Loading skill
Install any skill in seconds. Free to start, no credit card required.
Get Started Free →Apply when tempted to ask 'should I do X?' on reversible work. Proceed, present the result, let the human course-correct after the fact; reserve confirmation for irreversible actions.
.claude/skills/kunanonj-cursor-plugin-pstack-principle-never-block-on-the-human/SKILL.md| Test case | Without → With | Effect | Δ tokens | Δ turns |
|---|---|---|---|---|
| case-04 | ✗→✓ | ▲ Improved | 549% | 0% |
| case-01 | ✓→✗ | ▼ Worse | -32% | 0% |
| case-07 | ✓→✗ | ▼ Worse | -23% | 0% |
| case-09 | ✓→✗ | ▼ Worse | -53% | 0% |
| case-19 | ✓→✗ | ▼ Worse | -50% | 0% |
The human supervises asynchronously. Agents must stay unblocked: make reasonable decisions, proceed, and let the human course-correct after the fact. Code is cheap. Waiting is expensive.
Why: Every permission pause stalls the pipeline and makes the human the bottleneck. Since code changes are reversible and reviewable, a wrong decision usually costs less than blocking.
Pattern:
Boundaries:
| Case | Status | Duration (ms) | Turns | Tokens | Tool calls | ||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| Without | With | Δ | Without | With | Δ | Without | With | Δ | Without | With | Δ | ||
case-01 | pass→fail | 5,091 | 3,375 | -34% | 1 | 1 | 0% | 908 | 620 | -32% | 0 | 0 | — |
case-06 | pass→pass | 6,311 | 3,532 | -44% | 1 | 1 | 0% | 1,118 | 862 | -23% | 0 | 0 | — |
case-02 | pass→pass | 9,807 | 6,215 | -37% | 1 | 1 | 0% | 1,863 | 1,603 | -14% | 0 | 0 | — |
case-03 | fail→fail | 3,822 | 3,055 | -20% | 1 | 1 | 0% | 598 | 723 | +21% | 0 | 0 | — |
case-04 | fail→pass | 3,484 | 16,618 | +377% | 1 | 1 | 0% | 657 | 4,261 | +549% | 0 | 0 | — |
case-05 | fail→fail | 3,237 | 2,434 | -25% | 1 | 1 | 0% | 570 | 677 | +19% | 0 | 0 | — |
case-07 | pass→fail | 4,615 | 2,558 | -45% | 1 | 1 | 0% | 845 | 648 | -23% | 0 | 0 | — |
case-08 | fail→fail | 5,317 | 3,426 | -36% | 1 | 1 | 0% | 845 | 855 | +1% | 0 | 0 | — |
case-09 | pass→fail | 8,229 | 2,412 | -71% | 1 | 1 | 0% | 1,347 | 639 | -53% | 0 | 0 | — |
case-10 | fail→fail | 5,920 | 3,391 | -43% | 1 | 1 | 0% | 930 | 842 | -9% | 0 | 0 | — |
case-11 | fail→fail | 8,226 | 3,949 | -52% | 1 | 1 | 0% | 1,337 | 867 | -35% | 0 | 0 | — |
case-12 | fail→fail | 7,473 | 2,758 | -63% | 1 | 1 | 0% | 1,187 | 768 | -35% | 0 | 0 | — |
case-13 | fail→fail | 6,404 | 3,113 | -51% | 1 | 1 | 0% | 996 | 817 | -18% | 0 | 0 | — |
case-14 | fail→fail | 5,053 | 2,337 | -54% | 1 | 1 | 0% | 710 | 665 | -6% | 0 | 0 | — |
case-15 | fail→fail | 4,011 | 2,195 | -45% | 1 | 1 | 0% | 681 | 670 | -2% | 0 | 0 | — |
case-16 | fail→fail | 8,927 | 2,504 | -72% | 1 | 1 | 0% | 1,316 | 645 | -51% | 0 | 0 | — |
case-17 | fail→fail | 7,309 | 2,605 | -64% | 1 | 1 | 0% | 1,198 | 715 | -40% | 0 | 0 | — |
case-18 | fail→fail | 8,796 | 2,957 | -66% | 1 | 1 | 0% | 1,433 | 741 | -48% | 0 | 0 | — |
case-19 | pass→fail | 8,899 | 2,631 | -70% | 1 | 1 | 0% | 1,453 | 722 | -50% | 0 | 0 | — |
case-20 | pass→pass | 4,232 | 2,694 | -36% | 1 | 1 | 0% | 753 | 660 | -12% | 0 | 0 | — |
case-21 | pass→pass | 4,989 | 2,843 | -43% | 1 | 1 | 0% | 818 | 786 | -4% | 0 | 0 | — |
case-22 | pass→pass | 8,550 | 5,337 | -38% | 1 | 1 | 0% | 1,412 | 1,101 | -22% | 0 | 0 | — |
case-23 | pass→pass | 7,971 | 2,509 | -69% | 1 | 1 | 0% | 1,432 | 727 | -49% | 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. 23 cases were attempted. The headline lift of -60 percentage points is the difference between those two pass rates over the 23 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.
Other measured skills in the registry, with their headline benchmark lift.