Install any skill in seconds. Free to start, no credit card required.
Get Started Free →Structured clarification form for unresolved material requirements.
.claude/skills/nexu-io-discovery-question-form/SKILL.md| Test case | Without → With | Effect | Δ tokens | Δ turns |
|---|---|---|---|---|
| case-02 | ✗→✓ | ▲ Improved | -68% | 0% |
| case-01 | ✗→✓ | ▲ Improved | -15% | 0% |
| case-03 | ✗→✓ | ▲ Improved | 15% | 0% |
| case-09 | ✗→✓ | ▲ Improved | 58% | 0% |
| case-10 | ✗→✓ | ▲ Improved | -20% | 0% |
This atom defines the <question-form> protocol. It does not decide whether clarification is required. Follow the active skill and core prompt's requirements-clarification policy. When they identify unresolved information that would materially change the design direction, content structure, or delivery format, surface the smallest possible set of questions that unblocks the workflow.
The questions are rendered as a <question-form> artifact inline in the originating assistant message. This is assistant text parsed by the host, not a plugin GenUI surface or a native tool call. Submitted answers return as the next user message, beginning with [form answers — <form-id>].
discovery pipeline stage only makes this protocol available; declaringor entering the stage does not trigger a form.
conversation, plugin inputs, memory, active skill, and design system.
Emit the form as a question-form block whose body is a JSON object with a top-level questions array. Do not emit a bare question object by itself; the renderer only recognizes the wrapped form contract.
html<question-form id="discovery" title="Quick brief — 30 seconds"> { "description": "I'll lock these in before building. Skip what doesn't apply — I'll fill defaults.", "questions": [ { "id": "audience", "label": "Who's the primary audience?", "type": "checkbox", "options": ["VC", "Customer", "Internal team"], "maxSelections": 2, "required": true } ] } </question-form>
Each entry in the top-level questions array uses:
id: stable answer key, for example audience.label: user-facing question copy.type: one of radio, checkbox, select, text, textarea,number, range, date, time, datetime-local, color, url, email, tel, file, switch, or direction-cards.
options: required for choice controls; strings are allowed, or objects withlocalized label and stable value.
allowCustom: leave unset or set to true for finite-choice controls sousers can type their own answer instead of accepting only generated options. Set allowCustom: false only when the downstream system needs an exact machine id.
customLabel / customPlaceholder: optional localized copy for that customanswer input.
maxSelections: include this for checkbox controls with a limited selectioncount.
required: set to true only when the answer is needed before work cancontinue.
The discovery atom completes when the next user message contains an answer for every required question. Treat those submitted answers as conversation context and do not ask the same questions again unless later input invalidates an answer.
| Case | Status | Duration (ms) | Turns | Tokens | Tool calls | ||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| Without | With | Δ | Without | With | Δ | Without | With | Δ | Without | With | Δ | ||
case-02 | fail→pass | 37,177 | 11,883 | -68% | 1 | 1 | 0% | 8,237 | 2,605 | -68% | 0 | 0 | — |
case-01 | fail→pass | 22,655 | 14,739 | -35% | 1 | 1 | 0% | 3,493 | 2,974 | -15% | 0 | 0 | — |
case-03 | fail→pass | 62,826 | 14,912 | -76% | 1 | 1 | 0% | 2,560 | 2,935 | +15% | 0 | 0 | — |
case-04 | pass→pass | 12,884 | 13,689 | +6% | 1 | 1 | 0% | 2,770 | 3,292 | +19% | 0 | 0 | — |
case-05 | pass→pass | 30,325 | 56,778 | +87% | 1 | 1 | 0% | 4,753 | 6,391 | +34% | 0 | 0 | — |
case-06 | pass→pass | 13,135 | 14,140 | +8% | 1 | 1 | 0% | 2,339 | 3,029 | +29% | 0 | 0 | — |
case-07 | pass→pass | 21,765 | 10,907 | -50% | 1 | 1 | 0% | 3,514 | 3,047 | -13% | 0 | 0 | — |
case-08 | pass→pass | 7,948 | 7,296 | -8% | 1 | 1 | 0% | 1,491 | 1,903 | +28% | 0 | 0 | — |
case-09 | fail→pass | 10,300 | 7,578 | -26% | 1 | 1 | 0% | 1,366 | 2,162 | +58% | 0 | 0 | — |
case-10 | fail→pass | 12,644 | 5,843 | -54% | 1 | 1 | 0% | 2,041 | 1,638 | -20% | 0 | 0 | — |
case-11 | fail→pass | 14,704 | 9,393 | -36% | 1 | 1 | 0% | 2,313 | 2,538 | +10% | 0 | 0 | — |
case-12 | fail→pass | 18,113 | 5,272 | -71% | 1 | 1 | 0% | 3,052 | 1,439 | -53% | 0 | 0 | — |
case-13 | pass→pass | 9,094 | 6,787 | -25% | 1 | 1 | 0% | 1,331 | 2,033 | +53% | 0 | 0 | — |
case-14 | pass→pass | 16,753 | 6,549 | -61% | 1 | 1 | 0% | 3,496 | 1,703 | -51% | 0 | 0 | — |
case-15 | pass→pass | 20,733 | 4,847 | -77% | 1 | 1 | 0% | 2,423 | 1,417 | -42% | 0 | 0 | — |
case-16 | fail→pass | 14,334 | 17,703 | +24% | 1 | 1 | 0% | 1,990 | 2,702 | +36% | 0 | 0 | — |
case-17 | fail→pass | 12,486 | 8,845 | -29% | 1 | 1 | 0% | 2,035 | 2,000 | -2% | 0 | 0 | — |
case-18 | fail→pass | 14,965 | 5,995 | -60% | 1 | 1 | 0% | 2,334 | 1,644 | -30% | 0 | 0 | — |
case-19 | fail→pass | 6,889 | 4,879 | -29% | 1 | 1 | 0% | 940 | 1,404 | +49% | 0 | 0 | — |
case-20 | fail→pass | 27,347 | 8,719 | -68% | 1 | 1 | 0% | 3,904 | 1,835 | -53% | 0 | 0 | — |
case-21 | fail→pass | 19,022 | 9,760 | -49% | 1 | 1 | 0% | 2,936 | 2,424 | -17% | 0 | 0 | — |
case-22 | fail→pass | 47,095 | 11,054 | -77% | 1 | 1 | 0% | 8,221 | 3,227 | -61% | 0 | 0 | — |
DecimalAI ran this skill against gemini-3.6-flash twice over the same eval suite — once with the skill loaded and once without — and compared the two runs case by case. 22 cases were attempted. The headline lift of +64 percentage points is the difference between those two pass rates over the 22 comparable cases.
Without the skill loaded, the model failed this case. With it loaded, the same prompt on the same model passed. This is one improved case from the latest verified run; every case, including any that regressed, is in the table above.
Other measured skills in the registry, with their headline benchmark lift.