help.sap.com/docs/btp/sap-business-technology-platform/regions
+  source_owner: SAP SE
+  topic_supported: Regional availability of services; routing decisions that depend on region scope
+  why_needed: Routing to live-tier skills must account for regional availability
+
+- BTP Account Model
+  https://help.sap.com/docs/btp/sap-business-technology-platform/account-model
+  source_owner: SAP SE
+  topic_supported: Global account → directory → subaccount hierarchy; scope of routing decisions
+  why_needed: Disambiguates which tenant scope a routed request applies to
+
+- Entitlements and Quotas
+  https://help.sap.com/docs/btp/sap-business-technology-platform/entitlements-and-quotas
+  source_owner: SAP SE
+  topic_supported: Service entitlement context for routing decisions involving provisioned vs. unprovisioned services
+  why_needed: Prevents routing to skills whose target service is not entitled in the target subaccount
+
+- BTP Cockpit
+  https://help.sap.com/docs/btp/sap-business-technology-platform/sap-btp-cockpit
+  source_owner: SAP SE
+  topic_supported: Cockpit navigation model; reference for subaccount-scoped routing
+  why_needed: Clarifies cockpit scope when routing to landscape discovery or provisioning skills
+
+- Cloud Management Tools Feature Set B
+  https://help.sap.com/docs/btp/sap-business-technology-platform/cloud-management-tools-feature-set-b
+  source_owner: SAP SE
+  topic_supported: Feature Set B account management APIs; relevant for routing to API-driven landscape skills
+  why_needed: Routing to API-based skills requires confirming Feature Set A vs B capability boundary
+
+## Grounding rule
+
+SAP Help Portal documentation describes product taxonomy and service scope. It does not prove which services are entitled, deployed, or reachable in the user's specific BTP account. Routing decisions must be validated against user-provided account context before a live-tier dispatch is proposed.

package/skills/sap/sap-maestro/references/safety-checklist.md

@@ -0,0 +1,39 @@
+# Safety checklist — SAP Maestro
+
+Use before any dispatch decision, especially when routing leads to a live-tier skill.
+
+## Non-negotiables
+
+- Do not accept, store, echo, or propagate SAP system credentials, client secrets, OAuth tokens, BTP subaccount IDs, S/4HANA system IDs, tenant identifiers, or user passwords.
+- Do not invent skill IDs or agent names. Only route to skills and agents declared in the catalog.
+- Do not bypass the live-guard gate. A routing classification alone is not authorization to access a live system.
+- Do not propose dispatch to a `mutating-runtime` skill without explicit step-by-step approval from the user.
+- Do not conflate domain guidance across routing buckets (e.g., do not apply S/4HANA transport logic to BTP platform requests).
+- Do not mark a routing decision as `catalog-evidence` unless the skill or agent is actually declared in this repository's catalog.
+- Do not skip the split-routing step for `cross-domain` requests. Routing a cross-domain request to a single-domain skill will produce wrong guidance.
+
+## What people get wrong
+
+- **Skipping classification**: Dispatching directly to a skill without classifying the domain first. This causes wrong-skill errors and conflated guidance.
+- **Treating routing as authorization**: Classification and routing tell you *which* skill to load. They do not authorize live-system access.
+- **Conflating BTP platform with S/4HANA**: BTP subaccount operations and S/4HANA system configuration are different domains with different live-tier rules.
+- **Routing `unrouted` domains to the closest-sounding skill**: If no routing table entry matches, return `unrouted`. Do not approximate.
+- **Forgetting the live-guard gate**: Every `gated` dispatch mode requires explicit user confirmation. This is not optional.
+
+## When to push back
+
+- Push back when the user asks to skip classification and go straight to a live-tier skill.
+- Push back when the user provides a system URL or credentials and asks to "just connect."
+- Push back when the routing table returns `unrouted` and the user asks you to pick the "closest" skill anyway.
+- Push back when the dispatch would be to `sap-guarded-transport-import` without the 17-step sequence acknowledged.
+
+## Evidence labels
+
+Use exactly one of:
+
+- `documentation-based` — grounded in SAP official docs or BTP taxonomy
+- `catalog-evidence` — grounded in a declared skill/agent entry in this repo
+- `user-provided evidence` — stated by the user in this session
+- `inference` — derived reasoning not directly confirmed by docs or catalog
+
+Inference must always be labeled. Never present inferred routing as documentation-based.

