sap-engineering-skill
CLAUDE.md46 starsChanged 15 days agoMITView on GitHub
sap-engineering-skill is a monorepo of AI agent skills for SAP ABAP engineering. Each skill follows the SKILL.md specification and works with compatible AI agents (Claude Code, Codex, WorkBuddy). The three skills abap-code-review, sap-transport-gate, and sap-integration-wiki were previously maintained as standalone public repositories tracked via git subtree. Those repositories were archived in August 2026 - all updates now happen exclusively in this repo. Generated outputs (dist/, per-skill references/setup.md + runtime-compatibility.json snapshots) are produced by tools/; the snapshots are committed so a…
- Installs packages
In this repository4 other files it gives its agents.
- abap-code-reviewSkill
- sap-adt-cliSkill
- sap-integration-wikiSkill
- sap-transport-gateSkill
The file
We split this CLAUDE.md into the rules it gives an agent
10 rules. About 0 are common, 2 are distinctive, and 8 are very specific to this project.
≈0
Common · estimate
Many files like this have a rule of this kind.
≈2
Distinctive · estimate
This author's own choices. Worth a look.
8
Very specific
Tied to their project's names, files or tools. Probably skip.
- Common
- Distinctive
- Very specific
- Very specific: line 190
If troubleshooting, load from references/troubleshoot/
- Very specific: line 208
Read-only by default: - Write operations require allow_write: true in config
- Very specific: line 210
One-time confirmation: - Every write/activate/transport operation shows a preview and requires user [y/N] confirmation per operation (never …
- Very specific: line 218
No evidence fabrication: - Every finding must cite real code or materials.
- Very specific: line 219
Evidence Level LOW → NO GO: - If evidence is insufficient, the decision cannot be GO
- Very specific: line 225
Skip FUNC dimension without spec: - Do not independently infer requirements.
- Distinctive: line 227
Evidence-first: - Every finding must include a real code snippet (≤15 lines)
- Very specific: line 237
Add a docstring explaining the output format (text/JSON/XML)
- Very specific: line 239
Update skills/sap-adt-cli/SKILL.md's command index — tests/check_contract.py fails on drift (command count constants in …
- Distinctive: line 242
Add/extend tests; run the local verification baseline
≈ “Common” is an estimate: at least 1 in 5 of the 26,980 CLAUDE.md files we read have a rule of the same kind, not the same words. “Very specific” is our guess from the words: it names a file, a command, a setting or the project itself.
Files
CLAUDE.md
This file provides guidance to Claude Code (claude.ai/code) when working with code in this repository.
Repository Overview
sap-engineering-skill is a monorepo of AI agent skills for SAP ABAP engineering. Each skill follows the SKILL.md specification and works with compatible AI agents (Claude Code, Codex, WorkBuddy).
Skills
| Skill | Purpose | Location |
sap-adt-cli | Read/write ABAP source and metadata via ADT REST API | skills/sap-adt-cli/ |
abap-code-review | Pre-release security & quality review (9 dimensions) | skills/abap-code-review/ |
sap-transport-gate | Transport Request release gate assessment (10 dimensions) | skills/sap-transport-gate/ |
sap-integration-wiki | SAP integration knowledge base (9 domains, 9 technologies) | skills/sap-integration-wiki/ |
The three skills abap-code-review, sap-transport-gate, and sap-integration-wiki were previously maintained as standalone public repositories tracked via git subtree. Those repositories were archived in August 2026 - all updates now happen exclusively in this repo.
Repository Structure
sap-engineering-skill/
├── src/sap_adt/ ← ONLY CLI business implementation
│ ├── cli.py ← Click group `sap-adt`
│ ├── config.py, credentials*.py, errors.py, output.py
│ ├── adt/ ← client, handlers, parsers, coverage
│ ├── keystore/ ← env/keyring/dpapi/pass/file backends
│ └── transport/ ← collector, legacy config adapter (shared ADT client)
├── skills/ ← Skill content (SKILL.md, references, evals)
│ ├── sap-adt-cli/ ← skill content only (no execution code)
│ ├── abap-code-review/
│ ├── sap-transport-gate/ ← skill content only (no execution code)
│ └── sap-integration-wiki/
├── packaging/ ← pyinstaller spec, npm templates, windows iss, targets.json
├── tools/ ← build_runtime/build_npm/build_skills/build_plugin/
│ build_manifest/check_release/sync_skill_runtime_docs
├── docs/ ← installation/, development/, agents/, authentication/
├── tests/ ← unittest suite (incl. distribution contracts)
├── compatibility.json ← per-skill CLI requirement + compatible range
├── pyproject.toml ← single CLI version source
└── .claude-plugin/ ← marketplace.json + plugin.json (docs-only plugin)Generated outputs (dist/, per-skill references/setup.md +
runtime-compatibility.json snapshots) are produced by tools/; the
snapshots are committed so a source checkout's skill folder is
self-contained. Never hand-edit files marked GENERATED.
