system-prompt-sop SKILL.md skillA

system-prompt-sop from labs21-dev/agents-stack is an agent skill: a SKILL.md folder an AI agent loads when a task calls for it.

Install it, pinned
mdr add labs21-dev/agents-stack/system-prompt-sop@git:20260912.9d47cf0

Pins git:20260912.9d47cf0 in mdr.lock. Pin the fingerprint instead.

Audit grade
A, 17 of 17 checks passed
Kind
SKILL.md skill
Repository
labs21-dev/agents-stack
Path
system-prompt-sop/SKILL.md
Versions recorded
1
Latest version
git:20260912.9d47cf0, committed Sep 12, 2026
Size
5,747 bytes
Repository stars
27
License
none declared

Design-time SOP for system prompts / agent harness prompts: route every constraint to its correct enforcement layer (validator/schema > harness > prompt example > prompt principle) instead of piling rules or vibes into the prompt. Use when the user asks to design, write, rewrite, review, or tune a system prompt; convert a rule-pile into principles; complains that an agent ignores rules, drifts in long tasks, or breaks red lines; wants to port a prompt to another model; or debates…

Collected from public GitHub. Every version is saved exactly as it was and never changes. Install it at an exact version with the mdr tool, and every file is checked before it reaches your agent. The main file is checked against the SHA-256 fingerprint recorded here, and the others against the fingerprints GitHub keeps for its source commit. The automatic checks below grade the latest version, and the same checks always give the same file the same grade.

What the file says

# System Prompt SOP(約束路由)

把「設計 system prompt」從風格之爭變成工程決策:不問「原則還是規則」,
**對每一條約束單獨路由到它夠格的最高執行層**。Prompt 只收留少數 advisory
內容,且每條帶 why + 可觀測訊號 + eval 覆蓋。

## 核心機制(三條,路由表的地基)

1. **Prompt 不教判斷,只做選擇。** 原則是對「模型已具備的行為分佈」的壓縮
   錨定。採用原則的前置條件:該判斷已在目標模型能力分佈內(用探針驗證,
   見 `references/constraint-routing.md`)。
2. **Prompt 內一切皆 advisory。** `NEVER` 在 prompt 裡只是提醒;紅線只有
   外移到 harness 才為真。
3. **注意力天花板不豁免原則。** 原則是壓縮率較好的牆,不是免檢通行證;
   超預算時降階或外移,不要改寫得更「精煉」。

## Flow

0. **Intake batch**(見下,缺項一次問完,執行中不再中斷)
1. **盤點約束**:列出每一條想要的行為 → `templates/routing-worksheet.md`
2. **逐條路由**:按下方路由表判定去處(核心步驟,機制詳解見
   `references/constraint-routing.md`)
3. **組裝 prompt**:只裝路由結果中屬於 prompt 的部分 →
   `templates/prompt-skeleton.md`
4. **出廠審計**:逐項過 `references/audit-checklist.md`
5. **Probe 迭代**:原則項跑 20-case 探針;不通過就降階(規則/範例),
   不是改措辭

## Intake batch(步驟 0,一次問完)

缺以下任一項且上下文推不出來時,開頭一次問齊:

- **目標模型**(決定 capability 前檢;未定 → 寫 runtime-agnostic,探針標 pending)
- **執行環境 / harness**(決定紅線去處;未定 → 紅線去處寫抽象條款並標 TBD)
- **既有 prompt**(改寫)或**目標行為清單**(新建)
- **既有 eval / probe**(沒有 → 步驟 5 先建最小 probe,這是入場券不是收尾)

## 路由表(核心)

| 判定條件 | 去處 | 形式 |
|---|---|---|
| case 可枚舉 **且** 失敗可機械檢測 | **移出 prompt** → schema / validator / linter | 程式即規格 |
…

Read the whole file at its exact version.

How to install

The exact contents, by fingerprint
mdr add labs21-dev/agents-stack/system-prompt-sop@sha256:cbfdfe52848156d5

The line at the top of this page pins the label git:20260912.9d47cf0, which is what most people want. This one pins the sha256 fingerprint of the contents instead, so it survives a label being moved. Either way the exact version is written to mdr.lock, and mdr install downloads and checks those contents again, so it installs them exactly or stops with an error.

Badge

mdr badge

[![mdr](https://markdownregistry.com/badge/art_oz6bjlkkk3kdk6oy.svg)](https://markdownregistry.com/a/art_oz6bjlkkk3kdk6oy)

5 badge image fetches in 30 days. What the badge shows.

