SAP Router Skill for Codex

Canonical source: .agents/. Karpathy wrapper: mandatory. Caveman compression: default. Do not copy or fork skill bodies here; regenerate from canonical source.

Runtime root:

• SAP_ROUTER_ROOT must point to the canonical sap-router-skill repository.
• change the working directory to SAP_ROUTER_ROOT before relative commands.
• fail closed if scripts/source_catalog.py is not present there.

Dynamic local discovery:

• search skills: python scripts/source_catalog.py search "task description"
• search MCPs: python scripts/mcp_launcher.py search --query "task description"
• bundled MCPs are disabled candidates until reviewed; no runtime GitHub lookup.

Local optimization:

• prefer rtk for supported verbose CLI commands.
• use Context Mode for large outputs, indexed fetches, and session checkpoints.

Parity proof:

• skills: 203 sha256:046f6d52886520160ec0c76373806fd76f42ea8b489ebbe57b54e61fd29919fd
• profiles: 38 sha256:e746d5f5374f567c00e178b16645e378521476723efc61383bdc562f25d9c827
• registries: 14 sha256:e98ac633c7685dfb67cb878c1ae16833b0dcfccac5865035f5862c5fde42caf5

Run: python scripts/generate_ide_assets.py check

--- project-doc ---

SAP Router Skill & Harness — Multi-IDE Agent Instructions

203 canonical skills available across Claude, Codex, Gemini Antigravity, and Cursor.

IDE	Skill Directory	Entry Point
Claude Code	.claude/skills/ (203 generated skills)	.claude/skills/run-sap-router-skill/SKILL.md
Antigravity (Gemini)	.gemini/skills/ (203 generated skills)	.gemini/skills/run-sap-router-skill/SKILL.md
Codex / OpenAI	.codex/AGENTS.md (shared canonical instructions)	.codex/AGENTS.md
Cursor	.cursor/AGENTS.md (shared canonical instructions)	.cursor/AGENTS.md

All skills auto-trigger by file context and keyword. See SKILL.md for master dispatch. Current (v7.1.0): sap_harness.py (controlled agent execution + eval suite) + ZROUTER Remote FileSystem adapter + Automation Pilot & Skill-Share pipelines.

SAP Harness v7.1 (Controlled Multi-Agent & Eval Engine)

# Autonomous mission execution
python scripts/sap_harness.py run --agent cpi --task "triage failed messages in CPI"

# Evaluation & Benchmark test suite (Offline Hermetic by default)
python scripts/sap_harness.py eval --suite all
python scripts/sap_harness.py benchmark

# Specialized subagent directory
python scripts/sap_harness.py agents

# Test ZROUTER Remote FileSystem integration (for vscode_abap_remote_fs)
python scripts/sap_harness.py test-remote-fs

# Automation Pilot MCP compiler & Skill packager
python scripts/automation_pilot_compiler.py --catalog cf --command ListCfOrgs
python scripts/skill_packager.py sap-cpi-flowpilot --slack-json

Implemented v7 operating model

Canonical source is .agents/skills plus .agents/registries/mcp-capabilities.json. Claude and Gemini skill trees are generated mirrors; Codex and Cursor reuse .agents/skills through AGENTS/rules/agent assets instead of maintaining duplicate skill trees.

Generate/check IDE assets:

python scripts/generate_ide_assets.py generate --targets all
python scripts/generate_ide_assets.py check

Capability routing is fail-closed:

python scripts/validate_catalog.py --strict
python scripts/mcp_launcher.py list --capability sap.cpi.message.read
python scripts/mcp_launcher.py list --capability sap.apim.proxy.read

CPI/APIM/Fiori/UI5/CAP now route through API/CLI/plugin MCPs first and use Web UI MCP fallbacks (integration-suite-ui-mcp, apim-ui-mcp) only when background access is blocked. The Web UI bridge uses the logged-in browser session and does not accept credential values.

apim-ui-mcp carries two API Management channels and reports which one each call used. OAuth client_credentials on an apiportal-apiaccess service key is SAP's documented path and wins whenever it is configured; the logged-in browser session is an undocumented fallback for reads and tests, running fetch() inside the page so cookies never leave the browser. Scope is API lifecycle only — proxies, products, policies, key value maps and their tests — never business data; SAP API Policy v.4.2026a §2.2.2 points agentic business-API access at the Integration Suite MCP Gateway instead, which apim_mcp_gateway_probe checks the tenant for.

