About clarify
clarify is an agent skill for turning an invoking prompt into an actionable request. It uses conversation and evidence to avoid unnecessary questions, and asks only about unresolved choices that could materially affect the outcome.
Key features
- 01Uses conversation and available evidence
- 02Asks only consequential questions
- 03Provides context with grouped questions
- 04Continues work not blocked by answers
Use cases
- 01Resolve scope or success criteria
- 02Clarify externally visible decisions
- 03Proceed without asking about routine choices
Install
Copy the command and run it in a terminal.
npx skills add https:/ / github.com/ sjunepark/ agent-scripts --skill clarifyAfter installing, type/clarifyin Claude Code to confirm it loaded.
Bundled filesagents/openai.yaml
Details
From the project READMEClarify
Make the request in the invoking prompt clear enough to carry out. Use earlier conversation and available evidence to interpret it and avoid repeated questions; do not import unrelated open decisions into the agenda.
Ask about missing information only when it could materially change the outcome, scope, constraints, or success criteria and cannot be resolved from evidence or delegated judgment. Investigate facts yourself. Leave routine implementation choices to the implementer; surface technical choices when their consequences are externally visible, costly to reverse, or reserved for the user.
Give the context needed to answer, then ask a small group of related questions. Use concrete alternatives when they help the user recognize a preference, and a structured question tool when available and appropriate. Explain consequential blind spots without turning them into a checklist of hypothetical concerns.
Stop asking once the authorized task is clear enough to execute. A well-specified request needs no question. Briefly state the recovered intent and any material assumptions; ask for correction only when a consequential assumption remains unresolved. Clear answers settle their items without another confirmation round.
Continue the original authorized task, holding only work that depends on an unresolved answer. When clarification alone was requested, the readback is the deliverable.