Angular Philosophies

Chapter 11 · AI

Using this guide with AI

Installing the guide as a skill, a plugin, rules files or an MCP server.

4 rules3 min readSource on GitHub

The guide is written so that a person can read it and a tool can follow it. Every chapter ends with a block of short, imperative rules, and those blocks are collected into formats that AI coding tools already understand. Pick the one your tool uses. They all carry the same rules, because they are generated from the same chapters.

FormatFilesWorks with
Claude Code plugin.claude-plugin/, .mcp.json, ai/skills/Claude Code
Claude skillai/skills/angular-philosophies/Claude Code, the Claude apps, anything that loads SKILL.md
AGENTS.mdai/AGENTS.mdCodex, Cursor, Copilot, Gemini CLI, Claude Code and most other agents
Cursor rulesai/cursor/*.mdcCursor
llms.txtllms.txt, ai/llms-full.txtAny tool that can read a URL
MCP serverai/mcp/server.mjsAny MCP client

The generated files are not edited by hand. To change a rule, change the chapter in docs/ and the files are rebuilt from it.

Claude Code plugin

The plugin bundles the skill and the MCP server, so one install gives an agent both the rules and a way to look up a chapter.

bash
/plugin marketplace add tomavic/angular-philosophies
/plugin install angular-philosophies@angular-philosophies

After that, an Angular task in any project can load the skill on its own. For "a new orders list page with filters" it points the agent at the store, data-access and testing chapters, and at the template files to copy: a view store with a resource and a mutation, provided from the component, and a spec written with the test toolkit.

Claude skill, without the plugin

Copy the folder into your skills directory. It is self-contained: the rules are in SKILL.md, the full chapters are in references/, and the code templates are in templates/.

bash
# for one project, from its root
mkdir -p .claude/skills && curl -sSL https://github.com/tomavic/angular-philosophies/archive/refs/heads/main.tar.gz | tar -xz --strip-components=3 -C .claude/skills angular-philosophies-main/ai/skills/angular-philosophies
# for every project: the same command with ~/.claude/skills in place of .claude/skills

AGENTS.md

ai/AGENTS.md is a drop-in file: the stack, a "read this first, by task" table and every rule, grouped by chapter. Put it at the root of your project, and in an area folder if only that area follows it.

bash
curl -o AGENTS.md https://raw.githubusercontent.com/tomavic/angular-philosophies/main/ai/AGENTS.md

Then add your project's own overrides under a heading of their own, above the generated rules. Where your project and this guide disagree, the project wins, and writing that down is how an agent learns which one to follow.

Cursor rules

Each ai/cursor/*.mdc file covers one chapter and is attached by file type: the store rules to **/*.store.ts, the component rules to **/*.component.ts and **/*.component.html, the testing rules to **/*.spec.ts, and so on. Copy them in:

bash
mkdir -p .cursor/rules && curl -sSL https://github.com/tomavic/angular-philosophies/archive/refs/heads/main.tar.gz | tar -xz --strip-components=3 -C .cursor/rules angular-philosophies-main/ai/cursor

llms.txt

llms.txt is an index of the chapters with a line about each, and llms-full.txt is every chapter in one file, with the rules and the code. Give a tool the URL, or paste the full file into a chat with a model that has no file access.

ts
https://raw.githubusercontent.com/tomavic/angular-philosophies/main/llms.txt
https://raw.githubusercontent.com/tomavic/angular-philosophies/main/ai/llms-full.txt

MCP server

ai/mcp/server.mjs is a small stdio server with no dependencies. It serves the chapters on demand, so an agent pulls only the one it needs instead of carrying everything in context.

ToolReturns
list_topicsEvery chapter with its title and one-line summary
get_topicA whole chapter, by name or number
get_rulesThe rules for one chapter, or all of them
searchThe chapters and rules that mention a word or phrase

Add it to a client with Node 22 or later:

bash
claude mcp add angular-philosophies -- node /path/to/angular-philosophies/ai/mcp/server.mjs

or in any client's JSON configuration:

json
{
  "mcpServers": {
    "angular-philosophies": { "command": "node", "args": ["/path/to/angular-philosophies/ai/mcp/server.mjs"] }
  }
}

The plugin registers the same server automatically.

Writing instructions for agents that work

What made these files effective, if you write your own:

  • Rules, not essays. One imperative sentence each, with the word "never" where it means never.
  • A map by task. "Adding a store: read chapter 3, copy order-list.store.ts." Agents follow a pointer better than a paragraph.
  • A reference implementation for each pattern. An agent copies a real file more reliably than it follows a description.
  • Say where your rules override the framework's defaults. Otherwise the agent applies the generic best practice it was trained on.
  • Make the rules checkable. A rule enforced by ESLint or the compiler holds even when the instructions are ignored.

Rules

These are the lines your AI tool receives. Each one is a sentence on its own.

  1. Treat the generated AI files (everything in ai/ except ai/mcp/server.mjs, plus llms.txt) as build output; change the chapter in docs/ instead.
  2. Put AGENTS.md at the project root and add project-specific overrides above the generated rules; the project wins over the guide.
  3. Keep agent instructions as one-sentence imperative rules with a task-to-chapter map and a reference file per pattern.
  4. State where project rules override framework defaults, and back rules with lint rules or types where possible.