package/skills/sap/sap-maestro/references/workflow-and-output.md

@@ -0,0 +1,73 @@
+# Workflow and output contract — SAP Maestro
+
+Use this reference for all routing, classification, and dispatch decisions.
+
+## Domain taxonomy
+
+The following top-level SAP domain buckets are used for classification:
+
+| Domain | Description | Example Products |
+|--------|-------------|-----------------|
+| `btp-platform` | SAP BTP platform services (not application-layer) | BTP Cockpit, Cloud Foundry, Kyma, BTP CLI |
+| `btp-integration` | Integration Suite and hybrid integration | SAP Integration Suite, API Management, Event Mesh |
+| `btp-ai` | AI and ML services on BTP | SAP AI Core, AI Launchpad, GenAI Hub, HANA ML |
+| `btp-data` | Data and analytics on BTP | SAP Datasphere, SAP Analytics Cloud, HANA Cloud |
+| `s4hana-core` | S/4HANA core ERP logic | Finance, Procurement, Manufacturing modules |
+| `s4hana-extensibility` | S/4HANA custom code, clean core, and extensibility | ABAP Cloud, RAP, BAdIs, key-user apps |
+| `s4hana-transport` | S/4HANA system landscape and transport management | TMS, CTS+, STMS, landscape definitions |
+| `successfactors` | SAP SuccessFactors HCM | Employee Central, Recruiting, Learning |
+| `ariba` | SAP Ariba procurement | Sourcing, Contracts, Buying |
+| `grc` | Governance, Risk, and Compliance | SAP GRC Access Control, Process Control |
+| `identity` | Identity and access management | SAP IAS, SAP IPS, Identity Provisioning |
+| `cross-domain` | Spans two or more of the above | Integration + S/4 + BTP in one request |
+
+## Routing table
+
+| Domain | Matched skill or agent | Dispatch mode | Live tier |
+|--------|----------------------|---------------|-----------|
+| `btp-platform` | `sap-live-readonly-landscape-discovery` | gated (read-only) | read-only-runtime |
+| `btp-integration` | `sap-live-readonly-landscape-discovery` | gated (read-only) | read-only-runtime |
+| `s4hana-extensibility` | `sap-clean-core-debt-review` | direct | none |
+| `s4hana-transport` | `sap-guarded-transport-import` | gated (mutating) | mutating-runtime |
+| `cross-domain` | split into domain-scoped sub-requests | split | varies |
+| `btp-ai`, `btp-data`, `successfactors`, `ariba`, `grc`, `identity` | `unrouted` — no matching skill declared | unrouted | N/A |
+
+## Dispatch modes
+
+- **direct**: Load the named skill immediately. No live-guard gate required (static advisory skill).
+- **gated (read-only)**: Confirm with the user that read-only live access is authorized and credentials are in scope before loading.
+- **gated (mutating)**: Full live-guard sequence applies. Do not dispatch without explicit step-by-step approval per `sap-guarded-transport-import` protocol.
+- **split**: Decompose the request into domain-scoped sub-requests. Route each sub-request independently through the table above.
+- **unrouted**: No declared skill covers this domain. Return the classification, explain the gap, and suggest the user consult SAP official documentation or engage an SAP partner.
+
+## Live-guard gate
+
+Before dispatching to any skill with `live tier` of `read-only-runtime` or `mutating-runtime`:
+
+1. Confirm the user has authorized live-system access in this session.
+2. Confirm the user understands which live system will be accessed.
+3. Confirm the downstream skill's live-environment rules have been reviewed.
+4. For `mutating-runtime`: confirm the 17-step guarded mutation sequence in `sap-guarded-transport-import` is understood and approved.
+
+Do not skip the live-guard gate. Do not propose live dispatch based on classification alone.
+
+## Workflow
+
+1. **Receive and parse** the incoming request.
+2. **Classify** into one or more domain taxonomy buckets.
+3. **Look up** the routing table. If `cross-domain`, decompose.
+4. **Select dispatch mode** per the table.
+5. **Apply live-guard gate** if dispatch mode is `gated`.
+6. **Return output** per the output contract below.
+
+## Output contract
+
+Return:
+
+1. Domain classification (bucket name + rationale)
+2. Evidence label (documentation-based / catalog-evidence / user-provided / inference)
+3. Routing table entry that fired
+4. Dispatch mode selected
+5. Live-guard gate status (not applicable / pending / cleared)
+6. Recommended next action (load skill / split request / escalate / unrouted explanation)
+7. Refusal trigger if any live-guard condition is unmet