Loading skill
Install any skill in seconds. Free to start, no credit card required.
Get Started Free →Canonical rule owner for repo-harness ChatGPT integration -- Oracle-first browser/GPT Pro consult and continuation, MCP Connector setup, MCP bridge planning handoff, and Connector invocation read-back evidence.
.claude/skills/ancienttwo-repo-harness-chatgpt/SKILL.md| Test case | Without → With | Effect | Δ tokens | Δ turns |
|---|---|---|---|---|
| case-01 | ✗→✓ | ▲ Improved | -57% | 0% |
| case-02 | ✗→✓ | ▲ Improved | -70% | 0% |
| case-03 | ✗→✓ | ▲ Improved | -7% | 0% |
| case-04 | ✗→✓ | ▲ Improved | -34% | 0% |
| case-05 | ✗→✓ | ▲ Improved | -40% | 0% |
Canonical rule owner for every repo-harness ChatGPT integration surface: Oracle browser consult, session continuation, MCP Connector/bridge setup and operation, and MCP invocation read-back evidence. Discoverable only after explicit ChatGPT setup; never implied by either install profile. Router-only: mode protocol lives under references/.
references/setup.md.references/setup.md, then references/orchestrate.md.
references/consult.md.references/continue.md.references/read-back.md.references/bridge.md.references/delegate.md.references/campaign-issues.mdtask, lease, writes, shell, or acceptance authority.
| Case | Status | Duration (ms) | Turns | Tokens | Tool calls | ||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| Without | With | Δ | Without | With | Δ | Without | With | Δ | Without | With | Δ | ||
case-01 | fail→pass | 9,729 | 2,519 | -74% | 1 | 1 | 0% | 1,707 | 736 | -57% | 0 | 0 | — |
case-02 | fail→pass | 15,503 | 1,893 | -88% | 1 | 1 | 0% | 2,594 | 777 | -70% | 0 | 0 | — |
case-03 | fail→pass | 6,844 | 3,705 | -46% | 1 | 1 | 0% | 1,099 | 1,017 | -7% | 0 | 0 | — |
case-04 | fail→pass | 10,772 | 6,417 | -40% | 1 | 1 | 0% | 1,700 | 1,114 | -34% | 0 | 0 | — |
case-05 | fail→pass | 7,160 | 1,732 | -76% | 1 | 1 | 0% | 1,150 | 693 | -40% | 0 | 0 | — |
case-06 | fail→pass | 19,298 | 4,242 | -78% | 1 | 1 | 0% | 3,143 | 1,205 | -62% | 0 | 0 | — |
case-07 | fail→pass | 7,075 | 1,980 | -72% | 1 | 1 | 0% | 1,128 | 733 | -35% | 0 | 0 | — |
case-08 | fail→pass | 10,338 | 1,993 | -81% | 1 | 1 | 0% | 1,755 | 781 | -55% | 0 | 0 | — |
case-09 | fail→pass | 11,939 | 4,722 | -60% | 1 | 1 | 0% | 1,860 | 801 | -57% | 0 | 0 | — |
case-10 | fail→pass | 8,124 | 1,574 | -81% | 1 | 1 | 0% | 1,288 | 673 | -48% | 0 | 0 | — |
case-11 | fail→pass | 12,279 | 5,373 | -56% | 1 | 1 | 0% | 1,913 | 943 | -51% | 0 | 0 | — |
case-12 | fail→pass | 8,961 | 95,301 | +964% | 1 | 1 | 0% | 1,507 | 1,079 | -28% | 0 | 0 | — |
case-13 | fail→pass | 15,468 | 5,020 | -68% | 1 | 1 | 0% | 2,372 | 851 | -64% | 0 | 0 | — |
case-14 | pass→pass | 12,140 | 2,737 | -77% | 1 | 1 | 0% | 1,867 | 778 | -58% | 0 | 0 | — |
case-15 | pass→pass | 10,931 | 2,662 | -76% | 1 | 1 | 0% | 1,673 | 851 | -49% | 0 | 0 | — |
case-21 | pass→pass | 9,158 | 4,854 | -47% | 1 | 1 | 0% | 1,859 | 1,451 | -22% | 0 | 0 | — |
case-16 | fail→fail | 7,693 | 2,975 | -61% | 1 | 1 | 0% | 1,280 | 913 | -29% | 0 | 0 | — |
case-17 | fail→pass | 14,964 | 2,112 | -86% | 1 | 1 | 0% | 2,716 | 767 | -72% | 0 | 0 | — |
case-18 | fail→pass | 9,278 | 1,835 | -80% | 1 | 1 | 0% | 1,408 | 772 | -45% | 0 | 0 | — |
case-19 | fail→pass | 6,432 | 2,030 | -68% | 1 | 1 | 0% | 1,122 | 775 | -31% | 0 | 0 | — |
case-20 | pass→pass | 7,993 | 5,985 | -25% | 1 | 1 | 0% | 1,683 | 1,732 | +3% | 0 | 0 | — |
case-22 | pass→pass | 5,515 | 3,278 | -41% | 1 | 1 | 0% | 1,075 | 1,070 | -0% | 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. The headline lift of +73 percentage points is the difference between those two pass rates over the 22 comparable cases.
The publisher has shipped newer versions since this run, so these numbers describe v2, 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.
| Model | Method | Date | Lift |
|---|---|---|---|
| gemini-3.6-flash | verified | 8/21/2026 | +48% |
Other measured skills in the registry, with their headline benchmark lift.