AgentixTech
Developer onboarding · dev environment

Set up your machine with one command.

Your Claude installs the toolchain, configures itself, adds the AgentixTech engineering workflow, and builds a cached code map — then verifies the whole thing. You approve one sign-in. Roughly 15 minutes, mostly hands-off.

Linux · macOS · Windows ~15 min ~80% unattended idempotent · safe to re-run
bash <(curl -fsSL https://onboarddev.agentixtech.ai/setup.sh)

Paste into Terminal. Installs Claude Code, then hands off to it. Re-run anytime — it skips what's done.

01

How it works

STEP 1

Run one command

Copy the line for your OS and paste it into your terminal. That's the only thing you type.

STEP 2

Claude sets up

It installs the toolchain, configures itself, and adds the engineering suite — following a hosted runbook, step by step.

STEP 3

You approve sign-in

The one manual moment: approve GitHub (and SSO) in your browser. Nothing sensitive is stored on this page.

STEP 4

Verified

A deterministic check runs and prints a pass/fail report. If anything's off, re-run — it resumes safely.

02

What gets installed

  • Toolchain — git, mise, Node 20 LTS, Python 3.12, GitHub CLIauto
  • Claude Code — the CLI plus a sensible CLAUDE.md and permission defaultsauto
  • Engineering suite — 5 workflow skills + graphify, active in every repoauto
  • Self-heal hook — keeps the suite current and offers onboarding onceauto
  • GitHub sign-in — device login in your browserneeds you
  • SSO / cloud — approve when your role needs itneeds you
  • Verification — a no-LLM check confirms every tool is present and correctauto
  • Repo map — graphify indexes your active project (see below)auto
03

graphify — why your Claude stays fast and cheap

On a large codebase, an AI assistant burns time and tokens re-reading files. graphify builds a small, ranked REPO_MAP — the important files and what they contain — and keeps it in the cached context. Claude reads the map, opens the two or three files that matter, and skips the rest.

1 · SCAN

Parse the repo, extract symbols, build a reference graph across files.

2 · RANK

Score files by how much the rest of the code depends on them.

3 · CACHE

Write a token-budgeted map Claude reads first — instead of the whole tree.

~0.1×
cost of a cached read vs a fresh one
<150ms
to map a mid-size repo
0 deps
pure standard library, every OS
04

Safety & consent

  • No secrets on this page. Sign-in happens in your browser, scoped to you.
  • Your config is merged, never clobbered. Existing CLAUDE.md / settings are preserved.
  • Sudo & auth are gated. Anything privileged pauses for your approval.
  • Idempotent. Re-running skips completed steps and self-heals.
  • Verifiable. A deterministic (non-AI) check is the source of truth for "done".
  • Readable. Inspect it first — SETUP.md, setup.sh, ONBOARDING.md.