Loading skill
Install any skill in seconds. Free to start, no credit card required.
Get Started Free →Converge the screened direction pool into a small, discussion-ready shortlist, writing `output/trace/IDEA_SHORTLIST.md`. **Trigger**: idea shortlist, shortlist directions, brainstorm shortlist, 方向 shortlist. **Use when**: you already have a direction pool and screening table and want the strongest 3-5 directions for the final memo.
.claude/skills/willoscar-idea-shortlist-curator/SKILL.md| Test case | Without → With | Effect | Δ tokens | Δ turns |
|---|---|---|---|---|
| case-04 | ✗→✓ | ▲ Improved | 102% | 0% |
| case-05 | ✗→✓ | ▲ Improved | -53% | 0% |
| case-06 | ✗→✓ | ▲ Improved | -20% | 0% |
| case-07 | ✗→✓ | ▲ Improved | 199% | 0% |
| case-08 | ✗→✓ | ▲ Improved | -21% | 0% |
Always read:
references/overview.mdRead by task:
references/ranking_rubric.md when customizing risk notes or deferral reasonsassets/rationale_templates.json when changing deterministic shortlist rationale wording or rule orderUse scripts/run.py only for:
Do not treat run.py as the place for:
idea-memo-writer)Goal: turn a direction pool into a small shortlist that is:
and remains:
uv run python .codex/skills/idea-shortlist-curator/scripts/run.py --workspace <workspace>--workspace <dir> (required)--unit-id <U###>--inputs <semicolon-separated>--outputs <semicolon-separated>--checkpoint <C#>uv run python .codex/skills/idea-shortlist-curator/scripts/run.py --workspace workspaces/brainstorm-llm-agents| Case | Status | Duration (ms) | Turns | Tokens | Tool calls | ||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| Without | With | Δ | Without | With | Δ | Without | With | Δ | Without | With | Δ | ||
case-01 | fail→fail | 5,105 | 6,715 | +32% | 1 | 1 | 0% | 331 | 957 | +189% | 0 | 0 | — |
case-02 | fail→fail | 4,172 | 5,416 | +30% | 1 | 1 | 0% | 194 | 615 | +217% | 0 | 0 | — |
case-03 | fail→fail | 24,377 | 5,269 | -78% | 1 | 1 | 0% | 3,788 | 690 | -82% | 0 | 0 | — |
case-04 | fail→pass | 2,506 | 1,632 | -35% | 1 | 1 | 0% | 320 | 647 | +102% | 0 | 0 | — |
case-05 | fail→pass | 9,806 | 2,508 | -74% | 1 | 1 | 0% | 1,602 | 751 | -53% | 0 | 0 | — |
case-06 | fail→pass | 4,793 | 1,121 | -77% | 1 | 1 | 0% | 637 | 508 | -20% | 0 | 0 | — |
case-07 | fail→pass | 2,606 | 2,440 | -6% | 1 | 1 | 0% | 272 | 814 | +199% | 0 | 0 | — |
case-08 | fail→pass | 6,393 | 2,790 | -56% | 1 | 1 | 0% | 1,122 | 890 | -21% | 0 | 0 | — |
case-09 | fail→pass | 6,850 | 2,206 | -68% | 1 | 1 | 0% | 1,139 | 714 | -37% | 0 | 0 | — |
case-10 | fail→pass | 9,675 | 2,127 | -78% | 1 | 1 | 0% | 1,542 | 732 | -53% | 0 | 0 | — |
case-11 | fail→pass | 9,881 | 1,734 | -82% | 1 | 1 | 0% | 1,429 | 659 | -54% | 0 | 0 | — |
case-12 | pass→pass | 15,159 | 1,409 | -91% | 1 | 1 | 0% | 1,044 | 566 | -46% | 0 | 0 | — |
case-13 | fail→pass | 9,832 | 2,358 | -76% | 1 | 1 | 0% | 1,294 | 766 | -41% | 0 | 0 | — |
case-14 | fail→pass | 8,225 | 1,708 | -79% | 1 | 1 | 0% | 1,183 | 609 | -49% | 0 | 0 | — |
case-15 | fail→pass | 6,461 | 2,752 | -57% | 1 | 1 | 0% | 976 | 863 | -12% | 0 | 0 | — |
case-16 | fail→pass | 5,413 | 2,303 | -57% | 1 | 1 | 0% | 924 | 839 | -9% | 0 | 0 | — |
case-17 | pass→pass | 17,492 | 9,614 | -45% | 1 | 1 | 0% | 2,504 | 1,785 | -29% | 0 | 0 | — |
case-18 | pass→pass | 14,568 | 9,810 | -33% | 1 | 1 | 0% | 2,175 | 1,782 | -18% | 0 | 0 | — |
case-19 | fail→fail | 9,669 | 6,080 | -37% | 1 | 1 | 0% | 1,322 | 1,246 | -6% | 0 | 0 | — |
case-20 | pass→pass | 9,343 | 5,717 | -39% | 1 | 1 | 0% | 1,357 | 1,225 | -10% | 0 | 0 | — |
case-21 | fail→pass | 10,575 | 1,690 | -84% | 1 | 1 | 0% | 1,544 | 607 | -61% | 0 | 0 | — |
case-22 | fail→pass | 14,923 | 4,215 | -72% | 1 | 1 | 0% | 2,796 | 1,034 | -63% | 0 | 0 | — |
case-23 | fail→pass | 24,225 | 14,365 | -41% | 1 | 1 | 0% | 3,680 | 2,051 | -44% | 0 | 0 | — |
case-24 | fail→pass | 7,223 | 14,798 | +105% | 1 | 1 | 0% | 1,040 | 1,822 | +75% | 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. 24 cases were attempted, and 22 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 +67 percentage points is the difference between those two pass rates over the 22 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.