Loading skill
Install any skill in seconds. Free to start, no credit card required.
Get Started Free →Rewrite `outline/claim_evidence_matrix.md` as a projection/index of evidence packs (NO PROSE), so claims/axes are driven by `outline/evidence_drafts.jsonl` rather than outline placeholders. **Trigger**: claim matrix rewriter, rewrite claim-evidence matrix, evidence-first claim matrix, matrix index, 证据矩阵重写, 从证据包生成矩阵.
.claude/skills/willoscar-claim-matrix-rewriter/SKILL.md| Test case | Without → With | Effect | Δ tokens | Δ turns |
|---|---|---|---|---|
| case-04 | ✗→✓ | ▲ Improved | -44% | 0% |
| case-06 | ✗→✓ | ▲ Improved | -34% | 0% |
| case-07 | ✗→✓ | ▲ Improved | -19% | 0% |
| case-10 | ✗→✓ | ▲ Improved | -40% | 0% |
| case-11 | ✗→✓ | ▲ Improved | -48% | 0% |
Purpose: make outline/claim_evidence_matrix.md a navigation/index layer that is driven by evidence packs (not by outline bullets).
Key property: keep the legacy, gate-checked format so downstream QA can rely on stable markers:
- Claim: (one per subsection) - Evidence: (>=2 per subsection)outline/subsection_briefs.jsonloutline/evidence_drafts.jsonlcitations/ref.biboutline/claim_evidence_matrix.md..., unicode ellipsis (…), TODO, (placeholder), or <!-- SCAFFOLD -->.uv run python .codex/skills/claim-matrix-rewriter/scripts/run.py --helpuv run python .codex/skills/claim-matrix-rewriter/scripts/run.py --workspace <workspace>uv run python .codex/skills/claim-matrix-rewriter/scripts/run.py --helpuv run python .codex/skills/claim-matrix-rewriter/scripts/run.py --workspace <workspace>--help.outline/subsection_briefs.jsonl, outline/evidence_drafts.jsonl, citations/ref.bib.outline/subsection_briefs.jsonl and outline/evidence_drafts.jsonl exist and blocking_missing is empty.citations/ref.bib exists and contains the cited keys.uv run python .codex/skills/claim-matrix-rewriter/scripts/run.py --workspace <workspace>citations/ref.bibFix:
citation-verifier (or fix the upstream notes) so every referenced key exists before rewriting.blocking_missingFix:
paper-notes / evidence-draft) first; the matrix should not “paper over” missing evidence.| Case | Status | Duration (ms) | Turns | Tokens | Tool calls | ||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| Without | With | Δ | Without | With | Δ | Without | With | Δ | Without | With | Δ | ||
case-01 | fail→fail | 32,868 | 5,527 | -83% | 1 | 1 | 0% | 6,190 | 842 | -86% | 0 | 0 | — |
case-02 | fail→fail | 4,855 | 5,740 | +18% | 1 | 1 | 0% | 268 | 830 | +210% | 0 | 0 | — |
case-03 | fail→fail | 14,445 | 6,097 | -58% | 1 | 1 | 0% | 2,192 | 837 | -62% | 0 | 0 | — |
case-04 | fail→pass | 13,218 | 3,868 | -71% | 1 | 1 | 0% | 2,152 | 1,198 | -44% | 0 | 0 | — |
case-05 | pass→pass | 9,438 | 3,364 | -64% | 1 | 1 | 0% | 1,374 | 1,076 | -22% | 0 | 0 | — |
case-06 | fail→pass | 11,532 | 4,028 | -65% | 1 | 1 | 0% | 1,940 | 1,282 | -34% | 0 | 0 | — |
case-07 | fail→pass | 9,024 | 6,950 | -23% | 1 | 1 | 0% | 1,397 | 1,138 | -19% | 0 | 0 | — |
case-08 | pass→pass | 12,836 | 2,137 | -83% | 1 | 1 | 0% | 2,066 | 870 | -58% | 0 | 0 | — |
case-09 | pass→pass | 12,842 | 3,149 | -75% | 1 | 1 | 0% | 1,876 | 1,036 | -45% | 0 | 0 | — |
case-10 | fail→pass | 11,078 | 3,054 | -72% | 1 | 1 | 0% | 1,737 | 1,036 | -40% | 0 | 0 | — |
case-11 | fail→pass | 11,120 | 2,582 | -77% | 1 | 1 | 0% | 1,681 | 867 | -48% | 0 | 0 | — |
case-12 | fail→pass | 4,785 | 2,188 | -54% | 1 | 1 | 0% | 668 | 742 | +11% | 0 | 0 | — |
case-13 | fail→pass | 8,961 | 1,839 | -79% | 1 | 1 | 0% | 1,338 | 863 | -36% | 0 | 0 | — |
case-14 | fail→pass | 5,720 | 1,309 | -77% | 1 | 1 | 0% | 794 | 751 | -5% | 0 | 0 | — |
case-19 | fail→pass | 10,350 | 1,511 | -85% | 1 | 1 | 0% | 1,543 | 776 | -50% | 0 | 0 | — |
case-15 | fail→pass | 10,054 | 4,130 | -59% | 1 | 1 | 0% | 1,466 | 1,076 | -27% | 0 | 0 | — |
case-16 | fail→pass | 10,199 | 3,043 | -70% | 1 | 1 | 0% | 1,565 | 984 | -37% | 0 | 0 | — |
case-17 | fail→pass | 9,173 | 2,423 | -74% | 1 | 1 | 0% | 1,441 | 839 | -42% | 0 | 0 | — |
case-18 | fail→pass | 6,831 | 4,144 | -39% | 1 | 1 | 0% | 1,081 | 1,062 | -2% | 0 | 0 | — |
case-20 | fail→fail | 4,925 | 6,062 | +23% | 1 | 1 | 0% | 698 | 766 | +10% | 0 | 0 | — |
case-21 | pass→fail | 11,715 | 8,675 | -26% | 1 | 1 | 0% | 1,988 | 2,122 | +7% | 0 | 0 | — |
case-22 | pass→pass | 6,727 | 5,009 | -26% | 1 | 1 | 0% | 1,213 | 1,434 | +18% | 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 18 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 +55 percentage points is the difference between those two pass rates over the 18 comparable cases. 2 cases got worse with the skill loaded, and they are included in that figure.
The publisher has shipped newer versions since this run, so these numbers describe v1, not the version currently listed.
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.