Loading skill
Install any skill in seconds. Free to start, no credit card required.
Get Started Free →Compact the current conversation into a handoff document for another agent to pick up.
.claude/skills/asymmetric-al-handoff/SKILL.md| Test case | Without → With | Effect | Δ tokens | Δ turns |
|---|---|---|---|---|
| case-04 | ✗→✓ | ▲ Improved | 1427% | 0% |
| case-06 | ✗→✓ | ▲ Improved | 8% | 0% |
| case-12 | ✗→✓ | ▲ Improved | 80% | 0% |
| case-19 | ✗→✓ | ▲ Improved | 235% | 0% |
| case-07 | ✓→✗ | ▼ Worse | -54% | 0% |
Write a handoff document summarising the current conversation so a fresh agent can continue the work. Save to the temporary directory of the user's OS - not the current workspace.
Include a "suggested skills" section in the document, which suggests skills that the agent should invoke.
Do not duplicate content already captured in other artifacts (specs, plans, ADRs, issues, commits, diffs). Reference them by path or URL instead.
Redact any sensitive information, such as API keys, passwords, or personally identifiable information.
If the user passed arguments, treat them as a description of what the next session will focus on and tailor the doc accordingly.
| Case | Status | Duration (ms) | Turns | Tokens | Tool calls | ||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| Without | With | Δ | Without | With | Δ | Without | With | Δ | Without | With | Δ | ||
case-10 | fail→fail | 5,616 | 8,154 | +45% | 1 | 1 | 0% | 876 | 864 | -1% | 0 | 0 | — |
case-01 | fail→fail | 7,450 | 7,901 | +6% | 1 | 1 | 0% | 1,126 | 598 | -47% | 0 | 0 | — |
case-02 | fail→fail | 10,226 | 8,094 | -21% | 1 | 1 | 0% | 1,576 | 814 | -48% | 0 | 0 | — |
case-03 | fail→fail | 9,869 | 6,880 | -30% | 1 | 1 | 0% | 1,570 | 668 | -57% | 0 | 0 | — |
case-04 | fail→pass | 4,135 | 15,604 | +277% | 1 | 1 | 0% | 151 | 2,306 | +1427% | 0 | 0 | — |
case-09 | pass→pass | 18,600 | 20,844 | +12% | 1 | 1 | 0% | 2,768 | 3,606 | +30% | 0 | 0 | — |
case-05 | fail→fail | 15,164 | 7,073 | -53% | 1 | 1 | 0% | 2,333 | 303 | -87% | 0 | 0 | — |
case-06 | fail→pass | 6,554 | 8,116 | +24% | 1 | 1 | 0% | 998 | 1,080 | +8% | 0 | 0 | — |
case-07 | pass→fail | 6,903 | 7,237 | +5% | 1 | 1 | 0% | 1,168 | 534 | -54% | 0 | 0 | — |
case-08 | pass→pass | 12,203 | 24,060 | +97% | 1 | 1 | 0% | 2,168 | 4,532 | +109% | 0 | 0 | — |
case-11 | pass→fail | 10,468 | 7,038 | -33% | 1 | 1 | 0% | 1,639 | 446 | -73% | 0 | 0 | — |
case-12 | fail→pass | 12,307 | 20,192 | +64% | 1 | 1 | 0% | 1,901 | 3,416 | +80% | 0 | 0 | — |
case-13 | pass→pass | 15,739 | 22,755 | +45% | 1 | 1 | 0% | 2,409 | 3,992 | +66% | 0 | 0 | — |
case-14 | fail→fail | 9,492 | 11,580 | +22% | 1 | 1 | 0% | 1,764 | 586 | -67% | 0 | 0 | — |
case-15 | pass→fail | 9,471 | 10,111 | +7% | 1 | 1 | 0% | 1,582 | 965 | -39% | 0 | 0 | — |
case-16 | pass→pass | 11,656 | 22,449 | +93% | 1 | 1 | 0% | 1,849 | 3,893 | +111% | 0 | 0 | — |
case-17 | pass→pass | 7,782 | 12,181 | +57% | 1 | 1 | 0% | 1,342 | 2,185 | +63% | 0 | 0 | — |
case-18 | pass→fail | 11,449 | 8,027 | -30% | 1 | 1 | 0% | 2,012 | 967 | -52% | 0 | 0 | — |
case-19 | fail→pass | 3,906 | 11,072 | +183% | 1 | 1 | 0% | 647 | 2,166 | +235% | 0 | 0 | — |
case-20 | pass→pass | 10,050 | 15,354 | +53% | 1 | 1 | 0% | 1,796 | 2,954 | +64% | 0 | 0 | — |
case-21 | pass→fail | 7,393 | 8,180 | +11% | 1 | 1 | 0% | 1,303 | 713 | -45% | 0 | 0 | — |
case-22 | pass→pass | 9,938 | 15,087 | +52% | 1 | 1 | 0% | 1,723 | 2,755 | +60% | 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, and 11 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 -5 percentage points is the difference between those two pass rates over the 11 comparable cases. 8 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.