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/archfill/tsumugi/agents-mdgit clone --depth 1 https://github.com/archfill/tsumugiWhat 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 | $0.01844 | $0.01844 |
| Opus 5 | $0.00922 | $0.00922 |
| Sonnet 5 | $0.00369 | $0.00369 |
| Haiku 4.5 | $0.00184 | $0.00184 |
Grade A, and why
tsumugi 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 — 171 lines — stays where its author put it; the contents beside it link to each section on GitHub.
tsumugi — AI 開発ガイド(必読)
新規実装の前に必ず最初に読むこと。書く場所と再利用ルールを規定する。
ディレクトリ構造(書く場所のルール)
| 層 | 役割 | 何を書く |
|---|---|---|
core/ |
ビジネスロジック(純粋) | use case、検索アルゴリズム |
data/ |
永続化 | Drizzle schema、repository |
external/ |
外部 I/O | embeddingなどの非LLMアダプタ |
interfaces/ |
入力アダプタ | MCP tool、REST handler、transport |
lib/ |
横断ユーティリティ | id、errors、config |
依存方向: interfaces → core → (data, external) → lib
逆方向の import 禁止(lib は循環依存禁止)。
詳細なファイル一覧はソースツリーを正とし、このガイドでは固定しない。新しい機能も上記の
依存方向と責務に従って配置する。external/llm/ や同等のserver-side LLM clientは作らない。
既存資産(再実装禁止)
ID 生成
lib/id.ts の newId(prefix) を必ず使う。
import { newId } from "../lib/id.js";
const id = newId("obs");
DB アクセス
data/repos/<entity>.ts 経由で行う。db を直接触らない(schema 定義 / migration ファイル以外)。
import { observationRepo } from "../../data/repos/observation.js";
await observationRepo.insert(row);
Embedder
external/embedding/singleton.ts の getEmbedder()。
import { getEmbedder } from "../../external/embedding/singleton.js";
const embedding = await getEmbedder().embed(text);
検索
core/search/hybrid.ts の hybridSearch()。bigm/vector を直接呼ばない。
import { hybridSearch } from "../../core/search/hybrid.js";
const hits = await hybridSearch(input);
設定
lib/config.ts の loadConfig()。process.env 直接読み禁止。
エラー
lib/errors.ts の TsumugiError 派生を使う。new Error() しない。
import { ValidationError, ExternalError } from "../../lib/errors.js";
throw new ValidationError("content is required");
採用パターン
Use case の形
- ファイル:
core/<domain>/<verb>.ts(例:core/observation/save.ts) - 入口で Zod parse、出力は型付きオブジェクト
- repo / external を直接 import で呼ぶ(DI コンテナ不要)
- 副作用は use case 内で完結
Repository の形
- ファイル:
data/repos/<entity>.ts export const xxxRepo = { method1, method2 }で集約- Drizzle の
$inferSelect/$inferInsertで型を取り出す
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 · 171 lines · 1,844 tokens per session scan A ccff415af410
tsumugi AGENTS.md is an instructions file published in the GitHub repository archfill/tsumugi (0 stars, last pushed 23d ago), licensed Apache-2.0. It adds 1,844 tokens to every session, about $0.0092 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 2026-08-31.
Other instructions, from other repositories
ariadne AGENTS.md
Instructions for mclaut/ariadne, covering agents.md — ariadne, what this is, build / test / lint, architecture and runtime layout (not the repo).
searching-apple-notes AGENTS.md
Instructions for yinanli1917-cloud/searching-apple-notes, covering codex project instructions, apple-notes-mcp, tech stack, directory structure and conventions.
searching-apple-notes CLAUDE.md
Instructions for yinanli1917-cloud/searching-apple-notes, covering apple-notes-mcp, tech stack, directory structure, conventions and codex harness.
kb-mcp-server CLAUDE.md
Claude Code instructions for code-runner-xx/kb-mcp-server, covering claude.md — kb-mcp-server 开发手册(规则书), 0. 项目铁律(必读), 1. 版本与技术选型(锁定), 2. 环境变量(.env.example) and 3. 目录结构.
Graft CLAUDE.md
Instructions for AEndrix03/Graft, a project described as: Local-first semantic cache for AI agents. A small C daemon + CLI that remembers what your agent learned across sessions. Plugs into Claude Code, Codex, Gemini CLI, and Claude Desktop / ChatGPT via MCP. No LLM calls, no SaaS, no API key.
ariadne CLAUDE.md
Instructions for mclaut/ariadne, covering claude.md — ariadne, what this is, build / test / lint, architecture and runtime layout (not the repo).