Decide whether a request is clear enough to act on, and if not, ask only the minimum necessary clarification questions before starting work.
SkillScore68Maintained
Automatic
◆Product Management pack
Last verified 8/3/2026
What it changes
Passed 15 of 22 test cases with the skill, and 3 of 22 without it.
Without skill14%
With skill68%
Checks it passed that the model alone missed
Limits the total number of clarifying questions asked to 3 or fewer.
States explicit assumptions for reversible minor details like button layout defaults.
13 fixed2 passed either way6 still failing1 broke
Pass-rate lift
+55 pp
Tokens
+0%
Agent turns
0%
Response time
-54%
Model gemini-3.6-flashMeasured 8/3/2026v115 of 22 cases passed with skill
Tokens, turns and time describe resource use, not a dollar price.
See it in action
One of 13 test cases the skill fixed. Same model, same prompt, with and without the skill.
1 of 13
Prompt
We want to replace our polling architecture with Socket.io WebSockets for live chat messaging. Message persistence and connection drop reconnection buffering are unstated. State safe assumptions for unread badge colors, and ask the core questions required for real-time state management.
What was checked, without and with the skill
Checked
Without skill
With skill
Limits the total number of clarifying questions asked to 3 or fewer.
Missed
Passed
Asks targeted high-impact questions about offline message buffering behavior.
Passed
Passed
States explicit assumptions for unread notification badge color defaults.
Passed
Passed
case-20 · gemini-3.6-flash · 8/3/2026 · answers are capped and redacted
What's inside
The instructions your model reads when the skill loads.
SKILL.md63 linessingle file
Clarify the Request Skill
Purpose
Ensure work starts on the right understanding by asking only the questions that would materially change the outcome — and making safe, stated assumptions for everything else.
When to Use
The request is ambiguous and multiple reasonable interpretations would lead to different results.
The task is high-impact, irreversible, or affects many files (architecture, auth, schema, deployment).
Required information (e.g. which data source, which design direction, which user role) is missing and can't be inferred from the project.
The user's goal isn't clear enough to plan correctly.
When Not to Use
The task is simple, low-risk, and a reasonable default exists (e.g. "fix this typo", "add a loading spinner to this button").
The project conventions already answer the ambiguity (e.g. styling approach is obvious from existing code).
Clarifying would only confirm something that's already inferable at low risk.
Required Inputs
The user's request as given.
Context already gathered from the Understand the Project Skill (if applicable).
Assumptions
The user prefers progress over interrogation; minimize friction.
Low-risk ambiguities should be resolved with a stated assumption, not a question.
High-risk ambiguities (data loss, security, architecture-wide changes) must always be clarified.
Step-by-Step Workflow
List open questions: identify every point where the request could be interpreted in more than one way.
Classify each open question as:
High impact — affects correctness, safety, architecture, or is hard to reverse.
Low impact — affects style/detail only, easy to change later.
Resolve low-impact questions yourself by choosing the most likely interpretation based on project conventions and stating the assumption briefly in your response.
For high-impact questions, draft the minimum set of questions needed (ideally 1, rarely more than 3).
Combine related questions into a single message rather than asking one at a time.
Ask the questions, then wait for the answer before proceeding with the affected parts of the task. Unaffected parts of the task can proceed in parallel if safe.
Record assumptions made for low-impact items so they can be reported later via the Global Agent Output and Reporting Skill.
Rules and Best Practices
Ask only questions that would materially change the result.
Prefer one well-formed question over several scattered ones.
State assumptions explicitly and briefly rather than silently picking one.