Loading skill
Install any skill in seconds. Free to start, no credit card required.
Get Started Free →Synthesize the shortlist into a discussion-ready research idea brainstorm memo, writing `output/REPORT.md`, `output/APPENDIX.md`, and `output/REPORT.json`. **Trigger**: idea memo, brainstorm memo, research direction memo, report md, 研究备忘录, brainstorm report. **Use when**: the shortlist exists and you want the final reader-facing brainstorm deliverable for PI / PhD discussion.
.claude/skills/willoscar-idea-memo-writer/SKILL.md| Test case | Without → With | Effect | Δ tokens | Δ turns |
|---|---|---|---|---|
| case-22 | ✗→✓ | ▲ Improved | -66% | 0% |
| case-04 | ✗→✓ | ▲ Improved | -55% | 0% |
| case-06 | ✗→✓ | ▲ Improved | -60% | 0% |
| case-07 | ✗→✓ | ▲ Improved | -53% | 0% |
| case-08 | ✗→✓ | ▲ Improved | -34% | 0% |
Goal: turn the shortlist into a discussion-ready memo.
Always read:
references/overview.mdreferences/report_structure.mdUse scripts/run.py only for:
tooling/ideation.py library functions for renderingDo not treat run.py as the place for:
The final memo should feel like:
and still feel like:
uv run python .codex/skills/idea-memo-writer/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-memo-writer/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-02 | fail→fail | 24,131 | 27,954 | +16% | 1 | 1 | 0% | 3,596 | 4,026 | +12% | 0 | 0 | — |
case-22 | fail→pass | 13,358 | 2,801 | -79% | 1 | 1 | 0% | 2,020 | 683 | -66% | 0 | 0 | — |
case-01 | fail→fail | 17,562 | 5,884 | -66% | 1 | 1 | 0% | 3,022 | 572 | -81% | 0 | 0 | — |
case-03 | fail→fail | 20,523 | 3,495 | -83% | 1 | 1 | 0% | 3,555 | 491 | -86% | 0 | 0 | — |
case-04 | fail→pass | 14,742 | 5,167 | -65% | 1 | 1 | 0% | 3,069 | 1,383 | -55% | 0 | 0 | — |
case-05 | pass→pass | 8,095 | 3,066 | -62% | 1 | 1 | 0% | 1,452 | 917 | -37% | 0 | 0 | — |
case-06 | fail→pass | 8,624 | 1,574 | -82% | 1 | 1 | 0% | 1,399 | 558 | -60% | 0 | 0 | — |
case-07 | fail→pass | 19,510 | 2,006 | -90% | 1 | 1 | 0% | 1,505 | 709 | -53% | 0 | 0 | — |
case-08 | fail→pass | 6,492 | 2,655 | -59% | 1 | 1 | 0% | 1,216 | 804 | -34% | 0 | 0 | — |
case-09 | pass→fail | 7,473 | 1,442 | -81% | 1 | 1 | 0% | 945 | 488 | -48% | 0 | 0 | — |
case-10 | pass→pass | 8,514 | 4,687 | -45% | 1 | 1 | 0% | 1,502 | 1,021 | -32% | 0 | 0 | — |
case-11 | pass→fail | 10,305 | 3,712 | -64% | 1 | 1 | 0% | 1,524 | 808 | -47% | 0 | 0 | — |
case-12 | fail→pass | 11,967 | 11,421 | -5% | 1 | 1 | 0% | 1,776 | 1,956 | +10% | 0 | 0 | — |
case-13 | fail→pass | 10,645 | 1,907 | -82% | 1 | 1 | 0% | 1,739 | 607 | -65% | 0 | 0 | — |
case-14 | fail→pass | 10,387 | 2,434 | -77% | 1 | 1 | 0% | 1,512 | 639 | -58% | 0 | 0 | — |
case-15 | fail→pass | 6,930 | 3,457 | -50% | 1 | 1 | 0% | 1,135 | 833 | -27% | 0 | 0 | — |
case-16 | fail→pass | 9,166 | 2,159 | -76% | 1 | 1 | 0% | 1,542 | 713 | -54% | 0 | 0 | — |
case-17 | fail→pass | 9,287 | 3,576 | -61% | 1 | 1 | 0% | 1,266 | 861 | -32% | 0 | 0 | — |
case-18 | fail→pass | 4,042 | 1,443 | -64% | 1 | 1 | 0% | 489 | 546 | +12% | 0 | 0 | — |
case-19 | fail→pass | 19,622 | 4,984 | -75% | 1 | 1 | 0% | 4,526 | 1,120 | -75% | 0 | 0 | — |
case-20 | fail→fail | 20,351 | 13,967 | -31% | 1 | 1 | 0% | 3,576 | 2,268 | -37% | 0 | 0 | — |
case-21 | fail→fail | 21,415 | 30,732 | +44% | 1 | 1 | 0% | 5,395 | 6,504 | +21% | 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 19 counted toward the lift figure. The other 3 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 +50 percentage points is the difference between those two pass rates over the 19 comparable cases. 3 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.