Skip to content
XADEN

Developer Guide

Kick off with your assistant

Turn your idea into approved requirements, a chosen stack and a first runnable skeleton. You answer questions; the assistant writes things down and you approve them.

Before you start

  • A project that has Xaden: created or adopted. You are in its folder.
  • Your assistant is installed and can read your files and run shell commands. Xaden does not support chat windows without file access.

Start the kickoff

Open your assistant in the project folder. Start it with:

/kickoff <what you want to build>
$kickoff <what you want to build>

This needs the command files of ChatGPT Codex in the project: create it with xaden create --assistants codex, or run xaden adopt in an existing project.

/kickoff <what you want to build>

This needs the command files of GitHub Copilot in the project: create it with xaden create --assistants copilot, or run xaden adopt in an existing project.

/kickoff <what you want to build>

This needs the command files of Mistral Vibe in the project: create it with xaden create --assistants vibe, or run xaden adopt in an existing project.

Run `xaden prompt kickoff` and follow it. My idea: <what you want to build>

Claude Code finds the /kickoff command in .claude/commands/, which xaden create and xaden adopt write for Claude Code users. GitHub Copilot, Mistral Vibe and Codex CLI users get the same command in their own place when --assistants names them or xaden adopt finds them (see Add Xaden to a repository).

For an existing repository, describe what the project already does instead of a new idea. The assistant asks questions and writes the requirements, the threat model and the architecture notes. Review what it writes, because it describes code you know better than it does.

The assistant’s first reply must start with CONTEXT CHECK <number>. Compare the number with the one xaden context prints in your terminal. A mismatch means the assistant started from stale or missing files. A match is a consistency aid only: it does not prove the rules were read or followed, so review its work.

What you decide

  1. The users. The assistant drafts three to five personas from your answers. Every later solution is checked against them.
  2. The requirements. One file per requirement, grouped in epics. Each says what the user wants to do and how to tell it works. You set the ones you approve to approved; nothing is built for a requirement that is still a draft.
  3. The threat model. What you protect, who could attack it, where untrusted input enters, and which security level you aim for. It goes into xaden/threat-model.md; the assistant also fills the TODO markers of SECURITY.md with facts you give it.
  4. The stack. The assistant records the decision, builds a minimal skeleton that runs, and adds Xaden’s preset for the stack (xaden preset python or xaden preset node), which also maps the checks. You install the tools it lists.

Check that it worked: xaden check runs the new checks and ends with a summary. A skipped check is named; install the tool and run it again.

Next

Build features.

These pages describe Xaden v0.3.0, released 2026-10-17.