Loading skill
Install any skill in seconds. Free to start, no credit card required.
Get Started Free →Sibyl 反思 agent - 分析迭代产出，分类问题，生成改进计划
.claude/skills/majiayu000-sibyl-reflection/SKILL.md| Test case | Without → With | Effect | Δ tokens | Δ turns |
|---|---|---|---|---|
| case-04 | ✗→✓ | ▲ Improved | -15% | 0% |
| case-05 | ✗→✓ | ▲ Improved | -66% | 0% |
| case-06 | ✗→✓ | ▲ Improved | 307% | 0% |
| case-07 | ✗→✓ | ▲ Improved | -83% | 0% |
| case-08 | ✗→✓ | ▲ Improved | -88% | 0% |
!SIBYL_WORKSPACE="$ARGUMENTS[0]" .venv/bin/python3 -c "from sibyl.orchestrate import render_skill_prompt; import os; ws = os.environ.get('SIBYL_WORKSPACE', ''); print(render_skill_prompt('reflection', workspace_path=ws))"
AGENT_NAME: sibyl-reflection AGENT_TIER: sibyl-heavy SIBYL_ROOT: /Users/cwan0785/sibyl-system
Workspace path: $ARGUMENTS0] Current iteration: $ARGUMENTS1]
| Case | Status | Duration (ms) | Turns | Tokens | Tool calls | ||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| Without | With | Δ | Without | With | Δ | Without | With | Δ | Without | With | Δ | ||
case-01 | fail→fail | 17,641 | 25,345 | +44% | 1 | 1 | 0% | 1,465 | 3,359 | +129% | 0 | 0 | — |
case-02 | fail→fail | 5,467 | 6,663 | +22% | 1 | 1 | 0% | 208 | 548 | +163% | 0 | 0 | — |
case-03 | fail→fail | 17,537 | 16,615 | -5% | 1 | 1 | 0% | 338 | 496 | +47% | 0 | 0 | — |
case-04 | fail→pass | 10,958 | 7,885 | -28% | 1 | 1 | 0% | 2,110 | 1,801 | -15% | 0 | 0 | — |
case-05 | fail→pass | 16,251 | 2,917 | -82% | 1 | 1 | 0% | 1,960 | 672 | -66% | 0 | 0 | — |
case-06 | fail→pass | 11,083 | 17,330 | +56% | 1 | 1 | 0% | 648 | 2,638 | +307% | 0 | 0 | — |
case-07 | fail→pass | 20,926 | 6,773 | -68% | 1 | 1 | 0% | 2,213 | 383 | -83% | 0 | 0 | — |
case-08 | fail→pass | 27,535 | 7,102 | -74% | 1 | 1 | 0% | 3,469 | 433 | -88% | 0 | 0 | — |
case-09 | fail→pass | 9,186 | 22,185 | +142% | 1 | 1 | 0% | 620 | 3,008 | +385% | 0 | 0 | — |
case-10 | fail→fail | 19,209 | 15,225 | -21% | 1 | 1 | 0% | 2,125 | 1,201 | -43% | 0 | 0 | — |
case-11 | fail→pass | 18,333 | 7,112 | -61% | 1 | 1 | 0% | 2,109 | 448 | -79% | 0 | 0 | — |
case-12 | pass→pass | 19,255 | 1,641 | -91% | 1 | 1 | 0% | 2,264 | 410 | -82% | 0 | 0 | — |
case-13 | fail→pass | 20,416 | 1,774 | -91% | 1 | 1 | 0% | 2,566 | 432 | -83% | 0 | 0 | — |
case-14 | fail→pass | 11,100 | 4,818 | -57% | 1 | 1 | 0% | 1,376 | 1,024 | -26% | 0 | 0 | — |
case-19 | fail→pass | 9,491 | 6,249 | -34% | 1 | 1 | 0% | 1,524 | 1,269 | -17% | 0 | 0 | — |
case-15 | pass→pass | 15,116 | 4,796 | -68% | 1 | 1 | 0% | 1,688 | 974 | -42% | 0 | 0 | — |
case-16 | fail→pass | 13,716 | 6,929 | -49% | 1 | 1 | 0% | 1,208 | 464 | -62% | 0 | 0 | — |
case-17 | fail→pass | 24,884 | 10,456 | -58% | 1 | 1 | 0% | 3,153 | 1,209 | -62% | 0 | 0 | — |
case-18 | fail→pass | 21,370 | 6,630 | -69% | 1 | 1 | 0% | 2,424 | 328 | -86% | 0 | 0 | — |
case-20 | pass→pass | 4,854 | 7,518 | +55% | 1 | 1 | 0% | 687 | 509 | -26% | 0 | 0 | — |
case-21 | pass→pass | 16,070 | 14,305 | -11% | 1 | 1 | 0% | 1,968 | 1,827 | -7% | 0 | 0 | — |
case-22 | pass→pass | 12,283 | 7,618 | -38% | 1 | 1 | 0% | 1,150 | 1,489 | +29% | 0 | 0 | — |
case-23 | pass→pass | 4,430 | 3,718 | -16% | 1 | 1 | 0% | 921 | 861 | -7% | 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 19 counted toward the lift figure. The other 4 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 +57 percentage points is the difference between those two pass rates over the 19 comparable cases.
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.