interview-me

Étoiles3

Forks0

Mis à jour1 juillet 2026 à 17:30

Interview the user until you have 100% confidence about what they actually want.

Installation

Installer avec Codex ou Claude Copiez ce prompt, collez-le dans Codex, Claude ou un autre assistant, puis laissez-le vérifier la page du skill et l'installer pour vous.

Exécuter dans Manus

Source

kapral18

kapral18/dotfiles

Ouvrir le dépôt GitHub Voir les dépôts du créateur

Téléchargement

Exécuter dans Manus

Métiers associésSOC

Basé sur la classification professionnelle SOC

Développeurs de logicielsProfessions informatiques et mathématiques·SOC 15-1252

SKILL.md

readonly

name	interview-me
description	Interview the user until you have 100% confidence about what they actually want.
disable-model-invocation	true

Interview Me

Interview me until you have 100% confidence about what I actually want, not what I think I should want.

The SOP (§3.0 Intent Loop) already owns the mechanics: investigate read-only first, ask exactly one fork-closing question at a time, wait, update the spec, repeat. This skill sharpens how you run that loop.

Answer it yourself before asking. Every question I could resolve by reading the code, running a probe, or checking history is a question you should not ask me. Ask only what evidence cannot settle.
Walk the decision tree, resolving dependencies in order. Later choices often depend on earlier answers; do not ask a question whose relevance a prior answer would eliminate. Ask the most branch-eliminating one first, then descend.
Give me your recommended answer with each question. State the option you would pick and why, so I can confirm with one word instead of composing a reply. If you have a real default, lead with it; if the fork is genuinely open, say so.
Stop when the forks are empty. Done means every remaining interpretation produces the same output and the success criteria are testable — not when I stop objecting.

Plus depuis ce dépôt

même dépôt

ralph

kapral18/dotfiles

Use when running autonomous verified tasks or spawning, steering, inspecting, resuming, or replanning Ralph multi-agent runs via ,ralph go, tmux dashboards, or role attach.

2026-07-103

improve-branch

kapral18/dotfiles

Manual-only workflow for proposing one evidence-backed, high-leverage improvement to the current branch, PR, or issue goal.

2026-07-013

improve-codebase

kapral18/dotfiles

Manual-only workflow for proposing one evidence-backed, high-leverage improvement to the current codebase.

2026-07-013

improve-local

kapral18/dotfiles

Manual-only workflow for proposing one evidence-backed, high-leverage improvement to the local changes.

2026-07-013

improve-targeted

kapral18/dotfiles

Manual-only workflow for proposing one evidence-backed, high-leverage improvement to a targeted part of the codebase.

2026-07-013

name	interview-me
description	Interview the user until you have 100% confidence about what they actually want.
disable-model-invocation	true

Interview Me

Interview me until you have 100% confidence about what I actually want, not what I think I should want.

Answer it yourself before asking. Every question I could resolve by reading the code, running a probe, or checking history is a question you should not ask me. Ask only what evidence cannot settle.
Walk the decision tree, resolving dependencies in order. Later choices often depend on earlier answers; do not ask a question whose relevance a prior answer would eliminate. Ask the most branch-eliminating one first, then descend.
Give me your recommended answer with each question. State the option you would pick and why, so I can confirm with one word instead of composing a reply. If you have a real default, lead with it; if the fork is genuinely open, say so.
Stop when the forks are empty. Done means every remaining interpretation produces the same output and the success criteria are testable — not when I stop objecting.