interview

Estrelas8

Forks0

Atualizado31 de maio de 2026 às 19:37

One-question-at-a-time interviewing discipline that drives /bootstrap and /intake to ~95% confidence before writing the wiki or an item card. Use when gathering requirements for a new project or fleshing out a vague work item.

Instalação

Instalar com Codex ou Claude Copie este prompt, cole no Codex, Claude ou outro assistente e deixe que ele revise a página da skill e instale para você.

Fonte

dxlbnl

dxlbnl/zod4-mock

Abrir repositório GitHub Ver repositórios do creator

Download

Ocupações relacionadasSOC

Baseado na classificação ocupacional SOC

Especialistas em gestão de projetosNegócios e Operações Financeiras·SOC 13-1082

SKILL.md

readonly

name	interview
description	One-question-at-a-time interviewing discipline that drives /bootstrap and /intake to ~95% confidence before writing the wiki or an item card. Use when gathering requirements for a new project or fleshing out a vague work item.
disable-model-invocation	false

Interview discipline

Reach ~95% confidence that you can write the artifact well — the bootstrap wiki, or a backlog item card — before you write it. A question is far cheaper than building the wrong thing; a vague artifact costs far more downstream.

The loop

One adaptive question at a time. When the next question depends on the last answer, ask a single question (via AskUserQuestion) and let the answer shape what you ask next. Batch only genuinely independent questions into one call.
Infer and confirm — don't ask cold. Propose a sensible default ("sounds like a TypeScript + React SPA — right?") and let the user correct it, rather than an open-ended prompt. Put the recommended option first.
Track confidence out loud. After each answer, restate your current understanding in one line, then ask the single highest-value remaining question. Stop when you are ~95% sure you can write the artifact well — not when you run out of things you could ask.
Short-circuit is always available. The user can say "that's enough, proceed" at any point. Record the remaining unknowns (bootstrap → the relevant wiki page's open questions; intake → the card's ## Notes) instead of blocking.
Cap the questions. Ask at most ~6 before you must summarize and proceed (or explicitly ask permission to keep going). Never interrogate.
Summarize before acting. Present the final understanding and get a yes before you write the artifact.

Red flags

You asked something you could have inferred and offered for confirmation.
You kept asking after you could already write a good artifact (past ~95%).
You batched questions whose answers actually depend on each other.
You blocked on a question the user was happy to defer.

Where this is used

/bootstrap Phase 1 — converge on project / stack / package-manager / testing / constraints / initial backlog, question-by-question.
/intake — for a vague item, ask the one or two questions that make it actionable; a clearly-specified ask files in one shot (don't add friction where there's none).

Mais deste repositório

mesmo repositório

bootstrap

dxlbnl/zod4-mock

Primary entry point for a freshly cloned Vibin seed repo. Interviews the user about the project, stack, package manager, constraints, and initial backlog, populates the wiki/ starter pages, scaffolds the chosen stack, writes the stack-specific permission profile, and hands off to the manager skill. Use when starting a new project, or when the user says "bootstrap", "set up the project", or "let's begin".

2026-05-318

intake

dxlbnl/zod4-mock

File a new work item into wiki/backlog/inbox/. Use whenever the user reports a bug, asks for a feature, or proposes a change of direction mid-run — instead of inline-patching. Any agent (including subagents) can invoke this. Triggered by /intake "<title>" or when an agent recognises new work that should be captured.

2026-05-318

manager

dxlbnl/zod4-mock

Orchestrates Vibin's wiki-driven, spec-driven, test-first build pipeline. The top-level session runs this skill to order the backlog and drive the spec-writer -> test-writer -> implementer -> reviewer pipeline (or the bug / research / chore tracks), one item at a time. Use after /bootstrap, or to resume building the backlog.

2026-05-318

migrate-vibin

dxlbnl/zod4-mock

Bring an existing Vibin project up to date with the latest seed by diffing the project's recorded seed commit against the latest Vibin on GitHub and applying the changes. Use when the user says "migrate", "migrate-vibin", "upgrade Vibin", "apply the latest seed changes", or after a new seed release. Reads the seed commit hash from .vibin-version, audits the project's actual state, applies the migrations added since, and reconciles seed-owned files while preserving local customizations.

2026-05-318

tdd-cycle

dxlbnl/zod4-mock

Reference for the red-green-refactor test-first discipline used by the test-writer and implementer agents. Use when you need the canonical explanation of how Vibin's tests-first workflow works, or to settle a question about test ordering, "minimum" code, or trivially-passing tests.

2026-05-318

status

dxlbnl/zod4-mock

One-screen summary of where the manager pipeline is — lane counts, the active item with its current stage, the last escalation if any, and the next three ready items. Read-only. Use any time you want to know what's happening without disturbing a running pipeline.

2026-05-278

name	interview
description	One-question-at-a-time interviewing discipline that drives /bootstrap and /intake to ~95% confidence before writing the wiki or an item card. Use when gathering requirements for a new project or fleshing out a vague work item.
disable-model-invocation	false

Interview discipline

The loop

One adaptive question at a time. When the next question depends on the last answer, ask a single question (via AskUserQuestion) and let the answer shape what you ask next. Batch only genuinely independent questions into one call.
Infer and confirm — don't ask cold. Propose a sensible default ("sounds like a TypeScript + React SPA — right?") and let the user correct it, rather than an open-ended prompt. Put the recommended option first.
Track confidence out loud. After each answer, restate your current understanding in one line, then ask the single highest-value remaining question. Stop when you are ~95% sure you can write the artifact well — not when you run out of things you could ask.
Short-circuit is always available. The user can say "that's enough, proceed" at any point. Record the remaining unknowns (bootstrap → the relevant wiki page's open questions; intake → the card's ## Notes) instead of blocking.
Cap the questions. Ask at most ~6 before you must summarize and proceed (or explicitly ask permission to keep going). Never interrogate.
Summarize before acting. Present the final understanding and get a yes before you write the artifact.

Red flags

You asked something you could have inferred and offered for confirmation.
You kept asking after you could already write a good artifact (past ~95%).
You batched questions whose answers actually depend on each other.
You blocked on a question the user was happy to defer.

Where this is used

/bootstrap Phase 1 — converge on project / stack / package-manager / testing / constraints / initial backlog, question-by-question.
/intake — for a vague item, ask the one or two questions that make it actionable; a clearly-specified ask files in one shot (don't add friction where there's none).