Ctxinit — Security skill for Claude Code

ctxinit

Initializes a project's engineering foundation: a single-source **`context/` knowledge hub** (binding rules, security law, component-architecture law, docs, plans, an ADR decision log, a technical-debt registry, a repo map) plus thin agent entry-point files (`CLAUDE.md`, `AGENTS.md`, `.cursor/rules/*.mdc`) that **reference** the hub instead of copying it. It also owns that decision log going forward — recording, querying, and superseding ADRs through a deterministic, dependency- free CLI — so the log never drifts and no decision silently contradicts an earlier one.

Modeled on a knowledge-hub design proven on a real production codebase, genericized for any project.

Why

Most projects end up with the same rules explained three different ways: once in `CLAUDE.md`, once in `.cursorrules`, once in someone's head. They drift. ctxinit sets up **one canonical source** (`context/rules/`) and makes every AI coding tool point at it instead of holding its own copy — so editing a rule once updates it everywhere, for every agent, forever.

What it works with

The `context/` hub itself is plain markdown — no tool lock-in. ctxinit generates a dedicated, correctly-shaped entry point for each of the three major AI coding tools, all pointing at the same rules:

Tool	Entry point ctxinit generates	How that tool uses it
Claude Code	CLAUDE.md	Auto-loaded every session; @context/rules/*.md imports pull the rules into context.
Codex (and other AGENTS.md readers)	AGENTS.md	Read at session start; references context/rules/ by path.
Cursor	.cursor/rules/context.mdc	A thin stub (alwaysApply: true) whose only job is to point at context/rules/ — Cursor needs a file to glob, so this exists to satisfy that, not to duplicate content.

Running `ctxinit`/`ctxinit --sync` *as a slash command* is native to Claude Code (it's a Claude Code Skill — see below). In Codex or Cursor you can still get the same resu