@manyone-mds/agents

MDS LLM context, agent prompts, and design-system knowledge base (the MDS Vault).

A static, markdown-only package that ships the MDS Vault — a structured LLM-facing knowledge base covering tokens, components, conventions, workflows, and architecture. Install this alongside @manyone-mds/components when building AI tooling, custom agents, or IDE assistants that need MDS context.

Install

npm install @manyone-mds/agents

No runtime code, no dependencies. Pure docs.

What ships

node_modules/@manyone-mds/agents/
├── mds-vault/        ← the knowledge base (17 folders, structured markdown)
└── templates/        ← drop-in agent-rules files for downstream projects
    ├── CLAUDE.md     ← Claude Code
    ├── AGENTS.md     ← Codex CLI, Cursor (current)
    ├── .cursorrules  ← Cursor (legacy)
    ├── GEMINI.md     ← Gemini CLI
    └── .claude/
        └── settings.json   ← optional Tier-2 auto-loading hook

Setup — three tiers of integration

The vault sits in node_modules after install. Most agent tools don't scan node_modules for docs on their own — you need one file in your project root to point the agent at the vault. The right file depends on which agent you use.

Tier 1 — Project-root rules file (minimum)

Copy the template that matches your agent into your project root.

| Agent | Command | |---|---| | Claude Code | cp node_modules/@manyone-mds/agents/templates/CLAUDE.md ./CLAUDE.md | | Codex CLI | cp node_modules/@manyone-mds/agents/templates/AGENTS.md ./AGENTS.md | | Cursor (current) | cp node_modules/@manyone-mds/agents/templates/AGENTS.md ./AGENTS.md | | Cursor (legacy) | cp node_modules/@manyone-mds/agents/templates/.cursorrules ./.cursorrules | | Gemini CLI | cp node_modules/@manyone-mds/agents/templates/GEMINI.md ./GEMINI.md |

All five template files are identical content with different filenames. Each agent reads its own conventional file at session start, finds the vault, and reads atoms on demand.

Multi-tool teams can copy several at once — they don't conflict.

Tier 2 — Auto-load vault on every prompt (Claude Code only)

If you want the vault's navigation map and rules injected into model context on every prompt (not just at session start), add the hook config:

mkdir -p .claude
cp node_modules/@manyone-mds/agents/templates/.claude/settings.json .claude/settings.json

Requires jq on $PATH. The hook prepends the vault README and _meta/vault-rules.md to every prompt so the agent always knows where to look. This is how the MDS development repo itself runs.

Tier 0 — Nothing in project root

The vault is installed but inert. No agent will find it. Don't ship this way.

Vault structure

The vault root README lists the folder map. Each folder has an index.md and atom files (one fact per file) with structured frontmatter:

• agents/ — LLM-specific guidance (rule enforcement, vibecoding defaults)
• components/ — one file per MDS component (props, states, slots, gotchas)
• patterns/ — composed UI patterns (e.g., MDSLoginMask)
• foundation/ — design tokens (colours, typography, spacing, etc.)
• conventions/ — abstract rules (naming, contract, css-modules, tokens-policy)
• workflows/ — operational how-tos (add component, change token, release)
• … and twelve more (see mds-vault/README.md)

More

Full design system docs and source: github.com/damian-888/MDS-v1.0.0.

License

MIT