Install any skill in seconds. Free to start, no credit card required.
Get Started Free →When a learner is present and the goal is to advance THEIR reasoning rather than hand them the result, reply with exactly one guiding question that targets the single next gap in their thinking — and never state the answer, never lecture the solution, and never smuggle the answer into the question itself. Use when someone is coaching, tutoring, or walking a student/colleague toward figuring something out on their own. Do NOT use when the user just wants the answer or worked solution delivered directly, when they want an explainer/lecture written for an audience, or when they want work graded against a rubric.
.claude/skills/socratic-question/SKILL.md| Test case | Without → With | Effect | Δ tokens | Δ turns |
|---|---|---|---|---|
| case-01 | ✗→✓ | ▲ Improved | — | — |
| case-03 | ✗→✓ | ▲ Improved | — | — |
| case-12 | ✗→✓ | ▲ Improved | — | — |
| case-08 | ✗→✓ | ▲ Improved | — | — |
| case-05 | ✗→✓ | ▲ Improved | — | — |
When a learner is stuck and someone asks you to help them understand, the default pull is to teach: explain the concept, walk the solution, and — trying to be gentle — wrap the answer inside a leading question ("So since the denominator is bigger, the fraction gets smaller, right?"). That feels like guiding, but it does the thinking for them. The learner nods along and leaves without having crossed the gap themselves. The discipline here is the opposite: hand back exactly one question, aimed at the one step they are actually stuck on, and let them do the crossing. Withholding the answer is the whole point — it is what turns a lecture into learning.
Activate when a learner is in the loop and the goal is to move their reasoning forward: "help me get my student to see why…", "walk her toward figuring it out", "coach him to the insight", "I'm tutoring someone who thinks X — get them to spot the flaw themselves", "don't just fix it, help them understand it."
Do NOT activate when the user wants the result, not the journey: they ask for the answer, the worked solution, or the fix directly ("just tell me", "give me the final number"). Do NOT activate to write an explainer or lecture for an audience — that is exposition, not dialogue. And do NOT activate to grade or score someone's work against a rubric — that is assessment. This is for the live, one-on-one moment where a person is one step away from getting it.
| Case | Status | Duration (ms) | Turns | Tokens | Tool calls | ||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| Without | With | Δ | Without | With | Δ | Without | With | Δ | Without | With | Δ | ||
case-01 | fail→pass | — | — | — | — | — | — | — | — | — | — | — | — |
case-03 | fail→pass | — | — | — | — | — | — | — | — | — | — | — | — |
case-11 | fail→fail | — | — | — | — | — | — | — | — | — | — | — | — |
case-12 | fail→pass | — | — | — | — | — | — | — | — | — | — | — | — |
case-08 | fail→pass | — | — | — | — | — | — | — | — | — | — | — | — |
case-05 | fail→pass | — | — | — | — | — | — | — | — | — | — | — | — |
case-02 | fail→pass | — | — | — | — | — | — | — | — | — | — | — | — |
case-07 | fail→pass | — | — | — | — | — | — | — | — | — | — | — | — |
case-10 | fail→fail | — | — | — | — | — | — | — | — | — | — | — | — |
case-06 | fail→pass | — | — | — | — | — | — | — | — | — | — | — | — |
case-13 | fail→pass | — | — | — | — | — | — | — | — | — | — | — | — |
case-04 | fail→fail | — | — | — | — | — | — | — | — | — | — | — | — |
case-09 | fail→pass | — | — | — | — | — | — | — | — | — | — | — | — |
DecimalAI ran this skill against gemini-3.5-flash twice over the same eval suite — once with the skill loaded and once without — and compared the two runs case by case. 13 cases were attempted. The headline lift of +77 percentage points is the difference between those two pass rates over the 13 comparable cases.
Other measured skills in the registry, with their headline benchmark lift.