---

name: run-sap-router-skill description: >- Run, smoke-test, and drive the SAP Router Skill CLI (sap_router.py, memory_manager.py, xls_to_bapi.py, template_repo.py, abap_serializer.py, cpi_iflow_packager.py). Use when asked to run, start, build, test, verify, or smoke-test the sap-router-skill, its routing engine, MEMORY.md compaction, or the XLS/CSV-to-BAPI converter. trigger: keywords: [run, smoke-test, router, cli, sap-router, skill, memory, bapi, xls, csv] intent: >- Use when running, smoke-testing, building, or verifying the SAP Router Skill CLI, routing engine, or data converters.

Run: sap-router-skill

Multi-IDE: Same driver.py works on Claude, Codex, Gemini Antigravity, and Cursor. Canonical driver: .agents/skills/run-sap-router-skill/driver.py. IDE wrappers and generated mirrors point to this canonical procedure.

The repository contains Python operational scripts under scripts/. The core smoke gate remains self-contained: no live SAP system, network, or credentials needed — routing is a static lookup table, memory_manager is local file I/O, xls_to_bapi parses CSV/XLSX. Everything runs offline.

Script	Purpose
sap_router.py	Routing engine: ADT-first; SAP GUI scripting fallback (mcp-sap-gui); sf-mcp
memory_manager.py	Session context file (MEMORY.md) lifecycle
xls_to_bapi.py	CSV/XLSX → BAPI JSON payload converter
template_repo.py	Offline ABAP template repository with {{placeholders}}
abap_serializer.py	Multi-format ABAP packer: .nugg, abapGit, ZDOWNLOAD XML

The agent path is the smoke driver: it launches every CLI surface against a throwaway workspace and asserts exit codes + output.

Repository-owned skill/MCP discovery

For ambiguous SAP tasks, search the unified local catalog before choosing a skill or MCP:

python scripts/source_catalog.py search "task description"
python scripts/mcp_launcher.py search --query "task description"

Bundled MCP snapshots are fail-closed (disabled_candidate) until reviewed and promoted in .agents/registries/mcps.json. No runtime GitHub lookup is allowed.

All paths below are relative to the unit root (sap-router-skill/).

Prerequisites

• Python 3.11+ for the router and controlled harness. On Windows the interpreter is python.
• No third-party packages required for CSV. XLSX support is optional — pip install openpyxl only if you pass .xlsx files; .csv always works.

Run (agent path) — smoke driver

python .agents/skills/run-sap-router-skill/driver.py

Exit 0 = all checks reported by the driver passed; 1 = a check failed (offending line printed). The final output reports the current check count. The driver creates and deletes its own temp workspace — it touches no project files (ZROUTER opt-in state is redirected to a temp file via SAP_ROUTER_OPTIN_FILE, so the smoke run never mutates the real decision). Use it as the regression gate after editing any script in scripts/.

Record the command, exit status and final counts from the actual run as verification evidence. Driver success covers its offline smoke checks; live SAP acceptance requires a response for the requested target and operation.

Run (manual, individual CLIs)

# Routing: dev op -> ARC-1 ADT; ADT-unavailable -> SAP GUI scripting (mcp-sap-gui); sf_* -> sf-mcp
python scripts/sap_router.py route --action read_source                       # -> ARC-1 (ADT)
python scripts/sap_router.py route --action MM_CREATE_MATERIAL                 # -> needs-functional-context (no BAPI fired)
python scripts/sap_router.py route --action MM_CREATE_MATERIAL --functional    # -> BAPI dispatch (functional context)

# Parallel subagent dispatch (same-wave agents launch concurrently)
python scripts/sap_router.py dispatch-plan --spec requirements.md              # wave plan for stages 2-8
python scripts/sap_router.py crew-dispatch --task "find the leak then review the diff"  # concurrent cavecrew agents

# ZROUTER opt-in (optional RFC accelerator; never auto-installed)
python scripts/sap_router.py zrouter offer      # show the install offer (default: decline)
python scripts/sap_router.py zrouter decline    # persisted; routing stays ADT-first -> SAP GUI scripting
python scripts/sap_router.py zrouter accept     # then: zrouter install
python scripts/sap_router.py zrouter status     # current opt-in + enabled flag

