Sc4sap:compare Programs

SC4SAP Compare Programs

Reads 2–5 ABAP programs that implement the same business scenario but in different variants, analyzes them across 10 business dimensions, and emits a side-by-side Markdown comparison targeted at functional consultants (not developers).

Companies often split one logical business flow (e.g. GR list) into 2–3 programs so the same data answers different questions — MM sees quantity, CO sees value; KR sees e-tax-invoice fields, EU sees VAT codes; month-end controllers see aggregates, warehouse clerks see live transactions.

This skill crystallizes why each variant exists so a consultant can:

• pick the right program for a new requirement instead of creating a 4th,
• map fit/gap across localizations,
• brief a handover or knowledge transfer without reading ABAP.

Every response triggered by this skill MUST begin with [Model: · Dispatched: ] per [../../common/model-routing-rule.md](../../common/model-routing-rule.md) § Response Prefix Convention.

Multi-phase skill. Before each Agent(...) dispatch, emit ▶ phase= () · agent= · model= per [../../common/model-routing-rule.md](../../common/model-routing-rule.md) § Phase Banner Convention.

Step 4b is the integration point for Type A teamMode (Cross-Module Consultant Panel). When module_set ≥ 2 (existing Step 4b gate), Round 1 runs always — positions are parsed for ownership conflicts (same program claimed PRIMARY by 2+ modules). On conflict, escalate to Rounds 2-3 per [team-mode.md](team-mode.md). No conflict → positions pass directly to Step 5 writer render (legacy path). Base protocol: [../../common/team-consultation-protocol.md](../../common/team-consultation-protocol.md).

• User says "compare programs", "what's the difference between A and B", "MM vs CO version", "country-specific programs", or equivalent in the user's language
• Consultant handover / AMS transition — need to document "when to use which"
• Fit/Gap analysis across country rollouts — same flow, different programs
• Rationalization / decommissioning — considering whether to merge duplicates

• Only one program → use /sc4sap:program-to-spec instead
• User wants code quality review (not business intent) → /sc4sap:analyze-code
• User wants to build a new program → /sc4sap:create-program
• More than 5 programs — break into multiple comparison sessions or escalate to /sc4sap:team

MANDATORY — runs as Step 0 before any MCP call or user interaction.

Invoke /sc4sap:trust-session with parent_skill=sc4sap:compare-programs to pre-grant MCP tool + file-op permissions (eliminates per-tool prompts during parallel program reads).

• If .sc4sap/session-trust.log already has a line within the last 24h, skip silently.
• Otherwise run it and surface the one-line confirmation.
• All Agent dispatches within this skill MUST pass mode: "dontAsk".

Full spec: see [../trust-session/SKILL.md](../trust-session/SKILL.md).

MANDATORY: Read the companion files below before executing. Each covers a self-contained section:

| Companion | Scope | |-----------|-------| | [comparison-scope.md](comparison-scope.md) | 10 business comparison dimensions + default scope + scope-selection prompt template | | [workflow.md](workflow.md) | 6-step execution flow (Input → Scope → Facts × N → Analyze → Render → Follow-up) | | [dispatch-prompts.md](dispatch-prompts.md) | Full Agent(...) prompt bodies for Steps 3, 4, 4b, 5 (kept out of workflow.md to honor the 200-line cap) | | [report-template.md](report-template.md) | Markdown report skeleton (Executive Summary → Matrix → Per-program detail → Recommendation) |

Per-step model allocation. Skill frontmatter pins the main thread to Haiku; each Agent(...) carries its own model (frontmatter or explicit override).

• Main orchestrator (Haiku 4.5) — Steps 1, 2, 6: intake, scope confirmation, follow-up menu. Holds NO source code / AST / screens in its own context; all MCP-heavy work runs inside agents.
• Facts extraction (sap-code-reviewer × N, Sonnet 4.6 via model: "sonnet" override) — Step 3 (absorbs the old "Read Phase"). Each reviewer reads ONE program itself (GetProgFullCode / GetAbapAST / screens / GUI status / text elements / where-used) and returns structural facts only — no quality scoring. Sonnet is sufficient because this pass is rule-based extraction, not novel code generation; matches the base tier of common/model-routing-rule.md § Tier 1.
• Analysis + narrative (sap-analyst × 1, Opus 4.7) — Step 4: a SINGLE dispatch covering module classification + dimension scoring + executive summary + recommendation. Keeps the analyst's context continuous across reasoning layers instead of fragmenting into 4 chained calls.
• Module specialists (conditional, sap-{module}-consultant × K, Opus 4.7) — Step 4b: when programs span 2+ modules (MM+CO, SD+FI, etc.), each distinct module gets a consultant dispatch to explain "what would a {module} user use this for". The analyst's scoring consumes these in its narrative.
• Rendering (sap-writer × 1, Haiku 4.5) — Step 5: renders the final Markdown using report-template.md. Pure formatting from structured state.

All Agent dispatches pass mode: "dontAsk" (trust-session already granted in Step 0).

Report language mirrors the user's current conversation language.

• User writes in Korean → report in Korean (section headers + body).
• User writes in English → report in English.
• User writes in Japanese → report in Japanese.
• If mixed or unclear, default to the user's last full sentence language.
• Do not ask — detect and proceed. Only ask if the user explicitly requests a specific language.

.sc4sap/comparisons/{prog1}__vs__{prog2}[__vs__{prog3}…]-{YYYYMMDD}.md

• Program names are uppercase, underscore-safe (slashes → _).
• If the filename exceeds 120 chars (5-program case), use .sc4sap/comparisons/compare-{YYYYMMDD}-{hash6}.md and list the programs inside the front-matter.

1. Load dimension catalog from comparison-scope.md.
2. Execute the 6 steps in workflow.md in order.
3. Render the artifact using report-template.md.

Do not skip the companion-file reads — the dimension list, step order, and report schema all live there.

This skill reads source code + DDIC metadata + where-used + screen/GUI-status/text-element metadata only. It does NOT call GetTableContents or GetSqlQuery. If the user asks for sample row data to illustrate a difference, refuse per common/data-extraction-policy.md and document the request in the report's Risk & Open Questions section instead.

• /sc4sap:program-to-spec — single-program reverse-engineering (vertical depth)
• /sc4sap:analyze-code — quality review (what's wrong, not what's different)
• /sc4sap:analyze-cbo-obj — CBO package inventory (complementary context for dimension 8)
• /sc4sap:deep-interview — use before comparison if user is unsure which programs to include

Task: {{ARGUMENTS}}

Source & license

This open-source skill is cataloged on AgentStack and links to its original source — we do not rehost the code.

• Author: babamba2
• Source: babamba2/superclaude-for-sap
• License: MIT

Install and usage instructions live in the source repository linked above.