Rechercher

PythonReactSEOPDF
GitHub Skill

ask-questions-if-underspecified

Clarify requirements before implementing changes or plans when a request is underspecified. Use when objectives, scope, constraints, acceptance criteria, or environment details are unclear and you need the minimum set of questions before proceeding.

graysurf☆ 2 Étoiles du dépôt0 Forks du dépôt

Installation

Suivez les instructions du dépôt pour votre agent. Consultez les fichiers et dépendances avant l’installation.

Voir la source ↗Guides

Instructions

Ask Questions If Underspecified

Contract

Prereqs:

  • User request is missing must-have details (objective/scope/constraints/done criteria) and needs clarification before proceeding.
  • User is available to answer questions or explicitly approve stated assumptions.

Inputs:

  • User request + any provided context (codebase, environment, constraints, examples).
  • Optional user preference for defaults vs custom answers (e.g., "defaults").

Outputs:

  • 1-5 numbered "Need to know" questions with short options and an explicit default.
  • If the user asks to proceed without answers: a short numbered assumptions list to confirm before starting work.

Exit codes:

  • N/A (conversation workflow; no repo scripts)

Failure modes:

  • User cannot provide required answers and will not approve assumptions (must stop; do not implement).
  • Constraints conflict or request stays ambiguous after Q/A (must clarify before proceeding).

Goal

Ask the minimum set of clarifying questions needed to avoid wrong work. Do not begin implementation until must-have answers are provided or the user explicitly approves stated assumptions.

Workflow

  1. Decide whether the request is underspecified

    • Check objective (what should change vs stay the same)
    • Check done criteria (acceptance, examples, edge cases)
    • Check scope (files/components/users in or out)
    • Check constraints (compatibility, performance, style, deps, time)
    • Check environment (language/runtime versions, OS, test runner)
    • Check safety/reversibility (migrations, rollout, risk)
  2. Ask must-have questions first

    • Ask 1 to 5 questions that remove whole branches of work
    • Use numbered questions with short options (a/b/c)
    • Provide a clear default (bold it) and a fast path reply like "defaults"
    • Allow "not sure - use default" when helpful
    • Separate "Need to know" from "Nice to know" only if it reduces friction
  3. Pause before acting

    • Do not run commands, edit files, or make a detailed plan that depends on missing info
    • A low-risk discovery read is allowed if it does not commit to a direction
  4. Confirm and proceed

    • Restate requirements and success criteria in 1 to 3 sentences
    • Begin work only after answers or explicit approval of assumptions

Response format

Use a compact, scannable structure. Example:

Need to know
1) Scope?
   a) **Minimal change**
   b) Refactor while touching the area
   c) Not sure - use default
2) Compatibility target?
   a) **Current project defaults**
   b) Also support older versions: <specify>
   c) Not sure - use default
Reply with: defaults (or 1a 2a)

If asked to proceed without answers

  • State assumptions as a short numbered list
  • Ask for confirmation
  • Proceed only after confirmation or corrections

Exemples

Consultez les exemples du dépôt source.

Voir la source ↗

Historique source

Date d’indexation
2026-01-20
Mise à jour de l’index
2026-02-02
Branch
main

Les dates concernent ce répertoire. Consultez le dépôt pour l’historique du code.

GitHub ↗

chat Comments (0)

chat_bubble_outline

No comments yet. Be the first to share your thoughts!