# Session memory lifecycle (writes MEMORY.md at the given path)
python scripts/memory_manager.py init --input MEMORY.md --sys S4H --client 100 --env DEV --usr DEVELOPER
python scripts/memory_manager.py add  --input MEMORY.md --module MM --action-name CreateMaterial --status OK --fields '{"obj":"MAT123","tr":"S4HK900001"}'
python scripts/memory_manager.py verify --input MEMORY.md   # exit 0 if valid
python scripts/sap_router.py status --memory-file MEMORY.md

# XLS/CSV -> BAPI payload (9 modules: MM, SD, FI, QM, PP, WM, CO, HCM, BASIS)
python scripts/xls_to_bapi.py template --output tmpl.csv --module MM --action CREATE_MATERIAL
python scripts/xls_to_bapi.py convert  --input data.csv  --module MM --action CREATE_MATERIAL

# Template repository (offline code templates with {{placeholders}})
python scripts/template_repo.py init
python scripts/template_repo.py seed
python scripts/template_repo.py list
python scripts/template_repo.py resolve --template MM_CREATE_MATERIAL --values '{"HEADER":"h","DESCRIPTION":"d"}'

# ABAP serializer (multi-format ABAP packer/unpacker)
python scripts/abap_serializer.py generate --source file.abap --name ZCL_FOO --type CLAS --format nugg --output out/
python scripts/abap_serializer.py package  --source file.abap --name ZCL_FOO --type CLAS --output out/
python scripts/abap_serializer.py extract  --input file.nugg --output out/
python scripts/abap_serializer.py split    --source multi_class.abap --output out/

Gotchas (battle scars from this session)

• Generated template is NOT valid input. xls_to_bapi.py template writes the field descriptions as row 2 (e.g. "Material type (e.g. FERT)"). Feeding that file straight back into convert/validate fails required-field validation. Replace row 2 with real data first.
• memory_manager.py add silently auto-inits a missing MEMORY.md with hard-coded defaults S4H/100/DEV and user $USERNAME. If you skip init, your session header is wrong, not absent — no error is raised.
• Hard caps enforced on write: max 20 active blocks (older ones roll into a ## ARCHIVE single-line summary), and the whole file must stay ≤100 lines. verify exits 1 if a block exceeds 2 detail lines or the file exceeds 100.
• 8 SAP modules field-defined: MM, SD, FI, QM, PP, WM, CO, HCM, BASIS. Unknown module/action exits 1 with an "Available:" list — not a crash.
• route never validates the action against a catalog — unknown actions fall through to the SAP GUI scripting fallback (mcp-sap-gui) default — never BLOCKED. It's a classifier, not a validator.
• Functional WRITE actions are gated — CREATE_*, CHANGE_*, POST_*, GOODS_MOVEMENT, etc. return needs-functional-context (no BAPI fired) unless --functional is passed. BAPIs fire only when a real functional action requires them — a bare token (smoke test) never auto-fires one. Pure reads and explicit *_GUI actions are not gated.
• ZROUTER is opt-in, never the engine. Routing never auto-probes or auto-installs it. --use-zrouter is honoured only after zrouter accept; a zrouter decline is persisted (in zrouter_optin.json) and routing keeps working ADT-first -> SAP GUI scripting with no ZROUTER dependency.
• template_repo.py resolve requires double quotes in --values JSON on Windows. Use single-quote wrapping on the outside: --values '{"KEY":"val"}' works on both Windows and Unix. Double-quote-wrapped JSON on cmd.exe may break.
• abap_serializer.py _class_to_type auto-detects from name prefix. ZCL_ → CLAS, ZIF_ → INTF, ZCX_ → CLAS, other Z* → PROG. Use explicit --type if your naming convention differs.

Troubleshooting

Symptom	Fix
ModuleNotFoundError: openpyxl	pip install openpyxl, or use .csv instead of .xlsx.
convert reports missing required fields on a fresh template	You fed the template's description row as data — see first gotcha.
verify exits 1 "exceeds 100 lines"	MEMORY.md got too big; run python scripts/memory_manager.py compact --input MEMORY.md.
route/status print nothing	status shells out to memory_manager.py show; pass a real --memory-file path.
template_repo add says "0 placeholders"	The ABAP file has no {{PLACEHOLDER}} markers. Add placeholders or use seed for pre-built templates.
abap_serializer extract "No source found"	.nugg file may be malformed — the CDATA block must contain valid source.