Mutating capabilities use plan/approval/commit semantics. Local approval state is handled by:

python scripts/approval_broker.py plan --capability sap.apim.proxy.deploy --target DEV

What this project is

40 Python operational scripts plus Node.js review gates — no live SAP system required credentials required. Everything runs offline:

Run / verify

python .claude/skills/run-sap-router-skill/driver.py
# Exit 0 = all 78 checks passed. No project files modified.

Requires Python 3.11+ for the router and controlled harness. No packages needed for CSV; pip install openpyxl for XLSX support only.

Routing rules (v4.2.0 — ADT-first, GUI-fallback, caveman-optimized)

Decision tree:

1. CAVEMAN DELEGATION? (fix/find/review, 1-2 files) → cavecrew-investigator/builder/reviewer
2. ADT AVAILABLE? (read_source, search, activate...) → ARC-1 / aibap / mcp-abap-adt
3. GUI FALLBACK? (SPRO, SM30, SU01, MM01, VA01...) → mcp-sap-gui (25 transactions)
4. FUNCTIONAL WRITE? (BAPI batch) → needs explicit --functional context (else needs-functional-context, NO BAPI fired); then BAPI-first / SAP GUI dispatch. ZROUTER RFC only if opted in (zrouter accept).
5. SPEC→PIPELINE? → sap_router.py pipeline (8 stages)
6. LLM / RAG? (prompt optimization, doc retrieval) → sap-llm-engineering skill + RAG routing (Pinecone, Supabase pgvector, Azure AI Search)

Memory session commands

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
python scripts/memory_manager.py verify --input MEMORY.md
python scripts/memory_manager.py compact --input MEMORY.md

Caps: 20 active blocks max, 100 lines total. Older blocks auto-archive.

XLS converter commands

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

Action-to-BAPI mapping (29 actions / 9 modules)

Functional WRITE actions need explicit --functional context (else needs-functional-context, no BAPI fired); they then dispatch BAPI-first / SAP GUI. ZROUTER RFC is opt-in only (zrouter accept) — never the default. ADT ops and code_search → ARC-1 ADT. sf_* → sf-mcp.

Template repository commands

python scripts/template_repo.py init                                                  # Create blank repo
python scripts/template_repo.py seed                                                  # Populate 12 starter templates
python scripts/template_repo.py list                                                  # List all templates
python scripts/template_repo.py add --file template.abap --module MM --action FOO     # Import from ABAP file
python scripts/template_repo.py resolve --template MM_CREATE_MATERIAL --values '{"HEADER":"h"}'  # Substitute placeholders
python scripts/template_repo.py export --template MM_CREATE_MATERIAL --format json    # Export for abapGit/nugget

ABAP serializer commands

python scripts/abap_serializer.py generate --source file.abap --name ZCL_FOO --type CLAS --format nugg --output out/
python scripts/abap_serializer.py generate --source file.abap --name ZCL_FOO --type CLAS --format abapgit --output out/
python scripts/abap_serializer.py package  --source file.abap --name ZCL_FOO --type CLAS --output out/  # All 3 formats
python scripts/abap_serializer.py extract  --input file.nugg --output out/
python scripts/abap_serializer.py split    --source multi_class.abap --output out/
python scripts/abap_serializer.py list-formats

ZROUTER_DISPATCH v2 additions

• zcl_zrouter_handler_abstract=>evaluate_expression uses GENERATE SUBROUTINE POOL + PERFORM for dynamic ABAP expression evaluation at runtime. Available in all handler subclasses.
• templates/ZCL_ABAP_REPL_V2.abap — improved SICF REPL with /UI2/CL_JSON, optional X-API-KEY, CSRF header check, mode: "expr" for lightweight GENERATE SUBROUTINE POOL eval vs mode: "full" for WRITE-capturing SUBMIT.

Key gotchas

• Template row 2 = field descriptions — not valid data row. Replace before converting.
• memory_manager add auto-inits a missing MEMORY.md silently with defaults.
• route is a classifier, not a validator — unknown actions fall to SAP GUI scripting. Functional writes are gated on --functional. ZROUTER is opt-in, never the default.
• template_repo.py works offline — resolves placeholders via Python string substitution. ABAP runtime eval uses GENERATE SUBROUTINE POOL (see ZROUTER_DISPATCH v2).
• abap_serializer.py _class_to_type auto-detects from name prefix: ZCL_→CLAS, ZIF_→INTF, ZCX_→CLAS, other Z*→PROG. Use explicit --type to override.

@RTK.md