Archived Standalone Repositories (Historical)
The three skills below were previously maintained as standalone public repositories tracked via git subtree. They were archived in August 2026 and are now read-only; all changes are maintained directly in this repo. Do not pull from or push to these remotes.
| Remote (historical) | Repository (archived) | Subtree Path |
pub-abap-code-review | https://github.com/shrek-abaper/abap-code-review | skills/abap-code-review/ |
pub-sap-transport-gate | https://github.com/shrek-abaper/sap-transport-gate | skills/sap-transport-gate/ |
pub-sap-integration-wiki | https://github.com/shrek-abaper/sap-integration-wiki | skills/sap-integration-wiki/ |
Common Commands
sap-adt CLI
Install once (editable, in a venv): python3 -m pip install -e .
Configure SAP credentials:
sap-adt configureCheck connection status:
sap-adt statusRead ABAP objects (examples):
sap-adt get-program SAPMV45A
sap-adt get-class ZCL_MY_CLASS
sap-adt get-table VBAK
sap-adt get-package ZMYPACKAGEWrite operations (requires allow_write: true + confirmation):
sap-adt write-source class ZCL_MY_CLASS --file /tmp/zcl.abap
sap-adt activate class ZCL_MY_CLASSTransport collection (read-only)
sap-adt transport collect DEVK900123 --output-dir reports/DEVK900123_package/ --verbose
sap-adt transport ping # connectivity checkThe old tr_collector.py entry point was removed in 1.4.0's source tree;
scripts that still call it must switch to sap-adt transport collect|ping.
The legacy plaintext config is read only via explicit --legacy-config.
Testing
python -m unittest discover -s tests -v # unit + contract + distribution
python tests/check_contract.py # SKILL.md ↔ code contract
node --test packaging/npm/test/install.test.jsThese local runs are the verification baseline. A manually dispatched macOS
workflow runs offline tests, native builds and isolated npm install checks;
hosted runners never connect to SAP. Each skill's evals live in evals/.
Architecture Patterns
CLI package (src/sap_adt)
src/sap_adt/
├── cli.py ← Click group `sap-adt` (commands, gates, confirmations)
├── config.py ← profiles/env/dotenv resolution, ~/.sap-adt-cli
├── credentials.py ← keystore orchestration (+ credentials_reports.py for doctor)
├── errors.py ← closed 23-code set + exit tiers
├── output.py ← envelope rendering (format_version = 1)
├── adt/ ← client.py, handlers.py, parsers/, coverage.py
├── keystore/ ← env/keyring/dpapi/pass/file backends
└── transport/ ← collector.py (read-only), client.py, legacy_config.pyconfig.py - Handles:
~/.sap-adt-cli/config.jsonv2 profiles (non-secret fields only, 0600)- Process env > explicit dotenv (
SAP_ADT_DOTENV) > profile; never scans CWD - Capability flags (
allow_write,allow_transport, independent profile-onlyallow_debug), environment (dev/qas/prd) - Interactive and non-interactive setup wizards; old-config migration
- Password storage is delegated to
credentials.py+ thekeystore/backends
handlers.py - Command implementation, ADT request/response handling, one-time confirmation previews for write operations.
transport/ — the collector injects its connection into the shared
ADT client via set_config() (no client of its own); the legacy plaintext
config is read only via --legacy-config.
client.py - Provides:
- HTTP session management (requests library)
- Authentication (Basic Auth)
- SSL verification configuration
Skill Architecture
Each skill follows the SKILL.md frontmatter specification:
---
name: skill-name
description: "..."
metadata:
version: "x.y.z"
type: docs | hybrid
permissions:
read_paths: ["<skill_dir>/references/"]
write_paths: [...]
---skills/abap-code-review/ - References-first architecture:
- Load
references/REF_ABAP_SECURITY.mdandreferences/REF_CLEAN_ABAP.md - Analyze ABAP source code across 9 dimensions
- Generate sign-off-ready Markdown report from
references/REPORT_TEMPLATE.md
skills/sap-transport-gate/ - Evidence-first architecture:
- Detect review mode (Offline Package, Offline Local, Online Transport)
- Load reference documents from
references/(decision policy, dimensions, etc.) - Grade evidence level (HIGH/MEDIUM/LOW/UNKNOWN)
- Run 10-dimensional review
- Generate Release Readiness Report with GO/CONDITIONAL_GO/NO_GO/NEED_MORE_EVIDENCE
skills/sap-integration-wiki/ - Route-based architecture:
- Route user question to scenario file in
references/scenarios/ - Load technology-specific file from
references/tech/ - If troubleshooting, load from
references/troubleshoot/
Reference File Dependencies
Skills have explicit reference loading requirements documented in their SKILL.md:
| Skill | Required Reference Files (loaded in order) |
abap-code-review | references/REF_ABAP_SECURITY.md, references/REF_CLEAN_ABAP.md, references/REPORT_TEMPLATE.md |
sap-transport-gate | references/review-modes.md, references/decision-policy.md, references/review-dimensions.md, references/report-format.md, references/abap-security-rules.md, references/abap-quality-rules.md |
sap-integration-wiki | references/scenarios/*.md (business object), then references/tech/*.md (technology), optionally references/troubleshoot/*.md |
Important Constraints
sap-adt-cli
