Loading skill
Install any skill in seconds. Free to start, no credit card required.
Get Started Free →Enable continuous learning mode for automatic insight extraction
.claude/skills/aiskillstore-learn-on/SKILL.md| Test case | Without → With | Effect | Δ tokens | Δ turns |
|---|---|---|---|---|
| case-02 | ✗→✓ | ▲ Improved | -6% | 0% |
| case-04 | ✗→✓ | ▲ Improved | -50% | 0% |
| case-05 | ✗→✓ | ▲ Improved | -64% | 0% |
| case-06 | ✗→✓ | ▲ Improved | -60% | 0% |
| case-07 | ✗→✓ | ▲ Improved | -48% | 0% |
Enable continuous learning mode. When active, the router will periodically extract insights during the session.
Activates continuous learning mode where:
This is useful for long sessions where you want to capture insights as you go without remembering to run /learn manually.
knowledge/state.jsonjson { "learning_mode": true, "learning_mode_since": "[current ISO timestamp]", "queries_since_extraction": 0 }
knowledge/state.jsonContinuous Learning: ENABLED
────────────────────────────
Learning mode is now active.
Extraction will trigger automatically:
- Every 10 queries, or
- Every 30 minutes of activity
Insights will be saved to:
- knowledge/learnings/patterns.md
- knowledge/learnings/quirks.md
- knowledge/learnings/decisions.md
Use /learn-off to disable, or /learn for manual extraction./learn manually while continuous mode is active/knowledge to see current learning status| Case | Status | Duration (ms) | Turns | Tokens | Tool calls | ||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| Without | With | Δ | Without | With | Δ | Without | With | Δ | Without | With | Δ | ||
case-01 | fail→fail | 13,051 | 4,524 | -65% | 1 | 1 | 0% | 1,930 | 1,173 | -39% | 0 | 0 | — |
case-02 | fail→pass | 10,903 | 3,262 | -70% | 1 | 1 | 0% | 1,001 | 936 | -6% | 0 | 0 | — |
case-03 | fail→fail | 9,128 | 3,557 | -61% | 1 | 1 | 0% | 1,543 | 1,004 | -35% | 0 | 0 | — |
case-04 | fail→pass | 10,117 | 2,908 | -71% | 1 | 1 | 0% | 1,548 | 774 | -50% | 0 | 0 | — |
case-05 | fail→pass | 14,206 | 3,639 | -74% | 1 | 1 | 0% | 2,351 | 840 | -64% | 0 | 0 | — |
case-06 | fail→pass | 13,939 | 2,173 | -84% | 1 | 1 | 0% | 1,516 | 613 | -60% | 0 | 0 | — |
case-07 | fail→pass | 9,590 | 2,325 | -76% | 1 | 1 | 0% | 1,507 | 783 | -48% | 0 | 0 | — |
case-08 | pass→pass | 4,371 | 1,540 | -65% | 1 | 1 | 0% | 611 | 594 | -3% | 0 | 0 | — |
case-09 | pass→pass | 6,835 | 2,590 | -62% | 1 | 1 | 0% | 1,068 | 781 | -27% | 0 | 0 | — |
case-10 | pass→pass | 12,616 | 1,846 | -85% | 1 | 1 | 0% | 1,772 | 585 | -67% | 0 | 0 | — |
case-11 | pass→pass | 12,331 | 1,769 | -86% | 1 | 1 | 0% | 1,719 | 562 | -67% | 0 | 0 | — |
case-12 | pass→pass | 4,517 | 1,569 | -65% | 1 | 1 | 0% | 750 | 575 | -23% | 0 | 0 | — |
case-13 | fail→pass | 5,504 | 2,952 | -46% | 1 | 1 | 0% | 833 | 871 | +5% | 0 | 0 | — |
case-14 | pass→fail | 8,369 | 7,860 | -6% | 1 | 1 | 0% | 1,346 | 774 | -42% | 0 | 0 | — |
case-15 | fail→pass | 9,227 | 2,774 | -70% | 1 | 1 | 0% | 1,399 | 778 | -44% | 0 | 0 | — |
case-16 | fail→pass | 13,030 | 4,659 | -64% | 1 | 1 | 0% | 1,934 | 628 | -68% | 0 | 0 | — |
case-17 | fail→pass | 16,055 | 5,131 | -68% | 1 | 1 | 0% | 2,309 | 1,124 | -51% | 0 | 0 | — |
case-18 | fail→pass | 27,797 | 1,980 | -93% | 1 | 1 | 0% | 592 | 680 | +15% | 0 | 0 | — |
case-19 | fail→pass | 3,853 | 4,636 | +20% | 1 | 1 | 0% | 509 | 851 | +67% | 0 | 0 | — |
case-20 | fail→pass | 6,724 | 4,231 | -37% | 1 | 1 | 0% | 1,013 | 875 | -14% | 0 | 0 | — |
case-21 | fail→pass | 3,768 | 3,588 | -5% | 1 | 1 | 0% | 583 | 1,095 | +88% | 0 | 0 | — |
case-22 | fail→pass | 10,604 | 1,902 | -82% | 1 | 1 | 0% | 1,655 | 590 | -64% | 0 | 0 | — |
case-23 | fail→pass | 5,651 | 8,282 | +47% | 1 | 1 | 0% | 740 | 1,139 | +54% | 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, and 21 counted toward the lift figure. The other 2 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 +61 percentage points is the difference between those two pass rates over the 21 comparable cases. 1 case got worse with the skill loaded, and it is 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.