Quickstart

June 15, 2026 · View on GitHub

Set up a big autonomous Claude Code run the right way: prepare a /goal file before you /clear, then run it in a fresh, full-context session.

goalify only authors the file — it does not run your task. You run it. The file deletes itself when the run finishes successfully.

0. Prerequisites

  • Claude Code, installed and working. (goalify is a Claude Code / Agent Skills skill.)
  • A task worth a full session — a refactor, a migration, a feature, an audit. For a one-line fix, just ask Claude to do it; you don't need a /goal file.

1. Install

Easiest is the plugin:

claude plugin marketplace add Aboudjem/10x
claude plugin install goalify@10x

Or drop in the skill manually:

git clone https://github.com/Aboudjem/goalify.git
mkdir -p ~/.claude/skills
cp -R goalify/skills/goalify ~/.claude/skills/goalify

This gives you /goalify, which authors the run file. You execute that file with Claude Code's built-in /goal <path> command (Claude Code 2.1.139+). Claude Code auto-discovers the skill. (A brand-new top-level skills directory may need a restart the first time before it is watched.) To update later, re-pull and re-copy; to remove, delete ~/.claude/skills/goalify.

2. Use it in 3 steps

  1. Ask in plain language: goalify this: <your big task> — e.g. "goalify this: migrate our Express API from callbacks to async/await and keep the tests green."
  2. Answer the short MCQ (only if goalify finds a genuine fork in scope/structure/risk). Then it writes one self-contained /goal file to an absolute path and prints two commands.
  3. Run the two commands it printed: 
    /clear
/goal <absolute-path-it-gave-you>
    A fresh session reads the file at full context, fans out its own agents, verifies, tests, and deletes the file when every success criterion is met.

That's it. Nothing in your repo changes during step 1 — goalify only researches and writes the plan. The work happens in the fresh /goal session, under the hard rules baked into the file (no hallucination, separate-agent verification, tests, gated destructive actions).

What you get

  • One self-contained file with a declarative goal, verified context (absolute paths), your locked decisions, dependency-ordered phases with fan-out guardrails, machine-checkable success criteria, a progress checklist, and a gated self-destruct.
  • A fresh-context run — the execute session starts at 100% context, not the dregs of a long chat.
  • Zero residue — the file removes itself on success (and survives to resume if not).

See a real example: examples/sample-goal-file.md.

When NOT to use it

  • A trivial change — just ask Claude to do it.
  • Open-ended exploration with no definable end state — goalify will decline; explore interactively instead.
  • Work you want done right now in this session — use autopilot / ultrawork / ralph.