Loading skill
Install any skill in seconds. Free to start, no credit card required.
Get Started Free →Reads the input content, identifies the key signals (decisions, action items, facts, commitments), and produces a compressed summary at the target length. Preserves entropy -- maximum meaning per unit of output.
.claude/skills/miosa-osa-summarize/SKILL.md| Test case | Without → With | Effect | Δ tokens | Δ turns |
|---|---|---|---|---|
| case-07 | ✗→✓ | ▲ Improved | 9% | 0% |
| case-09 | ✗→✓ | ▲ Improved | 37% | 0% |
| case-10 | ✗→✓ | ▲ Improved | 153% | 0% |
| case-14 | ✗→✓ | ▲ Improved | -29% | 0% |
| case-15 | ✗→✓ | ▲ Improved | 69% | 0% |
> Summarize content for a target audience. Writer-owned.
/summarize <source> [--audience <executive|technical|general>] [--length <short|medium|long>]Takes source content (document, transcript, data) and produces a structured summary tailored to the target audience's genre and comprehension level.
| Length | Words | Use Case | |--------|-------|----------| | short | 50-100 | Slack message, quick update | | medium | 150-300 | Email, brief, status update | | long | 400-800 | Report section, detailed overview |
bash/summarize meeting-transcript.md --audience executive --length short /summarize architecture-doc.md --audience technical --length medium /summarize quarterly-data.md --audience general --length long
| Case | Status | Duration (ms) | Turns | Tokens | Tool calls | ||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| Without | With | Δ | Without | With | Δ | Without | With | Δ | Without | With | Δ | ||
case-01 | pass→fail | 14,561 | 8,213 | -44% | 1 | 1 | 0% | 2,517 | 1,650 | -34% | 0 | 0 | — |
case-02 | fail→fail | 7,681 | 6,387 | -17% | 1 | 1 | 0% | 1,372 | 1,305 | -5% | 0 | 0 | — |
case-03 | fail→fail | 1,961 | 2,682 | +37% | 1 | 1 | 0% | 300 | 726 | +142% | 0 | 0 | — |
case-04 | pass→pass | 17,830 | 14,363 | -19% | 1 | 1 | 0% | 2,961 | 2,709 | -9% | 0 | 0 | — |
case-05 | fail→fail | 2,065 | 3,525 | +71% | 1 | 1 | 0% | 306 | 855 | +179% | 0 | 0 | — |
case-06 | fail→fail | 2,854 | 7,795 | +173% | 1 | 1 | 0% | 431 | 1,491 | +246% | 0 | 0 | — |
case-07 | fail→pass | 8,520 | 7,979 | -6% | 1 | 1 | 0% | 1,380 | 1,505 | +9% | 0 | 0 | — |
case-08 | pass→pass | 11,800 | 10,789 | -9% | 1 | 1 | 0% | 1,929 | 1,964 | +2% | 0 | 0 | — |
case-09 | fail→pass | 5,309 | 4,942 | -7% | 1 | 1 | 0% | 812 | 1,111 | +37% | 0 | 0 | — |
case-10 | fail→pass | 4,543 | 8,441 | +86% | 1 | 1 | 0% | 681 | 1,724 | +153% | 0 | 0 | — |
case-11 | pass→pass | 12,889 | 10,177 | -21% | 1 | 1 | 0% | 2,160 | 2,060 | -5% | 0 | 0 | — |
case-12 | fail→fail | 44,444 | 42,375 | -5% | 1 | 1 | 0% | 8,216 | 7,546 | -8% | 0 | 0 | — |
case-13 | fail→fail | 5,762 | 4,853 | -16% | 1 | 1 | 0% | 1,098 | 1,129 | +3% | 0 | 0 | — |
case-14 | fail→pass | 11,411 | 7,069 | -38% | 1 | 1 | 0% | 1,801 | 1,284 | -29% | 0 | 0 | — |
case-15 | fail→pass | 5,627 | 8,514 | +51% | 1 | 1 | 0% | 902 | 1,521 | +69% | 0 | 0 | — |
case-16 | fail→fail | 4,207 | 5,941 | +41% | 1 | 1 | 0% | 662 | 1,324 | +100% | 0 | 0 | — |
case-17 | fail→pass | 5,465 | 7,830 | +43% | 1 | 1 | 0% | 825 | 1,494 | +81% | 0 | 0 | — |
case-18 | fail→pass | 19,038 | 10,933 | -43% | 1 | 1 | 0% | 3,208 | 2,048 | -36% | 0 | 0 | — |
case-19 | fail→pass | 4,424 | 5,490 | +24% | 1 | 1 | 0% | 690 | 1,168 | +69% | 0 | 0 | — |
case-20 | fail→pass | 9,213 | 7,904 | -14% | 1 | 1 | 0% | 1,475 | 1,570 | +6% | 0 | 0 | — |
case-21 | fail→fail | 3,368 | 12,828 | +281% | 1 | 1 | 0% | 477 | 2,155 | +352% | 0 | 0 | — |
case-22 | fail→fail | 6,545 | 6,949 | +6% | 1 | 1 | 0% | 1,143 | 1,454 | +27% | 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 +36 percentage points is the difference between those two pass rates over the 22 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.