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 one editable HTML report targeted at functional consultants (not developers): a coloured verdict matrix with a Differences only switch, field-by-field differences of the selection screens and output lists, the screens drawn side by side, and a decision guide — versioned and editable like the program-to-spec and program-to-manual pages, with an English companion.

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.</Purpose>

<Response_Prefix> Every response triggered by this skill MUST begin with [Model: <main-model> · Dispatched: <sub-summary>] per ../../common/model-routing-rule.md § Response Prefix Convention. </Response_Prefix> <Target_Object>The target object is looked up on the active profile's system only. Not found → stop and ask per ../../common/multi-profile-artifact-resolution.md § Target object not on the active system — never substitute a similar name, never switch profiles on your own.</Target_Object>

<Phase_Banner> Multi-phase skill. Before each Agent(...) dispatch, emit ▶ phase=<id> (<label>) · agent=<name> · model=<Opus|Sonnet|Haiku> per ../../common/model-routing-rule.md § Phase Banner Convention. Resolve the dispatch mode first per ../../common/model-dispatch-mode.md: auto (default) uses the model below/in the rule; user-defined asks the user Opus / Sonnet / Haiku (recommended one first) before each dispatch group. Resolve the cost mode too per ../../common/cost-mode.md (standard default; eco → pass BOTH model and effort on every Agent(...) call per its lever-A table (omitting effort silently runs the agent at its frontmatter effort, usually high), Skill("sc4sap:eco", "--attach") first and on every continuation turn, and this skill's eco recipe). </Phase_Banner>

<Team_Mode> 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. No conflict → positions pass directly to Step 5 writer render (legacy path). Base protocol: ../../common/team-consultation-protocol.md. </Team_Mode>

<Use_When>

• 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 </Use_When>

<Do_Not_Use_When>

• 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 </Do_Not_Use_When>

<Session_Trust_Bootstrap> 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.

Full spec: see ../trust-session/SKILL.md. </Session_Trust_Bootstrap>

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

<Agent_Composition> Per-step model allocation. The main thread follows the session model (model: inherit); each Agent(...) carries its own model (frontmatter or explicit override).

• Main orchestrator (session model) — 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-checker × N, Haiku) — Step 3. Each checker reads ONE program (or one include batch of a large program) itself — source / screens / GUI status / text elements / where-used — and returns verbatim FACT lines only: no scoring, no narrative. Read-and-copy work belongs on Haiku per common/model-routing-rule.md § Tier 3; the Opus analyst does every interpretation.
• Analysis + narrative (sap-analyst × 1, Opus) — 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) — 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, model: "sonnet" override) — Step 5: writes the per-program screen specs (selection / ALV), renders them, and writes the report .md from report-template.md. Sonnet because the screen specs must be transcribed 1:1 and the decision guide is consultant prose.
• English companion (sap-writer × 1, Sonnet override) — Step 5.3, when the report language is not English (default on; the user can opt out).
• Main thread builds the HTML page with scripts/spec/md-to-html.mjs --editable (Step 5.1), the optional Excel copy with scripts/spec/md-to-xlsx.mjs (Step 5.1b), and runs the review loop.

SAP MCP permission prompts are auto-approved by the sc4sap permission-approver PreToolUse hook. </Agent_Composition>

<Language_Policy> 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. </Language_Policy>

<Output_Location> <base>/comparisons/<NAME>-v<ver>-<lang>.html (the page), .xlsx when excel was chosen, and .md (its source — kept for rebuilds and imports), <NAME>-<lang>.history.json, screen pictures under _assets/<NAME>-<lang>/<PROG>/, plus the same with -en for the English companion. <base> per ../../common/multi-profile-artifact-resolution.md.

• <NAME> = the program names joined by __vs__ (uppercase, / → _); over 120 characters → compare-<hash6> (rule in render-steps.md § Names).
• Versions 1.0, 1.1, … as in program-to-spec: a re-run of the same programs is the next minor version; a page edit by a reader comes back as its own version. </Output_Location>

<Execution_Summary>

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. </Execution_Summary>

<Data_Extraction_Safety> This skill reads source code + DDIC metadata + where-used + screen/GUI-status/text-element metadata only. It does NOT call GetTableContents; its one GetSqlQuery is the T100 lookup of the message texts the programs use (workflow Step 3b: fields SPRSL, ARBGB, MSGNR, TEXT, only the classes and numbers in the facts, logon language + E). Sample rows in the drawn screens are invented — never business data, never real customer, vendor or company names. 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. </Data_Extraction_Safety>

<Related_Skills>

• /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) </Related_Skills>

Task: $ARGUMENTS