Loading skill
Install any skill in seconds. Free to start, no credit card required.
Get Started Free →Canonical rule owner for PRD drafting, Sprint planning/execution, and native Goal-session preparation from repo-harness planning artifacts.
.claude/skills/ancienttwo-repo-harness-product/SKILL.md| Test case | Without → With | Effect | Δ tokens | Δ turns |
|---|---|---|---|---|
| case-04 | ✗→✓ | ▲ Improved | 48% | 0% |
| case-05 | ✗→✓ | ▲ Improved | 653% | 0% |
| case-06 | ✗→✓ | ▲ Improved | 794% | 0% |
| case-08 | ✗→✓ | ▲ Improved | -6% | 0% |
| case-09 | ✗→✓ | ▲ Improved | -4% | 0% |
Canonical rule owner for PRD, Sprint, and Goal. Router-only: shared preflight, mode selection, and cross-mode boundaries. Mode protocol lives under references/.
git rev-parse --show-toplevel.docs/spec.md and .ai/harness/policy.json when present.references/prd.md.references/sprint.md./goal session -> references/goal.md.> **Status**: Approved, or bypasses $think, repo-harness run capture-plan, task contracts, /check, or external acceptance on the user's behalf; explicit human approval always precedes implementation.tasks/todos.md is the deferred-goal ledger only; no mode treats it as an active backlog or goal queue.| Case | Status | Duration (ms) | Turns | Tokens | Tool calls | ||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| Without | With | Δ | Without | With | Δ | Without | With | Δ | Without | With | Δ | ||
case-01 | fail→fail | 11,387 | 5,909 | -48% | 1 | 1 | 0% | 1,910 | 636 | -67% | 0 | 0 | — |
case-02 | fail→fail | 7,858 | 4,665 | -41% | 1 | 1 | 0% | 1,309 | 558 | -57% | 0 | 0 | — |
case-03 | fail→fail | 4,453 | 5,529 | +24% | 1 | 1 | 0% | 789 | 665 | -16% | 0 | 0 | — |
case-04 | fail→pass | 3,963 | 4,169 | +5% | 1 | 1 | 0% | 667 | 984 | +48% | 0 | 0 | — |
case-05 | fail→pass | 3,276 | 4,169 | +27% | 1 | 1 | 0% | 139 | 1,047 | +653% | 0 | 0 | — |
case-06 | fail→pass | 4,270 | 6,126 | +43% | 1 | 1 | 0% | 142 | 1,269 | +794% | 0 | 0 | — |
case-07 | pass→pass | 9,176 | 3,823 | -58% | 1 | 1 | 0% | 1,756 | 1,036 | -41% | 0 | 0 | — |
case-08 | fail→pass | 6,215 | 4,477 | -28% | 1 | 1 | 0% | 1,143 | 1,077 | -6% | 0 | 0 | — |
case-09 | fail→pass | 5,534 | 2,936 | -47% | 1 | 1 | 0% | 794 | 762 | -4% | 0 | 0 | — |
case-10 | pass→fail | 11,059 | 10,425 | -6% | 1 | 1 | 0% | 1,829 | 2,045 | +12% | 0 | 0 | — |
case-11 | fail→pass | 13,177 | 2,037 | -85% | 1 | 1 | 0% | 2,044 | 666 | -67% | 0 | 0 | — |
case-12 | fail→pass | 9,251 | 1,927 | -79% | 1 | 1 | 0% | 1,613 | 640 | -60% | 0 | 0 | — |
case-13 | fail→pass | 9,591 | 1,753 | -82% | 1 | 1 | 0% | 1,515 | 606 | -60% | 0 | 0 | — |
case-14 | pass→pass | 8,270 | 1,920 | -77% | 1 | 1 | 0% | 1,340 | 597 | -55% | 0 | 0 | — |
case-15 | pass→pass | 7,582 | 2,310 | -70% | 1 | 1 | 0% | 1,300 | 662 | -49% | 0 | 0 | — |
case-16 | fail→pass | 2,706 | 7,143 | +164% | 1 | 1 | 0% | 353 | 1,114 | +216% | 0 | 0 | — |
case-17 | fail→pass | 12,879 | 1,585 | -88% | 1 | 1 | 0% | 1,399 | 532 | -62% | 0 | 0 | — |
case-18 | fail→pass | 3,478 | 3,360 | -3% | 1 | 1 | 0% | 543 | 899 | +66% | 0 | 0 | — |
case-19 | pass→pass | 4,458 | 2,060 | -54% | 1 | 1 | 0% | 801 | 669 | -16% | 0 | 0 | — |
case-20 | pass→pass | 3,322 | 1,487 | -55% | 1 | 1 | 0% | 596 | 547 | -8% | 0 | 0 | — |
case-21 | pass→pass | 3,032 | 2,207 | -27% | 1 | 1 | 0% | 406 | 688 | +69% | 0 | 0 | — |
case-22 | pass→pass | 2,967 | 2,582 | -13% | 1 | 1 | 0% | 450 | 662 | +47% | 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 17 counted toward the lift figure. The other 5 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 +45 percentage points is the difference between those two pass rates over the 17 comparable cases. 1 case got worse with the skill loaded, and it is 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.