Sc4sap:ask Consultant

SC4SAP Ask Consultant

Single entrypoint for asking a SAP module consultant agent an operational question. The skill auto-selects the right consultant(s) from the question text + project config, honors the configured SAP environment, and returns a faithful answer — no code generation, no object creation, just the consultant's judgment.

Invoke /sc4sap:trust-session with parent_skill=sc4sap:ask-consultant so the consultant agent's read-only MCP calls (SearchObject, GetTable, GetPackage, GetWhereUsed, etc.) proceed without prompting.

• If .sc4sap/session-trust.log already has a line within the last 24h, skip silently.

Full spec: see ../trust-session/SKILL.md.

• .sc4sap/config.json → sapVersion (ECC / S4 On-Prem / S4 Cloud Public / S4 Cloud Private), abapRelease, industry, country, activeModules
• .sc4sap/sap.env → SAP_URL, SAP_CLIENT, SAP_LANGUAGE, SAP_INDUSTRY, SAP_COUNTRY, SAP_ACTIVE_MODULES (as fallback)

Pass these to the consultant via the dispatch prompt so its answer reflects the actual landscape. If any key is missing, ask the user before dispatching — do NOT let the consultant invent assumptions.

The consultant agent's Mandatory_Baseline (Tier 1 + Tier 2 per ../../common/context-loading-protocol.md) ensures it loads spro-lookup.md, customization-lookup.md, active-modules.md at session start and reads only the configs/{MODULE}/*.md file(s) the question needs.

1. Explicit mention: "MM 물어봐" / "ask SD" / "FI 컨설턴트" → dispatch that consultant directly.
2. Keyword inference: map keywords to modules (PO/EKKO/EBELN → MM; invoice/BKPF/BSEG → FI; sales order/VBAK → SD; cost center/KOSTL → CO; MRP/production order/AFKO → PP; WBS/PROJ → PS; maintenance order → PM; inspection lot → QM; cash management/treasury → TR; payroll/infotype → HCM; warehouse/bin → WM; freight order → TM; InfoObject/DataSource → BW; sourcing/Ariba Network → Ariba; dump/transport/kernel → BC).
3. Multi-module: if 2-3 modules match with similar signal strength, dispatch all in parallel. Example: "MM PO가 FI에 어떻게 전기되는지" → dispatch MM + FI in parallel; the skill then composes the answer.
4. Unclear: ask the user which module first — one question, one round.

Supported consultants (15 total):

• Module: sap-sd-consultant, sap-mm-consultant, sap-fi-consultant, sap-co-consultant, sap-pp-consultant, sap-ps-consultant, sap-pm-consultant, sap-qm-consultant, sap-tr-consultant, sap-hcm-consultant, sap-wm-consultant, sap-tm-consultant, sap-bw-consultant, sap-ariba-consultant
• Basis: sap-bc-consultant

1. Trust bootstrap — § <Session_Trust_Bootstrap>. (Haiku · skill-to-skill)
2. Environment load — read config.json + sap.env; surface resolved values on the FIRST turn only (one line: SAP: <version> · <industry> · <country> · active: <modules>). (Haiku · main thread)
3. Module routing — apply § <Module_Routing>. If ambiguous, ask one question and stop. (Haiku · main thread)
4. Consultant dispatch — one Agent(...) per resolved module (parallel in a single message when 2–3 modules match). Emit phase banner per dispatch:

   ▶ phase=4 (consultant-<MODULE>) · agent=sap-<module>-consultant · model=Opus 4.7
   

   Dispatch shape:

   Agent({
     subagent_type: "sc4sap:sap-<module>-consultant",   // frontmatter already pins claude-opus-4-7
     description: "<MODULE> consultation — <topic>",
     prompt: <user question + environment context + expected format>
   })
   

   teamMode variant: when N ≥ 2, Step 4 doubles as Round 1 of teamMode — use the Round 1 spawn shape in team-rounds.md § Round 1 (adds team_name, name, charter-file reference; consultants write POSITION files instead of returning directly).
5. Synthesis (conditional — only when ≥ 2 consultants replied) — dispatch sap-writer with model: "sonnet" override. Writer's base model is Haiku (pure formatting); Sonnet override gives the light cross-domain reasoning needed to detect agreement / disagreement between consultant answers without jumping to Opus. Emit banner:

   ▶ phase=5 (synthesis) · agent=sap-writer · model=Sonnet 4.6
   

   Dispatch shape:

   Agent({
     subagent_type: "sc4sap:sap-writer",
     model: "sonnet",   // override base Haiku — synthesis needs light reasoning
     description: "Cross-module synthesis — <modules>",
     prompt: """
       Synthesize the following consultant answers for the user question.
       Identify shared points, flag disagreements (with a one-line "WHY they differ" note),
       and produce a cross-module summary. Do NOT re-answer the question — only compose
       from the supplied consultant outputs.
   
       User question: <question>
       Environment: <sapVersion> · <industry> · <country> · modules: <list>
   
       Consultant answers:
       <MODULE_A>: <answer_a>
       <MODULE_B>: <answer_b>
       [...]
     """
   })
   

   teamMode variant: if Round 1 POSITIONs diverged (per team-rounds.md § Divergence check), do NOT run the legacy synthesis above — follow team-rounds.md Rounds 2-3 then team-mode.md § Synthesis (task-list–driven writer dispatch). On single-consultant case: SKIP Step 5 entirely — main thread (Haiku) just forwards the consultant's answer to Step 6.
6. Return & follow-up — present the final answer (single-consultant: verbatim; multi-consultant: synthesis output as the body + per-module subsections). Offer follow-up paths: /sc4sap:create-program (if the answer leads to a new build), /sc4sap:program-to-spec (if user wants the existing asset documented), /sc4sap:analyze-code (if quality review needed). (Haiku · main thread)

No writes: this skill never calls Create* / Update* / Delete* / Activate* / CreateTransport. If the consultant's answer suggests a change, the user must run a separate creation / modification skill.

No row extraction: GetTableContents and GetSqlQuery are NOT used. Schema / DDIC reads are fine (GetTable, GetStructure, GetDataElement, GetDomain, SearchObject).

[Model: <main-model> · Dispatched: Opus×<n> (<consultant names>)]

🧭 Consultant: sap-<module>-consultant
🌐 Environment: <sapVersion> · <industry or "—"> · <country or "—"> · active modules: <list>

<consultant's faithful answer>

---
💡 Next steps (optional):
- /sc4sap:create-program — if this leads to a new build
- /sc4sap:program-to-spec — to document an existing asset
- /sc4sap:analyze-code — to review existing code

For multi-module dispatches, the 🧭 Consultant line lists all names, the body leads with the sap-writer synthesis (Sonnet) — shared points, disagreements, cross-module summary — followed by one verbatim subsection per consultant.

Dispatch-summary examples in the prefix:

• Single consultant: [Model: Haiku 4.5 · Dispatched: Opus×1 (sap-mm-consultant)]
• Multi-consultant: [Model: Haiku 4.5 · Dispatched: Opus×2 (sap-mm-consultant, sap-fi-consultant), Sonnet×1 (sap-writer synthesis)]
• Multi-consultant teamMode: [Model: Haiku 4.5 · Dispatched: Opus×2 (sap-mm-consultant, sap-fi-consultant), Sonnet×1 (sap-writer team-synthesis) · Team: 2 members × 2 rounds]

Task: {{ARGUMENTS}}