Getting it into your agent
One page per mod, every tool's command on it. A separate URL per tool would split the same page into five that compete with each other.
npx agentmods add instructions/boxlogodev/sapstack/agents-mdgit clone --depth 1 https://github.com/BoxLogoDev/sapstackWhat it costs to keep this loaded
Counted locally with the o200k_base tokenizer, which is exact for GPT models; Claude uses its own tokenizer and its counts differ. Treat this as one consistent yardstick across the catalogue rather than a bill. Prices are per million input tokens.
| Model | Per session | Once invoked |
|---|---|---|
| Fable 5 | $0.07088 | $0.07088 |
| Opus 5 | $0.03544 | $0.03544 |
| Sonnet 5 | $0.01418 | $0.01418 |
| Haiku 4.5 | $0.00709 | $0.00709 |
Grade A, and why
sapstack AGENTS.md scanned grade A with 0 findings against 26 rules in 11 categories — prompt injection, anti-refusal, data exfiltration, privilege escalation, supply chain, agent snooping, system-prompt leakage, SSRF and excessive agency — measured yesterday.
A static scan of the body, not an audit. Every finding is printed with the line that produced it so you can judge whether it matters here. A mod is markdown that instructs an agent; that is exactly why what it instructs is worth reading.
Nothing flagged
None of the 26 patterns this scan looks for appear in this file: no shell pipes, no recursive deletes, no credential paths, no hidden text, no instruction-override or anti-refusal phrasing, no agent-config snooping. That is not a guarantee, it is the absence of the things that are checkable.
How it starts
The opening of the file, as written. The whole thing — 499 lines — stays where its author put it; the contents beside it link to each section on GitHub.
AGENTS.md — sapstack (v2.4.0)
이 파일은 sapstack의 프로젝트 운영 계약 정본입니다. Codex CLI·Kiro· Claude Code·Windsurf 등 모든 AI 에이전트가 이 파일을 읽습니다. Universal Rules, 응답 포맷, 라우팅 규칙은 여기에만 존재하며,
CLAUDE.md·.windsurfrules등 도구별 파일은 해당 도구 라우팅만 담습니다.
이 순서로 읽는다
| # | 무엇 | 어디 |
|---|---|---|
| 1 | Rules — 반드시 지킬 것 | 이 파일 + DESIGN.md(시각) |
| 2 | State — 지금 어디까지 왔나 | STATE.md ← 작업 시작 전 필수 |
| 3 | Decisions — 이 판단이 아직 유효한가 | decisions/active/ (뒤집힌 것은 superseded/) |
| 4 | Knowledge — 필요할 때만 | plans/ · aidlc-docs/ · docs/ |
규칙이 바뀌면 이 파일, 상태가 바뀌면 STATE.md, 판단이 바뀌면 decisions/.
한 사실은 한 곳에만 산다. 상위 규약: ~/.claude/workflows/project-memory.md
Claude Code 사용자는 이 파일과
plugins/*/skills/*/SKILL.md를 읽습니다.
이 저장소는 무엇인가?
sapstack은 SAP 운영 자문을 위한 통합 지식·에이전트·커맨드 플러그인 모음입니다. 23개 SAP 모듈(FI/CO/TR/MM/SD/PP/PM/QM/WM/EWM/HCM/SFSF/ ABAP/S4Migration/BTP/BASIS/BC/GTS/Cloud/IBP/SAC/Ariba/Integration-Cloud)
- 1개 메타 플러그인(sap-session, Evidence Loop 오케스트레이터)로 구성됩니다 (합계 24).
- 지식 원본 형식: Claude Code plugin marketplace (
plugins/*/skills/*/SKILL.md) - 이 파일: 모든 AI 에이전트가 공유하는 운영 계약 정본
- 저장소: https://github.com/BoxLogoDev/sapstack
🎯 Universal Rules (모든 SAP 답변에 적용)
철학: 이 규칙들은 sapstack Advisor Ethos를 강제합니다. 규칙의 이유는
ETHOS.md참조 — Ground-truth over plausibility, Evidence over confidence, No hardcoding, ECC≠S/4, Field language, Operator decides.
아래 규칙은 절대 위반 금지입니다:
- Never hardcode company codes, G/L accounts, cost centers, or org units. 사용자가 제공한 값을 사용하고, 제공이 없으면 물어보세요.
- Always ask for environment before answering:
- SAP Release (ECC 6.0 EhPx / S/4HANA release year)
- Deployment model (On-Premise / RISE / Public Cloud)
- Industry sector
- Always distinguish ECC vs S/4HANA behavior where they differ (e.g. BSEG vs ACDOCA, FD32 vs UKM_BP, F.05 vs FAGL_FC_VAL).
- Transport request required for any configuration change in dev/QA/prod.
- Simulate before actual run — AFAB, F.13, FAGL_FC_VAL, KSU5, MR11, F110 등은 반드시 Test Run 선행.
- Never recommend SE16N data edits in production.
- Always provide T-code + menu path for every action.
- Use field language, not dictionary Korean — 한국어 응답은 현장체 우선:
- 현장 외래어를 1순위로: "코스트 센터", "페이먼트 메소드", "트포", "미고"
- 외래어(공식 번역, 필드코드) 이중 병기: "코스트 센터 (원가센터, KOSTL)"
- 발화체 수용: "돌렸는데", "뜨네요", "안 돼요", "박아주세요", "튕겨요"
- T-code·약어 원형 유지 (
F110,PO,GR— 풀어 쓰지 말 것) - 업무 시점은
D-1,월마감 D+3,가결산같은 업계 표준 표기 - 상세:
plugins/sap-session/skills/sap-session/references/korean-field-language.md - 동의어 사전:
data/synonyms.yaml(58 용어 + 10 약어 + 15 업무 시점)
What this file has done since we first saw it
Hashed on every crawl. A supply-chain change to an agent config is a question of when, not whether, so the history is kept rather than the latest state alone.
- yesterday First seen · 499 lines · 7,088 tokens per session scan A 67bd343c896b
sapstack AGENTS.md is an instructions file published in the GitHub repository BoxLogoDev/sapstack (18 stars, last pushed 2d ago), licensed MIT. It adds 7,088 tokens to every session, about $0.0354 per session on Opus 5. A static security scan graded it A with 0 findings. No closer match exists in the catalogue, so it is treated as the original; first seen 2026-08-30.
Other instructions, from other repositories
abap2UI5 AGENTS.md
Instructions for abap2UI5/abap2UI5, covering agents.md — ai assistant guide for abap2ui5, where knowledge lives, project overview, related repositories and architecture.
vibing-steampunk copilot-instructions.md
Instructions for oisee/vibing-steampunk, covering copilot instructions for vsp, project, build, test, lint, build and run all unit tests.
onebase CLAUDE.md
Instructions for ivanarama/onebase, covering claude.md, сборка и запуск, cli + сервер (без cgo) — основной бинарь, gui с нативным окном (cgo + webview2 на windows) and тесты.
mcp-erpnext AGENTS.md
Instructions for Casys-AI/mcp-erpnext, covering agents.md, project overview, project structure & module organization, build, test, and development commands and run all tests (also: deno task test).
abap2UI5 CLAUDE.md
Instructions for abap2UI5/abap2UI5: All project guidance lives in AGENTS.md — the single source of truth for this repository (architecture and the layered design, the src/00-99 packages, the generated app/webapp mirror, the build and delivery pipeline, the CI gates and ABAP code style).
balaka CLAUDE.md
Instructions for artivisi/balaka, covering claude instructions, project overview, current status, key files and development guidelines.