clarify
/clarify
Use before interpreting supplied data or reporting metrics, averages, rates, growth or comparisons. Resolve material ambiguity through the host's interactive question tool before choosing a result; showing alternatives does not settle the choice. Also use for ambiguous factual requests or deliverables and explicit requests to clarify assumptions. Reuse settled definitions and collect only missing inputs.
- Author
- warwick-bit ↗
- Repository
- llm-accuracy ↗
- Licence
- MIT
- Last repository change
- 27.10.2026
Assessment and security
- Files in this package
- 2
- Scripts
- 0
- HTML comments
- 0
- URLs in SKILL.md
- 0
No signals detected. Repository changed on 27.10.2026, without references to retired models or APIs.
MIT, open: use and modify while retaining author attribution.
What it does and when to use it
Use before interpreting supplied data or reporting metrics, averages, rates, growth or comparisons. Resolve material ambiguity through the host's interactive question tool before choosing a result; showing alternatives does not settle the choice. Also use for ambiguous factual requests or deliverables and explicit requests to clarify assumptions. Reuse settled definitions and collect only missing inputs.
Original author description
Use before interpreting supplied data or reporting metrics, averages, rates, growth or comparisons. Resolve material ambiguity through the host's interactive question tool before choosing a result; showing alternatives does not settle the choice. Also use for ambiguous factual requests or deliverables and explicit requests to clarify assumptions. Reuse settled definitions and collect only missing inputs.
How it works
SKILL.md has no level-two headings · 94 lines, 949 words
SKILL.md preview
--- name: clarify description: Use before interpreting supplied data or reporting metrics, averages, rates, growth or comparisons. Resolve material ambiguity through the host's interactive question tool before choosing a result; showing alternatives does not settle the choice. Also use for ambiguous factual requests or deliverables and explicit requests to clarify assumptions. Reuse settled definitions and collect only missing inputs. --- # Clarify Resolve consequential ambiguity before relying on an assumed answer. This is an advisory workflow, not a guarantee that all assumptions or errors are found. First inspect the request, earlier answers, available sources and any approved definitions. Do not ask the user to repeat a settled choice or supply information you can obtain with available, authorized read-only tools. Data can reveal fields and alternative results; it cannot decide which business meaning the user intends. For data work, check only unresolved choices that affect the requested result: metric and units, population and exclusions, grain or duplicate handling, denominator, dates and time zone, comparison, source/version and success criterion. If useful and feasible, calculate alternatives from available evidence to explain the effect of the choice. Do not invent figures or require a full analysis before asking a simple defining question. When the user asks for one result and an unresolved definition changes that result, ask before choosing a headline answer. A familiar convention or a plausible use for each alternative does not establish the user's intended meaning. Keep definition choices neutral unless their stated goal supports a recommendation. Calculating or showing both interpretations does not settle the user's intended meaning; use those alternatives to help frame the question. If the user requested both interpretations or explicitly authorized assumptions, follow that request without an extra definition question. When a question is needed, make an actual call to the host's interactive question tool before ending the turn. Writing options in a chat message is not a tool call. Check the tools available in this conversation and their actual schemas and mode restrictions; do not infer availability from the app name. The tool may be named `AskUserQuestion`, `request_user_input`, `request_user_input_async`, or have another name or namespace. Use the available question capability even when its name differs from these examples. In a Codex CLI terminal, prefer `request_user_input` for the native selectable menu when permitted. In other clients, prefer an asynchronous question when the client supports rendering its interactive choices, so independent work can continue. Reading this skill or describing the intended question does not complete this step. A listed tool or an `accepted` result does not prove that a selector appeared. If the user reports no selector, try another available, permitted question tool; do not repeat the failed route. Do not change modes, install an integration or invent a tool to ask. For a few meaningful alternatives, offer concise, distinct multiple-choice answers with a brief supported consequence for each. Describe category choices by their inclusion rule; do not invent lists of members or exclusions from a category name. Use examples only when their membership is established. Put a recommendation first only when the user's goal or evidence supports it; label and explain that recommendation. Allow a free-text alternative through the host's built-in facility. If the choice is open-ended, use a free-text question if the tool supports one. When a request needs both a definition choice and a missing value, keep the definition as selectable options and collect the actual value through a suitable free-text input or a follow-up. If the tool requires multiple choices but only an amount, date or other open value is missing, ask a short chat question instead of inventing choices, ranges or an "I'll type it" option. Such a selection is not the missing value. A partial answer settles only what the user actually supplied; retain that part and ask only for what is still needed. Missing numbers do not turn a meaningful definition choice into an entirely prose question. Follow the host's question limits and avoid duplicating its built-in Other option. Only fall back to a short chat question when no suitable tool is available or permitted, or the available tools failed. Do not repeat a tool question in chat. Ask the smallest set of questions needed to remove the material ambiguity. Follow the user's configured language, tone, accessibility and question-format preferences, within the host tool's constraints. Otherwise use plain English: short questions, familiar words and choices that explain the practical difference. Avoid unexplained jargon and internal implementation details; explain a necessary technical term briefly. Preserve the intended meaning when simplifying a question. Wait for the answer before dependent calculations or conclusions; continue independent work where useful. A recommended or preselected option, elapsed time, an empty response or dismissal is not an answer. If the user explicitly asks you to proceed with assumptions, state the chosen assumptions and keep dependent results conditional. Missing source evidence still remains unavailable. Apply the answer to the task and retain its meaning in the conversation or existing task brief. Keep the selected measure, units, population and period in the result. Plain-English labels must preserve the calculation: a total is not a per-item figure without the relevant count. Apply the same evidence boundary to the entire answer and later follow-ups. When a total's composition is unknown, describe its scope using only the agreed category name: do not supply examples of presumed members or exclusions. A caveat is still a claim; do not predict unspecified adjustments or invent a cause to justify an authorized assumption. Once the needed answers arrive, give the requested result, calculation and material qualifications, then stop. Do not append a generic advice section. When interpretation or advice is requested, tie each recommendation to the stated purpose and available evidence. Do not present one measured quantity as another operational quantity without the required relationship and inputs. Omit additions that need missing context rather than asking questions solely to support optional commentary. Do not create a new persistent store. Revisit the choice only when the user changes it or new evidence creates a material conflict. If nothing material remains unresolved, proceed without a ceremonial confirmation.
Files in this package (2)
- SKILL.md 7 KB
- agents/openai.yaml 1 KB
Versions
| Repository | Score | Licence |
|---|---|---|
| This page · warwick-bit/llm-accuracy | 91 | MIT |
| ericrisco/rsc-harness | 100 | MIT |
| growupanand/ConvoForm | 100 | Apache-2.0 |
| mcouthon/agents | 100 | MIT |
| ramboz/jig | 100 | MIT |
| ramboz/jig | 100 | MIT |
| ramboz/jig | 100 | MIT |
| umputun/cc-thingz | 100 | MIT |
| ronakrupani/clarify | 100 | MIT |
Database snapshot: 13.11.2026 · package version 17313d28ee8a (SHA-256 digest of the entire package)
Downloads in the last 30 days: 0 · ranking
Report a problem with this skill: harmful content, incorrect descriptions or an author’s removal request.
Similar skills
- 100clarify-and-plan
Use for ambiguous, multi-step, or high-impact requests where the agent must clarify assumptions, compare 2-3 viable approaches, break work into phases, and get phase-one alignment before implementation. Trigger on vague "build", "refactor", "add feature", "figure out",…