| 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 taxonomycatalog-evidence — grounded in a declared skill/agent entry in this repouser-provided evidence — stated by the user in this sessioninference — 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:
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.
