Architecture

agent-rules-kit is content + a tiny composer. The rules live as markdown packs; the CLI selects packs, composes them into one document, and writes it to the target file.

flowchart LR
  R["rules/*.md<br/>(core, testing, security, git, refactoring)"] --> C["composeRules(ids)"]
  SEL["--packs core,testing"] --> C
  C --> DOC["single markdown"]
  T["--target agents|claude|cursor"] --> W["write file"]
  DOC --> W
  W --> OUT["AGENTS.md / CLAUDE.md / .cursorrules"]

Module map

Module	Responsibility
rules/*.md	The rule packs themselves — reviewable markdown
packs.ts	Pack registry + loader that reads the markdown bodies
compose.ts	Resolve pack ids (incl. all) and compose the document
targets.ts	Map a target to its output filename
cli.ts	Argument parsing, init / list commands

Design principles

• Content first — the value is the rules; the code is a thin composer around them.
• Reviewable — packs are plain markdown, so improvements are normal PRs.
• No lock-in — output is a plain file your assistant already reads; regenerate any time.
• Tiny & dependency-free — pure functions, testable by reading the shipped rules/.