Install any skill in seconds. Free to start, no credit card required.
Get Started Free →Reset to smart interview depth (default). Use when user says '/smart-interview' or wants to reset interview mode.
.claude/skills/majiayu000-smart-interview/SKILL.md| Test case | Without → With | Effect | Δ tokens | Δ turns |
|---|---|---|---|---|
| case-09 | ✗→✓ | ▲ Improved | -32% | 0% |
| case-02 | ✗→✓ | ▲ Improved | 23% | 0% |
| case-08 | ✗→✓ | ▲ Improved | -32% | 0% |
| case-10 | ✗→✓ | ▲ Improved | -40% | 0% |
| case-11 | ✗→✓ | ▲ Improved | -30% | 0% |
You are resetting the interview depth to SMART (default) for this session.
This session will use smart interview depth (the default):
IF "new project" + no existing code:
→ Full depth (all questions)
ELIF "modify" + existing code found:
→ Medium depth (5-8 questions)
ELIF "micro" OR estimated <100 lines:
→ Quick depth (Q1, Q2, Q6, Q12)
ELIF bypass signals (typo, rename, explicit path + <10 lines):
→ Bypass with one triage questionInterview depth: SMART (reset to default)
- Will auto-detect appropriate depth
- Greenfield → full, Modify → medium, Micro → quickContinue with standard front-door flow, which will use smart detection.
Use Task tool to spawn front-door:
yamlTask: subagent_type: general-purpose prompt: | INTERVIEW DEPTH: smart (default) Execute front-door skill with smart detection. Use the depth detection logic: - New project + no code → full depth - Modify existing → medium depth (5-8 questions) - Micro task → quick depth (Q1, Q2, Q6, Q12) Start by triaging the user's intent.
smart-interview, smart, default, reset, auto, normal
| Case | Status | Duration (ms) | Turns | Tokens | Tool calls | ||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| Without | With | Δ | Without | With | Δ | Without | With | Δ | Without | With | Δ | ||
case-09 | fail→pass | 15,613 | 9,291 | -40% | 1 | 1 | 0% | 1,626 | 1,112 | -32% | 0 | 0 | — |
case-01 | fail→fail | 16,208 | 10,904 | -33% | 1 | 1 | 0% | 977 | 1,331 | +36% | 0 | 0 | — |
case-02 | fail→pass | 10,683 | 3,661 | -66% | 1 | 1 | 0% | 863 | 1,065 | +23% | 0 | 0 | — |
case-03 | fail→fail | 9,888 | 4,048 | -59% | 1 | 1 | 0% | 727 | 907 | +25% | 0 | 0 | — |
case-04 | pass→pass | 13,647 | 10,002 | -27% | 1 | 1 | 0% | 1,286 | 2,112 | +64% | 0 | 0 | — |
case-05 | pass→pass | 5,520 | 5,326 | -4% | 1 | 1 | 0% | 735 | 1,177 | +60% | 0 | 0 | — |
case-06 | pass→pass | 19,525 | 23,573 | +21% | 1 | 1 | 0% | 2,987 | 3,023 | +1% | 0 | 0 | — |
case-07 | pass→pass | 17,003 | 5,757 | -66% | 1 | 1 | 0% | 1,751 | 1,107 | -37% | 0 | 0 | — |
case-08 | fail→pass | 18,551 | 11,437 | -38% | 1 | 1 | 0% | 1,995 | 1,354 | -32% | 0 | 0 | — |
case-10 | fail→pass | 17,716 | 6,981 | -61% | 1 | 1 | 0% | 1,896 | 1,134 | -40% | 0 | 0 | — |
case-11 | fail→pass | 8,386 | 4,163 | -50% | 1 | 1 | 0% | 1,506 | 1,054 | -30% | 0 | 0 | — |
case-12 | fail→pass | 16,728 | 9,550 | -43% | 1 | 1 | 0% | 2,773 | 1,353 | -51% | 0 | 0 | — |
case-13 | fail→pass | 13,781 | 8,987 | -35% | 1 | 1 | 0% | 1,680 | 1,061 | -37% | 0 | 0 | — |
case-14 | fail→pass | 9,337 | 8,327 | -11% | 1 | 1 | 0% | 1,383 | 906 | -34% | 0 | 0 | — |
case-15 | fail→pass | 18,058 | 9,189 | -49% | 1 | 1 | 0% | 2,057 | 1,144 | -44% | 0 | 0 | — |
case-16 | fail→pass | 21,566 | 5,662 | -74% | 1 | 1 | 0% | 1,728 | 1,355 | -22% | 0 | 0 | — |
case-17 | fail→pass | 46,869 | 19,173 | -59% | 1 | 1 | 0% | 3,353 | 2,446 | -27% | 0 | 0 | — |
case-18 | pass→pass | 19,796 | 14,699 | -26% | 1 | 1 | 0% | 3,024 | 1,922 | -36% | 0 | 0 | — |
case-19 | fail→pass | 51,126 | 8,433 | -84% | 1 | 1 | 0% | 1,554 | 989 | -36% | 0 | 0 | — |
case-20 | fail→pass | 13,916 | 9,251 | -34% | 1 | 1 | 0% | 1,839 | 1,110 | -40% | 0 | 0 | — |
case-21 | fail→pass | 16,211 | 7,253 | -55% | 1 | 1 | 0% | 1,606 | 778 | -52% | 0 | 0 | — |
case-22 | fail→pass | 14,727 | 7,268 | -51% | 1 | 1 | 0% | 1,399 | 764 | -45% | 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 +68 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.