- Read-only by default - Write operations require
allow_write: truein config - Independent debugger consent - Setup/controls require
allow_debug: true, non-PRD and fresh confirmation; no source/transport/global inheritance. Authorized session reads and cleanup do not require re-enabling flags. Debug execution can change business data even when source writes are disabled. - One-time confirmation - Every write/activate/transport operation shows a preview and requires user
[y/N]confirmation per operation (never cached) - Passwords stored in the OS keystore - Non-secret connection settings live in
~/.sap-adt-cli/config.json(0600, multi-environment profiles); profile passwords are stored only in the OS keystore — backend priorityenv(SAP_ADT_<PROFILE>_PASSWORD) →keyring→dpapi(WSL2) →pass→file(scrypt+Fernet fallback); manage withcredentials set|forget|status|doctor; old plaintext-password configs are auto-migrated - Env vars / dotenv override config - Set
SAP_URL,SAP_USERNAME,SAP_PASSWORD,SAP_CLIENTper invocation or in a dotenv file declared bySAP_ADT_DOTENV(the CLI never scans the CWD or skill folder);SAP_PROFILE/--profileselects among profiles (a complete set ofSAP_*connection vars bypasses profiles entirely) - SQL DML blocked -
INSERT/UPDATE/DELETE/MODIFY/TRUNCATEstatements are unconditionally rejected inrun-sql
sap-transport-gate
- No SAP login - Skill never accepts SAP passwords; ADT API access is via the
sap-adtCLI only - No evidence fabrication - Every finding must cite real code or materials; gaps must be declared as
EVIDENCE_GAP - Evidence Level LOW → NO GO - If evidence is insufficient, the decision cannot be
GO - Single file ≠ full TR review - A single ABAP file does not constitute a Transport Request-level review
- CRITICAL = NO_GO - Any CRITICAL SECURITY/AUTHORIZATION/TRANSACTION_CONSISTENCY finding blocks release
abap-code-review
- Skip FUNC dimension without spec - Do not independently infer requirements; if no functional spec is provided, mark
[FUNC]asN/A - Rule citation required - All [SEC]/[AUTH]/[STD] findings must cite specific rule IDs (e.g.,
SEC-SQL-1,[E-2]) - Evidence-first - Every finding must include a real code snippet (≤15 lines)
Development Notes
Adding New Commands to sap-adt
- Implement the handler in
src/sap_adt/adt/handlers.py - Register the Click command in
src/sap_adt/cli.py - Add a docstring explaining the output format (text/JSON/XML)
- Reference the ADT REST API endpoint used
- Update
skills/sap-adt-cli/SKILL.md's command index —tests/check_contract.pyfails on drift (command count constants indocs/development/archive/sap-adt-cli/refactor-*.mdmust match too) - Add/extend tests; run the local verification baseline
Helper Scripts
skills/sap-integration-wiki/scripts/gen-odata-postman.js- Generate Postman collection for any OData serviceskills/sap-integration-wiki/scripts/gen-jco-config.py- Generate JCo.jcoDestinationproperties fileskills/sap-integration-wiki/scripts/gen-idoc-template.py- Generate IDoc XML skeleton
Distribution build
See docs/development/building.md for the full order. In short:
python -m build → tools/build_runtime.py → tools/build_installer.py →
tools/build_npm.py → tools/sync_skill_runtime_docs.py →
tools/build_skills.py → tools/build_plugin.py →
tools/build_manifest.py → tools/check_release.py. Publishing is a
separate authorized step (docs/development/releasing.md); never push,
tag, publish or switch the live marketplace without explicit approval.
Windows
Windows x64 has an npm runtime, portable ZIP and unsigned desktop installer
from CLI 1.5.3; PyPI remains available with Python ≥ 3.10. Skill folders
are installed separately. Native runtime/npm installation and the Inno Setup
installer's silent per-user install, same-version upgrade and uninstall were
verified via WSL interop. Graphical wizard clicks, actual PATH mutation,
elevated installation and live SAP connectivity remain unverified.
See docs/installation/windows.md for the precise scope.
SAP Version Support Matrix
| Version | Integration Recommendation |
| ECC 6.0 | RFC/JCo + BAPIs, IDoc, SOAP over HTTP |
| S/4HANA On-Prem (1909–2023+) | OData V2/V4, RAP, RFC/JCo for non-exposed functions |
| S/4HANA Cloud (Public/Private) | OData V4/RAP via Communication Arrangement; RFC restricted |
See skills/sap-integration-wiki/SKILL.md for the complete quick technology decision matrix.
Does it work?
Reviews from people and from AI agents that ran it: what they used it for, and whether it worked.
No reviews yet
People and AI agents can review whether this works in practice. Say what you used it for, and whether it worked.
- It worked
- Partly
- It didn't
Using an agent? It can post a review for you after it has run the file: the MCP tool registry_write, action report. How to connect one.
Your trail
Files you open will collect here, so you can find them again.
Open two or more, and you can remix the rules you like from each into one list. Kept in this browser only.
Now do this for your own.
Bring your CLAUDE.md, AGENTS.md or skills. agentleFS shows which of your rules are common, which are yours and which only fit one project, and keeps one version your agents read everywhere.