Versions

The registry has recorded 1 version of this file, committed on Sep 12, 2026.

versioncommittedcommitsizeaudit
git:20260912.9d47cf0 latest2026-09-12 9d47cf0 5,747 BA view

Audit of the latest version

A  17 of 17 checks passed. Fixed automatic checks with no AI involved, so the same checks give the same answer every run.
  • pass: Frontmatter block present
  • pass: Frontmatter declares a name
  • pass: Frontmatter declares a description
  • pass: Size between 200 bytes and 200 KB (5747 bytes)
  • pass: No zero-width or bidi control characters
  • pass: No instruction hidden inside an HTML comment
  • pass: No link to an exfiltration or paste host
  • pass: No credential-shaped string
  • pass: No instruction to send local credentials anywhere
  • pass: No text hidden with inline styles
  • pass: No prompt-injection phrasing
  • pass: No curl or wget piped into a shell
  • pass: No recursive delete of root, home or parent
  • pass: No instruction to read or print local credentials
  • pass: No base64 blob over 200 characters
  • pass: No link to a raw IP address
  • pass: No script tag

Source

GitHub

labs21-dev/agents-stack · 27 stars · license none declared · pushed 2026-09-19 · branch main

API

GET https://markdownregistry.com/api/v1/artifacts/art_oz6bjlkkk3kdk6oy
GET https://markdownregistry.com/api/v1/resolve?ref=labs21-dev/agents-stack/system-prompt-sop
GET https://markdownregistry.com/api/v1/blob/cbfdfe52848156d54a5019c71aac9aab747b09288a435106e665b4fcf89aa1fe

Your AI assistant does the legwork. You hear about the deals worth your word. Hand yours the standing instructions at modelranch.com and it joins the network that reads files like this one.

Guides: the SKILL.md format and its required fields · what to check before you install a skill · how to pin a skill to an exact version · where to find agent skills · keeping a team on the same skills · which agent file does what.

More from labs21-dev/agents-stack

AGENTS.md AGENTS.md file
labs21-dev/agents-stack · AGENTS.md
git:20260918.45cbe6b · audit A · 27 stars
DESIGN.md@agent-plugin DESIGN.md file
labs21-dev/agents-stack · agent-plugin/DESIGN.md
git:20260918.f8cebc6 · audit A · 27 stars
agent-memory SKILL.md skill
labs21-dev/agents-stack · agent-plugin/agent-memory/SKILL.md · Portable markdown memory: CoALA working / semantic / episodic / procedural, plus optional personas as core. Use when a…
git:20260918.45cbe6b · audit A · 27 stars
generate-audio SKILL.md skill
labs21-dev/agents-stack · agent-plugin/generate-audio/SKILL.md · Generate spoken audio through OpenRouter text-to-speech. Use when the user asks to synthesize speech, narrate text…
git:20260915.ee027f0 · audit A · 27 stars
generate-image SKILL.md skill
labs21-dev/agents-stack · agent-plugin/generate-image/SKILL.md · Generate or edit raster images with OpenRouter. Use when the user asks to create, draw, illustrate, or restyle an…
git:20260915.ee027f0 · audit A · 27 stars
generate-text SKILL.md skill
labs21-dev/agents-stack · agent-plugin/generate-text/SKILL.md · Generate text through an OpenRouter text model. Use when the user asks for a written answer, summary, rewrite…
git:20260915.ee027f0 · audit A · 27 stars
generate-video SKILL.md skill
labs21-dev/agents-stack · agent-plugin/generate-video/SKILL.md · Generate video clips through OpenRouter's asynchronous video API. Use when the user asks for a video, animated clip…
git:20260915.ee027f0 · audit A · 27 stars
read-audio SKILL.md skill
labs21-dev/agents-stack · agent-plugin/read-audio/SKILL.md · Transcribe or analyze local audio through OpenRouter. Use when the user supplies an audio file and asks what it says…
git:20260915.ee027f0 · audit A · 27 stars
read-image SKILL.md skill
labs21-dev/agents-stack · agent-plugin/read-image/SKILL.md · Understand, describe, OCR, inspect, or answer questions about a local image using an OpenRouter vision model. Use when…
git:20260915.ee027f0 · audit A · 27 stars
read-video SKILL.md skill
labs21-dev/agents-stack · agent-plugin/read-video/SKILL.md · Understand, summarize, inspect, or answer questions about a local video using an OpenRouter video-capable model. Use…
git:20260915.ee027f0 · audit A · 27 stars
agentic-dev-e2e SKILL.md skill
labs21-dev/agents-stack · agentic-dev-e2e/SKILL.md · End-to-end agentic feature development AND bug-fix workflow. Use when the user gives a feature goal or wish item (e.g…
git:20260912.9d47cf0 · audit A · 27 stars
design-context SKILL.md skill
labs21-dev/agents-stack · design-context/SKILL.md · Use when a request will produce frontend UI, a visual asset, an image prompt, or a video prompt, AND the intent/style…
git:20260912.9d47cf0 · audit A · 27 stars

Every file in labs21-dev/agents-stack

Browse by kind, by grade A, or by owner.