Install any skill in seconds. Free to start, no credit card required.
Get Started Free →落实 CLAUDE.md / AGENTS.md 中的“主动探索”原则,在向用户提问前自动使用 Grep、Read、Bash、WebSearch 等工具获取信息
.claude/skills/cnfjlhj-proactive-explorer/SKILL.md| Test case | Without → With | Effect | Δ tokens | Δ turns |
|---|---|---|---|---|
| case-11 | ✗→✓ | ▲ Improved | 21% | 0% |
| case-12 | ✗→✓ | ▲ Improved | 18% | 0% |
| case-23 | ✗→✓ | ▲ Improved | 2% | 0% |
| case-24 | ✗→✓ | ▲ Improved | -9% | 0% |
| case-02 | ✓→✗ | ▼ Worse | 163% | 0% |
Use this skill to reduce avoidable user questions.
Core principle: search local context before asking the user to restate information the system can gather on its own.
rg, targeted file reads, and focused shell commands.Use when any of the following are true:
Do not use when:
AGENTS.md, project docs, and relevant config.rg and targeted reads.Before asking a clarifying question, be able to state:
| Case | Status | Duration (ms) | Turns | Tokens | Tool calls | ||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| Without | With | Δ | Without | With | Δ | Without | With | Δ | Without | With | Δ | ||
case-01 | fail→fail | 31,327 | 18,054 | -42% | 1 | 1 | 0% | 215 | 836 | +289% | 0 | 0 | — |
case-02 | pass→fail | 25,027 | 16,336 | -35% | 1 | 1 | 0% | 356 | 935 | +163% | 0 | 0 | — |
case-03 | fail→fail | 18,342 | 11,460 | -38% | 1 | 1 | 0% | 261 | 769 | +195% | 0 | 0 | — |
case-04 | pass→fail | 18,045 | 27,140 | +50% | 1 | 1 | 0% | 1,960 | 814 | -58% | 0 | 0 | — |
case-05 | fail→fail | 13,598 | 11,653 | -14% | 1 | 1 | 0% | 1,470 | 834 | -43% | 0 | 0 | — |
case-06 | pass→pass | 24,484 | 8,164 | -67% | 1 | 1 | 0% | 778 | 1,703 | +119% | 0 | 0 | — |
case-07 | pass→fail | 6,004 | 36,460 | +507% | 1 | 1 | 0% | 944 | 882 | -7% | 0 | 0 | — |
case-08 | fail→fail | 4,964 | 19,126 | +285% | 1 | 1 | 0% | 697 | 783 | +12% | 0 | 0 | — |
case-09 | fail→fail | 15,693 | 15,676 | -0% | 1 | 1 | 0% | 1,745 | 2,169 | +24% | 0 | 0 | — |
case-10 | pass→pass | 9,360 | 11,931 | +27% | 1 | 1 | 0% | 1,454 | 1,564 | +8% | 0 | 0 | — |
case-11 | fail→pass | 23,559 | 9,207 | -61% | 1 | 1 | 0% | 920 | 1,116 | +21% | 0 | 0 | — |
case-12 | fail→pass | 13,853 | 13,337 | -4% | 1 | 1 | 0% | 1,331 | 1,569 | +18% | 0 | 0 | — |
case-13 | pass→pass | 17,260 | 11,946 | -31% | 1 | 1 | 0% | 1,236 | 1,158 | -6% | 0 | 0 | — |
case-14 | fail→fail | 11,198 | 46,187 | +312% | 1 | 1 | 0% | 1,618 | 861 | -47% | 0 | 0 | — |
case-15 | fail→fail | 10,329 | 16,730 | +62% | 1 | 1 | 0% | 1,532 | 737 | -52% | 0 | 0 | — |
case-16 | pass→pass | 18,878 | 5,435 | -71% | 1 | 1 | 0% | 1,480 | 1,292 | -13% | 0 | 0 | — |
case-17 | pass→pass | 23,966 | 14,531 | -39% | 1 | 1 | 0% | 1,181 | 1,158 | -2% | 0 | 0 | — |
case-18 | fail→fail | 39,218 | 13,191 | -66% | 1 | 1 | 0% | 1,265 | 1,491 | +18% | 0 | 0 | — |
case-19 | fail→fail | 32,111 | 52,487 | +63% | 1 | 1 | 0% | 1,708 | 2,212 | +30% | 0 | 0 | — |
case-20 | pass→pass | 11,883 | 13,051 | +10% | 1 | 1 | 0% | 885 | 1,117 | +26% | 0 | 0 | — |
case-21 | fail→fail | 15,822 | 14,972 | -5% | 1 | 1 | 0% | 1,441 | 924 | -36% | 0 | 0 | — |
case-22 | pass→fail | 14,669 | 16,822 | +15% | 1 | 1 | 0% | 1,240 | 840 | -32% | 0 | 0 | — |
case-23 | fail→pass | 15,373 | 11,824 | -23% | 1 | 1 | 0% | 1,425 | 1,454 | +2% | 0 | 0 | — |
case-24 | fail→pass | 14,403 | 14,416 | +0% | 1 | 1 | 0% | 1,054 | 964 | -9% | 0 | 0 | — |
case-25 | pass→pass | 6,450 | 8,332 | +29% | 1 | 1 | 0% | 633 | 923 | +46% | 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. 25 cases were attempted, and 14 counted toward the lift figure. The other 11 produced results that are not comparable between the two arms, so they are excluded from the headline rather than averaged into it. The headline lift of 0 percentage points is the difference between those two pass rates over the 14 comparable cases. 4 cases got worse with the skill loaded, and they are included in that figure.
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.
| Model | Method | Date | Lift |
|---|---|---|---|
| gemini-3.6-flash | verified | 8/20/2026 | — |
Other measured skills in the registry, with their headline benchmark lift.