Loading skill
Install any skill in seconds. Free to start, no credit card required.
Get Started Free →Canonical rule owner for installing, migrating, upgrading, repairing, scaffolding, and capability-configuring the repo-harness workflow in a repository.
.claude/skills/ancienttwo-repo-harness-setup/SKILL.md| Test case | Without → With | Effect | Δ tokens | Δ turns |
|---|---|---|---|---|
| case-04 | ✗→✓ | ▲ Improved | -48% | 0% |
| case-05 | ✗→✓ | ▲ Improved | -31% | 0% |
| case-06 | ✗→✓ | ▲ Improved | -52% | 0% |
| case-07 | ✗→✓ | ▲ Improved | -51% | 0% |
| case-08 | ✗→✓ | ▲ Improved | -59% | 0% |
Canonical rule owner for init, migrate, upgrade, repair, scaffold, and capability configuration. Router-only: shared preflight, mode selection, and cross-mode boundaries. Mode protocol lives under references/.
pwd, or an explicit --repo argument).bun scripts/inspect-project-state.ts --repo <repo> --format text when available.references/init.md.references/migrate.md.references/upgrade.md.references/repair.md.references/scaffold.md.references/capability.md.HOME) state from a repo-scoped mode; user-level setup is the separate repo-harness update command.ownership=known_generated files.scaffold.create-project-dirs, direct scripts/init-project.sh, hooks-init, docs-init) as public commands.| Case | Status | Duration (ms) | Turns | Tokens | Tool calls | ||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| Without | With | Δ | Without | With | Δ | Without | With | Δ | Without | With | Δ | ||
case-01 | fail→fail | 4,412 | 5,557 | +26% | 1 | 1 | 0% | 199 | 684 | +244% | 0 | 0 | — |
case-02 | fail→fail | 3,965 | 2,394 | -40% | 1 | 1 | 0% | 292 | 666 | +128% | 0 | 0 | — |
case-03 | fail→fail | 4,456 | 2,845 | -36% | 1 | 1 | 0% | 153 | 884 | +478% | 0 | 0 | — |
case-04 | fail→pass | 8,961 | 2,004 | -78% | 1 | 1 | 0% | 1,420 | 737 | -48% | 0 | 0 | — |
case-05 | fail→pass | 6,709 | 2,165 | -68% | 1 | 1 | 0% | 1,059 | 731 | -31% | 0 | 0 | — |
case-06 | fail→pass | 10,582 | 2,115 | -80% | 1 | 1 | 0% | 1,499 | 720 | -52% | 0 | 0 | — |
case-07 | fail→pass | 8,788 | 2,013 | -77% | 1 | 1 | 0% | 1,496 | 734 | -51% | 0 | 0 | — |
case-08 | fail→pass | 9,505 | 2,160 | -77% | 1 | 1 | 0% | 1,601 | 664 | -59% | 0 | 0 | — |
case-09 | pass→pass | 7,470 | 2,070 | -72% | 1 | 1 | 0% | 1,143 | 680 | -41% | 0 | 0 | — |
case-10 | fail→pass | 11,629 | 1,399 | -88% | 1 | 1 | 0% | 2,091 | 622 | -70% | 0 | 0 | — |
case-11 | fail→pass | 7,167 | 2,502 | -65% | 1 | 1 | 0% | 1,327 | 809 | -39% | 0 | 0 | — |
case-12 | fail→pass | 11,451 | 2,074 | -82% | 1 | 1 | 0% | 1,981 | 744 | -62% | 0 | 0 | — |
case-13 | fail→pass | 10,086 | 3,340 | -67% | 1 | 1 | 0% | 1,746 | 944 | -46% | 0 | 0 | — |
case-14 | pass→pass | 13,041 | 3,003 | -77% | 1 | 1 | 0% | 2,161 | 916 | -58% | 0 | 0 | — |
case-15 | fail→pass | 11,451 | 2,066 | -82% | 1 | 1 | 0% | 2,054 | 655 | -68% | 0 | 0 | — |
case-16 | fail→pass | 8,701 | 2,462 | -72% | 1 | 1 | 0% | 1,475 | 824 | -44% | 0 | 0 | — |
case-17 | pass→pass | 11,435 | 1,792 | -84% | 1 | 1 | 0% | 1,999 | 639 | -68% | 0 | 0 | — |
case-18 | fail→pass | 9,761 | 3,121 | -68% | 1 | 1 | 0% | 1,831 | 905 | -51% | 0 | 0 | — |
case-19 | fail→pass | 11,486 | 1,769 | -85% | 1 | 1 | 0% | 2,196 | 648 | -70% | 0 | 0 | — |
case-20 | pass→pass | 9,427 | 1,563 | -83% | 1 | 1 | 0% | 1,599 | 602 | -62% | 0 | 0 | — |
case-21 | fail→pass | 4,386 | 1,578 | -64% | 1 | 1 | 0% | 689 | 609 | -12% | 0 | 0 | — |
case-22 | fail→pass | 15,284 | 2,210 | -86% | 1 | 1 | 0% | 2,768 | 741 | -73% | 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 20 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 +68 percentage points is the difference between those two pass rates over the 20 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.