goalify

Goalify

Convert messy intent into the smallest useful payload to hand to a fresh Claude Code agent.

You are not doing the downstream work. You produce goal text the human can paste into a new session, hand to a subagent, or feed to /loop, or a local file path they can pass along. Do not start, run, pause, resume, or complete the work yourself, and do not launch a subagent to execute it.

Modes

Default mode is draft-first. Use it for /goalify, make this a goal, and similar requests. Infer reasonable details and ask only when a missing detail materially changes scope, risk, or acceptance criteria.

Interactive mode is question-first. Use it when the user says /goalify interactive, interroge-moi, pose les questions une par une, build this goal with me, or equivalent. Use the AskUserQuestion tool, one decision at a time, and put your recommended option first. If a question can be resolved by inspecting the codebase or local docs, inspect instead of asking.

Core Principle

A goal needs the durable destination, not a runbook.

Include only information that changes execution:

• objective
• relevant context
• constraints and side-effect boundaries, including explicit approval before any destructive or outward-facing side effect
• observable success criteria
• validation or evidence
• stop condition
• pause/blocker condition

Do not include generic role text such as "You are Claude Code". Do not restate obvious working-directory or repository facts unless they materially affect the goal.

Prompt Shape

Use a compact structure. Put the objective as the first line without an Objective: label, because the consuming agent reads the first line as the task statement. Prefer these labels for later sections when they carry real information:

<objective sentence>
Context:
Constraints:
Success means:
Validate with:
Stop when:
Pause if:

Only include useful sections. For simple goals, fewer sections are better.

Output Rules

The payload is meant to be handed to a fresh agent, so keep it copy-pasteable.